Microbial fermentation process intelligent monitoring and regulation method and system based on large language model

Abstract

The invention belongs to the technical field of bioengineering and artificial intelligence crossing, and particularly discloses a microbial fermentation process intelligent monitoring and regulation method and system based on a large language model. According to the method, sensor data of the fermentation system are accessed and analyzed in real time, a rule engine and an anomaly detection model are periodically called, anomaly grading is performed on the data, anomaly interpretation and suggestions are generated in combination with an RAG knowledge base, future key parameter trends are predicted based on a time sequence prediction model (LSTM), trend deviation analysis is performed, and fitting quality indexes are generated. And generating and executing a regulation strategy. The method has significant progress in the aspects of complex anomaly diagnosis, strategy dynamic optimization, fine regulation and control, knowledge reuse and the like.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of bioengineering and artificial intelligence, specifically relating to a method and system for intelligent monitoring and regulation of microbial fermentation processes based on a large language model. Background Technology

[0002] Monitoring and regulating microbial fermentation processes presents significant unique challenges compared to general industrial control, primarily due to the strong coupling of biological variables, dynamic nonlinear temporal characteristics, and environmental sensitivity. Multiple parameters, such as strain metabolic activity, substrate consumption, and product synthesis, are highly correlated; key parameters like pH and dissolved oxygen exhibit non-steady-state changes over time; and the processes respond dramatically to external disturbances such as temperature and aeration rates, resulting in a narrow regulatory window. These characteristics pose substantial challenges to traditional technologies in this field.

[0003] Current mainstream technologies primarily rely on PLC / DCS systems and rule engines, which have four significant limitations. First, there is a noticeable lag in anomaly detection. Traditional static threshold alarms struggle to capture gradual parameter changes in a timely manner; by the time an alarm is triggered, the anomaly has persisted for several hours. The strong coupling characteristics of biological systems mean that a single parameter anomaly is accompanied by coordinated changes in multiple parameters, but traditional systems lack the ability to perform correlation analysis. The nonlinear characteristics of fermentation processes lead to complex and varied anomaly patterns, making it difficult for fixed rules to comprehensively cover them. Data shows that traditional methods for detecting metabolic flux anomalies in antibiotic fermentation have a detection delay of 45-90 minutes, severely impacting product yield and increasing the difficulty of regulation. Second, regarding regulation strategies, fixed PID parameters cannot adapt to the differences in characteristics at different fermentation stages, and feeding strategies only support simple timed or quantitative modes. This mismatch between the control algorithm and the dynamic characteristics of the biological process stems from the lack of a stage-adaptive mechanism. For example, in antibiotic fermentation, the requirements for feeding rates differ significantly between the logarithmic growth phase and the stationary phase, but traditional systems cannot automatically adjust parameters. Third, insufficient knowledge utilization is another prominent drawback. Historical experimental data is typically stored as isolated files, failing to achieve structured associations; expert experience is also not transformed into callable knowledge rules. This is mainly due to the lack of knowledge graph construction technology for the fermentation field, resulting in a severe disconnect between SOP documents and real-time systems. In practice, technicians often need to query information across multiple isolated systems, leading to low decision-making efficiency. Fourth, the inefficiency of human-computer interaction also restricts system performance. Alarm information usually only displays simple parameter out-of-limit prompts, such as "pH abnormal," lacking root cause analysis capabilities. When making control decisions, operators have to rely on manually consulting paper records. This situation reflects the system's inadequacy in integrating natural language processing technology, failing to achieve knowledge visualization and interactive diagnosis.

[0004] These limitations render existing systems inadequate when dealing with complex fermentation processes. For example, in the case of dissolved oxygen anomalies, traditional methods not only suffer from delayed responses but may also lead to erroneous control due to misjudgments. As industries such as biopharmaceuticals place increasingly higher demands on fermentation processes, the development of a new generation of systems with multimodal perception, intelligent temporal modeling, knowledge-enhanced decision-making, and human-machine collaborative interfaces has become an urgent need. Summary of the Invention

[0005] To address the problems existing in the prior art, the purpose of this invention is to provide a method and system for intelligent monitoring and regulation of microbial fermentation processes based on Large Language Model (LLM).

