Biological pesticide continuous production intelligent regulation method and system

Abstract

The application discloses a kind of biological pesticide continuous production intelligent regulation methods and systems, it is related to industrial automation technical field. Including have: S1: construct digital twin model, and construct virtual attack and defense test environment in the digital twin model;S2: by the digital twin model, test the control logic to be applied, corresponding test results are acquired and verified, while through reinforcement learning model, obtain the control logic after testing;S3: according to the control logic after testing, obtain the virtual verification result of the digital twin model corresponding, set feedforward compensation instruction, compensate the optimization instruction setting value of regulation and control the control logic after testing is set.This application can automatically optimize, solidify control strategy, avoid the problem that traditional method can only be found and repaired serious design defects when factory field debugging, shorten engineering debugging cycle, reduce the risk that production is interrupted, equipment is damaged even safety accident due to control logic defect.

Description

Technical Field

[0001] This invention relates to the field of industrial automation technology, specifically to an intelligent control method and system for continuous production of biological pesticides. Background Technology

[0002] As global agriculture shifts towards green and sustainable practices, biopesticides, with their advantages of being environmentally friendly, highly specific to targets, and having low residues, are gradually becoming an important alternative to chemical pesticides. However, the industrialization of biopesticides still faces severe challenges, with the core pain points concentrated in the "inefficiency" and "instability" of the production process—traditional intermittent production models are unable to meet the urgent needs of modern agriculture for biopesticides that are "highly active, low-cost, and scalable."

[0003] Currently, the production of biopesticides mainly relies on traditional batch fermentation and intermittent synthesis processes. These processes use common equipment and control logic from the chemical industry, separating microbial fermentation, product extraction, and formulation processing into independent batch operations. Under this model, the production process suffers from three common technical bottlenecks: First, process parameter control is rudimentary; key parameters such as temperature, pH, and dissolved oxygen concentration rely heavily on manual experience for adjustment, making it difficult to achieve precise dynamic responses. This leads to susceptibility to interference in microbial metabolism, resulting in significant batch-to-batch fluctuations in the conversion rate and activity of the target product. Second, mass and heat transfer efficiency is low; the bioheat and metabolic byproducts generated during fermentation are difficult to remove in a timely manner, easily causing local environmental deterioration, inhibiting microbial growth and activity, and even posing a risk of contamination. Third, production continuity is poor; non-productive operations such as equipment cleaning, sterilization, and feeding between batches account for more than 30%, which not only reduces equipment utilization but also further deteriorates process stability due to repeated start-ups and shutdowns.

[0004] Chinese invention patent application CN120993859A discloses an intelligent collaborative control system for slow-release fertilizer production. The system includes a data acquisition module that collects multi-source production data signals in real time from the batching, granulation, coating, and drying stages. These multi-source production data signals include at least raw material ratios and material temperatures. An intelligent optimization module receives the multi-source production data signals and optimizes them based on preset nutrient release targets. A production execution control module receives collaborative control command signals and generates process control signals accordingly. A digital twin module connects to both the data acquisition module and the intelligent optimization module, receiving multi-source production data signals to update the virtual production model's state and providing simulation and prediction feedback signals to the intelligent optimization module. A quality tracking module connects to both the data acquisition module and the production execution control module, receiving both multi-source production data signals and process control signals. This intelligent collaborative control system for slow-release fertilizer production can solve the problems of unstable production quality, high energy consumption, and lack of collaborative optimization in slow-release fertilizer production.

[0005] However, the above and similar technical solutions still have the following shortcomings: before being put into actual production, it is difficult to fully verify the correctness, coordination and robustness of the control logic of the intelligent regulation system for biological pesticides. As a result, design defects can only be exposed during the on-site debugging and operation phase in the factory. This will not only prolong the engineering debugging cycle, increase development costs and uncertainties, but may even lead to production interruption, equipment damage and even safety accidents. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent control method and system for continuous production of biological pesticides, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent control of continuous production of biological pesticides, comprising: S1: Virtual Environment Construction: Based on the equipment physical dynamics, microbial growth and metabolic dynamics, and material and energy flow of the actual continuous production process of biological pesticides, a digital twin model is constructed, and a virtual attack and defense test environment is constructed in the digital twin model; S2: Verification and solidification process: The control logic to be applied is tested using the digital twin model, and the corresponding test results are obtained and verified. At the same time, based on the verified test results, the policy network parameters of the control logic to be applied are adjusted using a reinforcement learning model to obtain the control logic after testing. S3: Coupled Deployment Control: Based on the post-test control logic, obtain the virtual verification result corresponding to the digital twin model, and based on the virtual verification result, set a feedforward compensation instruction, and based on the feedforward compensation instruction, compensate and control the optimization instruction setting value of the post-test control logic.