[0006] To achieve the objectives of this invention, the technical solution is as follows: In a first aspect, the present invention provides a method for intelligent monitoring and regulation of microbial fermentation processes based on a large language model, comprising the following steps: S1 can access and analyze sensor data from the fermentation system in real time, such as temperature, pH, dissolved oxygen (DO), and feed flow rate. S2 periodically calls the rule engine and anomaly detection model to classify the collected sensor data into anomalies, such as no anomaly, mild anomaly, or severe anomaly. S3 pushes the anomaly classification results to users, and generates anomaly explanations and suggestions based on the RAG knowledge base for those classified as mild or severe anomalies. S4. For anomalies classified as mild, predict future trends of key parameters based on the time series prediction model (LSTM), perform trend deviation analysis, generate fitting quality indices, and generate control strategies; for anomalies classified as severe, stop automatic operation. S5, for abnormalities classified as mild, execute the control strategy generated in S4; S6 records all control operations and abnormal events, generates task reports, and synchronizes them to the fermentation control system.

[0007] Furthermore, the rule engine described in S2 is designed for microbial fermentation processes and is a rule-based decision-making system used to analyze and judge real-time sensor data to assist in anomaly detection and control strategy generation.

[0008] The information accessed by the rules engine includes:

[0009] Furthermore, the anomaly detection model described in S2 employs a combination of threshold rules and statistical anomaly detection, with anomaly classification dynamically adjusted based on the degree of deviation and the scope of impact. The anomaly detection model in S2 not only addresses the issue of lagging anomaly detection but also more broadly resolves the shortcomings of traditional PLC / DCS and rule engines in areas such as lack of multi-parameter correlation, nonlinear complexity, variable anomaly patterns, and false alarms and false negatives.

[0010] This invention performs multi-dimensional data correlation analysis by periodically calling a rule engine and anomaly detection model. It considers the complex coupling relationships between sensor data (pH, DO, temperature, feed flow rate, etc.), enabling anomaly classification to consider not only single parameter deviations but also risk indicators of coordinated deviations. Simultaneously, it adjusts thresholds based on fermentation stage, historical batch data, and predicted trends, making anomaly detection more sensitive and robust.

[0011] More specifically, based on the anomaly classification of S2, its meaning and the corresponding actions of S4 include:

[0012] Furthermore, the RAG knowledge base described in S3 includes experimental SOPs, a collaborative knowledge base (WIKI) for anomalies, a fermentation protocol library, and reference documents / literature. The generated suggestions include links to the knowledge sources and reference documents. The RAG knowledge base is not entirely derived from user historical data, but rather is a multi-source, structured knowledge system. Specific details are as follows:

[0013] Furthermore, S3 pushes anomaly classification results, anomaly explanations, and suggestions to users through an instant messaging (IM) platform.

[0014] Furthermore, the control strategies generated by S4 include, but are not limited to, process parameter adjustments such as feeding strategies, aeration strategies, stirring strategies, and temperature adjustment strategies. The strategy selection is dynamically adjusted based on the anomaly type and historical results.

[0015] Furthermore, S4 also includes providing a visual interface for users, allowing them to modify control strategies or send control commands directly.

[0016] Secondly, this invention provides an intelligent monitoring and control system for microbial fermentation processes based on a large language model, comprising: The data sensing module is used to access and analyze sensor data from the fermentation system in real time. The monitoring module is used to periodically detect data anomalies and issue tiered alarms. The diagnostic module is used to generate anomaly explanations and suggestions by combining the RAG knowledge base; The prediction module is used to predict parameter trends based on time series models and generate fitting quality indicators. The control module is used to execute control strategies; The coordination module is used to schedule tasks from various modules and generate task reports.

[0017] Thirdly, the present invention provides a computer-readable storage medium including a stored program, wherein, when the program is running, it controls the device where the computer-readable storage medium is located to execute the intelligent monitoring and control method for microbial fermentation process based on a large language model as described in the first aspect.

[0018] Fourthly, the present invention provides an electronic device, including a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the electronic device is triggered to execute the intelligent monitoring and control method for microbial fermentation process based on a large language model as described in the first aspect.

[0019] Compared with the prior art, the advantages and positive effects of the present invention are as follows: This invention provides a novel intelligent monitoring and control method for microbial fermentation processes. For the first time, it applies a multi-agent architecture to the monitoring and control of microbial fermentation processes, enabling collaborative analysis of multi-dimensional data and supporting trend prediction and fitting for complex nonlinear processes, providing a reference for strategy optimization. This invention also integrates a large language model and RAG technology to enhance the intelligence level of anomaly identification, diagnosis, and control suggestions. The control strategy generated by this invention supports both automatic execution and manual confirmation, ensuring safety and controllability.

[0020] In summary, this invention provides a method and system for intelligent monitoring and regulation of microbial fermentation processes based on Large Language Model (LLM), which represents a significant advancement in complex anomaly diagnosis, dynamic strategy optimization, fine-grained regulation, and knowledge reuse. Attached Figure Description

[0021] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the system module connections of the present invention.

[0024] Figure 2 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0025] To better understand the above-mentioned objectives, features, and advantages of the present invention, the solutions of the present invention will be further described below. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.

[0026] Many specific details are set forth in the following description in order to provide a full understanding of the invention, but the invention may also be practiced in other ways different from those described herein; obviously, the embodiments in the specification are only some embodiments of the invention, and not all embodiments.

[0027] The preferred embodiments of the present invention will now be described in detail with reference to specific examples. It should be understood that the following examples are given for illustrative purposes only and are not intended to limit the scope of the invention. Those skilled in the art can make various modifications and substitutions to the present invention without departing from its spirit and essence.

[0028] Example 1 See Figure 1 As shown, this embodiment provides an intelligent monitoring and control system for microbial fermentation processes based on a large language model, including: The data sensing module is used to access and analyze sensor data from the fermentation system in real time. The monitoring module is used to periodically detect data anomalies and issue tiered alarms. The diagnostic module is used to generate anomaly explanations and suggestions by combining the RAG knowledge base; The prediction module is used to predict parameter trends based on time series models and generate fitting quality indicators. The control module is used to execute control strategies; The coordination module is used to schedule tasks from various modules and generate task reports.

[0029] Example 2 This embodiment uses the antibiotic fermentation production process as a specific application scenario to illustrate the specific implementation process of the method of the present invention for trend prediction, generation and execution of control strategies for the fermentation process when the system determines the current state as slightly abnormal during the fermentation operation.

[0030] I. Application Scenarios and Experimental Conditions Production location: Antibiotic production workshop; Fermentation target: Streptomyces microorganisms; Fermentation method: Batch feeding aerobic fermentation; Fermentation equipment: a mechanically stirred fermenter with an effective working volume of 50 m³; Fermentation stage: mid-to-late logarithmic growth phase (fermentation time approximately 36 hours).

[0031] II. The main control parameters for the fermentation process are as follows: Temperature control range: 30.0~32.0 ℃; pH control range: 6.8~7.2; Dissolved oxygen (DO) control target: 30%~40%; Initial ventilation rate: based on the tank's design rating; Feeding method: continuous feeding, with the feeding flow rate set according to the cell growth rate.

[0032] III. Intelligent monitoring and control methods for microbial fermentation processes based on large language models, such as... Figure 2 As shown, it includes the following steps: S1. Real-time access and analysis of sensor data from the fermentation system. Specifically: During fermentation, the system continuously collects the following sensor data through an industrial IoT interface with a sampling period of 1 minute: fermentation tank temperature, fermentation broth pH value, dissolved oxygen concentration (DO), feed pump instantaneous flow rate, stirring speed, and aeration rate.

[0033] The collected data is sequentially processed by timestamp alignment and unit standardization, and single outliers that deviate significantly from the physical reasonable range are removed to form a continuous and stable structured data sequence.

[0034] S2. Periodically invoke the rule engine and anomaly detection model to classify anomalies in the collected sensor data. Specifically: 1. Analysis was performed using 60 consecutive minutes of runtime data. The results are as follows: Dissolved oxygen decreased gradually from 38% to 29%, with no abrupt changes. During the same period, the pH decreased from 7.06 to 6.93; The temperature remained within the range of 31.0–31.3 ℃ without significant fluctuations; The stirring speed and aeration rate remained at the set values, and the upper limit constraint was not triggered.

[0035] 2. Use a rule engine to determine the above data: Each individual parameter remains within the allowable range of the process; Dissolved oxygen and pH decreased slowly and synchronously, which is consistent with the typical characteristics of the enhanced bacterial metabolism phase; No equipment malfunctions or rapid loss of control risks were detected.

[0036] 3. The statistical anomaly detection model further analyzes the difference between the parameter change rate and historical batch data, confirming that the deviation is below the severe anomaly threshold. Based on the deviation magnitude, change trend, and impact range, the system classifies the current state as a mild anomaly.

[0037] S3. Push the anomaly classification results to the user, and generate anomaly explanations and suggestions based on the RAG knowledge base. Specifically: After classifying mild anomalies, the RAG knowledge base is invoked for retrieval. The retrieval content includes: Typical dissolved oxygen changes during the later stages of the logarithmic growth phase of antibiotic fermentation; Case studies of treatments where dissolved oxygen levels decreased slowly in historical batches; Operating instructions for the increased oxygen consumption stage in the enterprise's process SOP.

[0038] 2. Based on the search results, the system generates an explanation text for the anomaly, clearly stating: The current synergistic decrease in dissolved oxygen and pH is consistent with the enhanced metabolic activity of the bacteria, which is a slight deviation that can be recovered through process fine-tuning.

[0039] The explanation of the anomaly and the corresponding source of knowledge are displayed in the monitoring interface and recorded in the system log.

[0040] S4. Based on the time series forecasting model (LSTM), predict the future trends of key parameters, perform trend deviation analysis, generate fitting quality indicators, and generate control strategies. Specifically: 1. The system selects dissolved oxygen, pH, and feed flow rate as prediction targets, and calls an LSTM time-series prediction model trained based on the user's historical batches of similar products to predict the parameter change trends for the next 60 minutes. The prediction results show: If the current ventilation and feeding conditions are maintained, dissolved oxygen will drop below 25% in about 45 minutes; The predicted curve shows a clear similarity to historical low DO batches; The model's coefficient of determination R² for this prediction window is 0.90.

[0041] 2. Based on the prediction results and process constraints, the system generates the following control strategy: Without changing the stirring speed, increase the aeration rate by 5% from the original set value; The continuous feed rate was reduced by 3% to reduce the oxygen load in the short term. The observation period for the control effect was set to 30 minutes.

[0042] S5. Implementation and Effectiveness Monitoring of Control Strategies The control module converts the above control strategies into control commands and sends them to the fermentation control system to adjust the opening degree of the aeration valve and the speed of the feed pump.

[0043] After the strategy is executed, the system continuously monitors changes in key parameters, and the results are as follows: After about 15 minutes of regulation, dissolved oxygen stopped decreasing and gradually increased. During the 30-minute observation period, dissolved oxygen remained stable in the range of 32% to 35%. The pH decrease trend has slowed significantly and has not approached the lower limit threshold again; Temperature and other parameters remain stable.

[0044] No new abnormal signals were detected, indicating that the control strategy was implemented effectively.

[0045] S6. Data Recording and Experiment Report Generation The system automatically generates a complete record of this minor anomaly handling, including: Anomaly triggering time and classification results; Raw data and predicted curves of key parameters during the abnormal period; The content, execution time, and execution parameters of the control strategy; Comparison of parameter stability before and after regulation.

[0046] The above records are synchronously stored in the fermentation control system and user process database for subsequent process analysis, model optimization and knowledge reuse.

[0047] Example 3 This embodiment uses the industrial enzyme preparation production process as a specific application scenario to illustrate the specific implementation process of the method of the present invention in stopping automatic control behavior, prompting abnormal risks, and supporting manual intervention when the monitoring system determines the current state as a serious abnormality during the fermentation process.

[0048] I. Application Scenarios and Experimental Conditions Production location: Industrial enzyme preparation production workshop; Fermentation target: Recombinant Bacillus; Fermentation method: High-density aerobic fermentation; Fermentation equipment: a mechanically stirred fermenter with an effective working volume of 30 m³; Fermentation stage: Stable expression period of the target enzyme (fermentation time approximately 52 hours).

[0049] II. The main control parameters for the fermentation process are as follows: Temperature control range: 28.0–30.0 ℃; pH control range: 6.5–7.0; Dissolved oxygen (DO) control target: not less than 25%; Ventilation mode: continuous ventilation, adjustable ventilation volume; Stirring method: multi-stage stirring, with speed set in stages.

[0050] III. Intelligent monitoring and control methods for microbial fermentation processes based on large language models, such as... Figure 2 As shown, it includes the following steps: S1. Real-time access and analysis of sensor data from the fermentation system. Specifically: During fermentation, the system continuously collects parameters such as temperature, pH, dissolved oxygen, aeration rate, stirring speed, and feed flow rate inside the fermenter through an industrial communication interface with a sampling period of 30 seconds.

[0051] Before entering the monitoring module, the collected data undergoes timestamp synchronization, unit unification, and rationality verification, and is stored in a structured form for anomaly detection and event logging.

[0052] S2. Periodically invoke the rule engine and anomaly detection model to classify anomalies in the collected sensor data. Specifically: 1. During the stable expression period, the system continuously detected the following abnormal phenomena: Dissolved oxygen levels dropped rapidly from 28% to 9% within 8 minutes; During the same period, the temperature of the fermentation broth rose from 29.2 ℃ to 31.0 ℃, exceeding the upper limit set by the process. The pH value dropped from 6.8 to 6.4, exceeding the lower limit allowed by the process. The current ventilation rate and stirring speed are already at the maximum allowable operating state of the equipment, and the oxygen supply capacity cannot be further increased.

[0053] 2. Based on preset severe anomaly judgment rules, the rules engine confirms that multiple key parameters simultaneously exceed the safe operating range; the statistical anomaly detection model compares historical fermentation batch data and determines that the rate of change of this anomaly is highly consistent with recorded fermentation runaway events. Based on the comprehensive assessment results of the anomaly magnitude, rate of change, and potential process risks, the system classifies the current state as a severe anomaly.

[0054] S3. Push the anomaly classification results to the user, and generate anomaly explanations and suggestions based on the RAG knowledge base. Specifically: 1. After confirming the severity anomaly classification result, the system calls the RAG knowledge base for targeted retrieval. The retrieval content includes: A typical abnormal case of a sudden drop in dissolved oxygen during high-density fermentation of industrial enzymes; Record of process risk analysis under conditions of limited ventilation capacity; Enterprise safety production and abnormal handling regulations.

[0055] 2. Based on the above information, the system generates an exception description, clearly stating: The current fermentation state has exceeded the limits of automatic control capabilities. Continuing to perform automatic control operations carries a significant risk of amplification, which may lead to an irreversible decrease in enzyme activity or batch failure.

[0056] The anomaly description and risk warning are pushed to users simultaneously through the monitoring interface and instant messaging, and are marked as high-priority alarm events.

[0057] S4. Automatic Operation Abort and System Protection In the event of a severe abnormal situation, the system immediately executes automatic operation suspension measures, including: The automatic control of process parameters such as ventilation, feeding, and stirring by the control module is suspended. Stop the strategy generation and execution process based on the time series prediction model; Lock automatic control permissions and retain only the manual operation interface.

[0058] In this state, the system continuously collects and displays key parameters in real time, providing data support for manual investigation and decision-making.

[0059] S5. Human intervention support and process recording The system provides a critical anomaly handling record entry in the monitoring interface for manual entry of the following information: The cause of the anomaly has been manually confirmed; The manual handling measures implemented; Time for the abnormal state to be resolved.

[0060] The above information is associated with the corresponding serious anomalies and stored as an important basis for subsequent analysis and auditing.

[0061] S6. Exception Event Archiving and Data Synchronization The system has fully archived this serious anomaly event, and the archived content includes: Records of key parameter changes before and after the anomaly occurred; The basis and timeframe for determining severe anomalies; Automatic operation stoppage record; The process of human intervention and the final result.

[0062] The relevant data is synchronized to the enterprise's MES system and anomaly case knowledge base for subsequent process risk assessment, system boundary revision, and safety management optimization.

[0063] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for intelligent monitoring and control of microbial fermentation processes based on a large language model, characterized in that, Includes the following steps: S1, real-time access and analysis of sensor data from the fermentation system; S2 periodically calls the rule engine and anomaly detection model to classify the collected sensor data into anomalies, such as no anomaly, mild anomaly, or severe anomaly. S3 pushes the anomaly classification results to users, and generates anomaly explanations and suggestions based on the RAG knowledge base for those classified as mild or severe anomalies. S4. For anomalies classified as mild, predict future trends of key parameters based on time-series prediction models, perform trend deviation analysis, generate fitting quality indices, and generate control strategies. For anomalies classified as severe, automatic operation shall be suspended. S5, for abnormalities classified as mild, execute the control strategy generated in S4; S6 records all control operations and abnormal events, generates task reports, and synchronizes them to the fermentation control system.

2. The method according to claim 1, characterized in that, The rule engine described in S2 is designed for microbial fermentation processes and calls information including real-time sensor data, historical process data, threshold rules, statistical anomaly detection rules, and experiential knowledge rules.

3. The method according to claim 1, characterized in that, The anomaly detection model described in S2 uses a combination of threshold rules and statistical anomaly detection, and the anomaly classification is dynamically adjusted based on the degree of deviation and the scope of influence.

4. The method according to claim 1, characterized in that, The RAG knowledge base described in S3 includes experimental SOPs, a collaborative knowledge base of abnormal cases, a fermentation protocol library, and reference documents / literature. The generated suggestions are accompanied by knowledge sources and links to reference documents.

5. The method according to claim 1, characterized in that, In S3, anomaly classification results, anomaly explanations, and suggestions are pushed to users via an instant messaging platform.

6. The method according to claim 1, characterized in that, The control strategies generated by S4 include, but are not limited to, process parameter adjustments such as feeding strategies, aeration strategies, stirring strategies, and temperature adjustment strategies. The strategy selection is dynamically adjusted based on the anomaly type and historical results.

7. The method according to claim 1, characterized in that, S4 also includes providing users with a visual interface that allows them to modify control strategies or send control commands directly.

8. A smart monitoring and control system for microbial fermentation processes based on a large language model, characterized in that, include: The data sensing module is used to access and analyze sensor data from the fermentation system in real time. The monitoring module is used to periodically detect data anomalies and issue tiered alarms. The diagnostic module is used to generate anomaly explanations and suggestions by combining the RAG knowledge base; The prediction module is used to predict parameter trends based on time series models and generate fitting quality indicators. The control module is used to execute control strategies; The coordination module is used to schedule tasks from various modules and generate task reports.

9. A computer-readable storage medium, characterized in that, The storage medium includes a stored program that, when executed, performs the method according to any one of claims 1-7.

10. An electronic device, characterized in that, It includes a memory and a processor, wherein when the processor executes a program in the memory, it triggers the method of any one of claims 1-7.

Images