[0008] Furthermore, a virtual attack and defense testing environment is constructed within the digital twin model, including: S1.1: Constructing a digital twin model: Constructing a digital twin model through equipment physical dynamics modeling, microbial growth and metabolic dynamics modeling, and material and energy flow modeling; S1.2: Constructing the test environment: Based on the digital twin model, a dynamic simulation engine is set up, and a generator and discriminator are set up through a generative adversarial network architecture. At the same time, the test scenario parameters are set through the generator, and the stability score corresponding to the test scenario parameters is output through the discriminator.

[0009] Furthermore, constructing digital twin models includes: S1.1.1: Equipment physical dynamics modeling: Based on the physical boundaries of the continuous production line for biopesticides, unit equipment is set up, and a physical dynamics model of the equipment is constructed based on the collected physical equipment data; S1.1.2: Microbial growth and metabolic kinetics modeling: Based on the target microbial species corresponding to the biopesticide, the corresponding microbial species and process data are collected and obtained, and the microbial species and process data are combined with the unstructured model, structured model and metabolic network model corresponding to the target microbial species to construct a microbial growth and metabolic kinetics model; S1.1.3: Material and Energy Flow Modeling: Based on the full flow diagram with process unit nodes and connection and distribution nodes, set the corresponding overall material balance equation, component material balance equation and energy balance equation, and combine the collected historical process data with the overall material balance equation, component material balance equation and energy balance equation to construct the material and energy flow model.

[0010] Furthermore, the control logic obtained after the test includes: S2.1: Virtual Test: Based on the control logic to be applied, the generator sets the corresponding test scenario parameters, uses the test scenario parameters as input to the digital twin model, and outputs the corresponding simulation control commands. S2.2: Verification and Division: The simulation control command is used as the input of the discriminator, and the corresponding stability score is obtained. At the same time, the formal verification of the control logic to be applied is performed through the constructed formal tool, and the corresponding negative samples and counterexamples are determined according to the stability score and the formal verification results. S2.3: Defect Correction: Based on the negative samples and counterexamples, set up a corresponding failure case package, and use the failure case package and reinforcement learning model to train the control logic to be applied offline to obtain the corresponding training reward value. At the same time, use the training reward value to update and correct the policy network parameters of the control logic to be applied to obtain the control logic after testing.

[0011] Furthermore, based on the control strategy, decision rules, and security constraints corresponding to the control logic to be applied, corresponding system states, transformation relationships, and attributes to be verified are set. Then, through model checking tools or theorem provers, the system states, transformation relationships, and attributes to be verified are combined to construct a formal tool.

[0012] Furthermore, the formalization tool is used to perform formal verification on the control logic to be applied, and the control logic to be applied that fails formal verification is set as a negative example, while the control logic to be applied that passes formal verification is set as a positive example.

[0013] Furthermore, the stability score is compared with a preset score threshold. Test case data corresponding to stability scores below the preset score threshold are set as negative samples, and test case data corresponding to stability scores not lower than the preset score threshold are set as positive samples.

[0014] Furthermore, the optimized instruction setting value for compensating and regulating the control logic after the test includes: S3.1: Logic Interconnection: Through industrial real-time Ethernet, the industrial control system and the edge server of the control logic after testing interact in real time with process data. At the same time, based on the real-time process data, the corresponding optimization instruction setting value is determined, and the optimization instruction setting value is sent to the process control layer in the industrial control system through the communication protocol. S3.2: Synchronous Operation: Based on the real-time input data of the actual biopesticide production line, the digital twin model is simulated and deduced, and virtual verification results are collected. At the same time, based on the virtual verification results, the post-test control logic generates corresponding feedforward compensation instructions, and feedforward adjustment is performed on the optimization instruction set value of the actual biopesticide production line according to the feedforward compensation instructions.

[0015] Furthermore, the engineered and packaged post-test control logic is set in the production scheduling and control layer or the monitoring control layer of the industrial control system, and is located above the process control layer.

[0016] A continuous production intelligent control system for biological pesticides uses any one of the above-mentioned methods for continuous production intelligent control of biological pesticides.

[0017] Compared with the prior art, the beneficial effects of the present invention are: Firstly, before the control system is applied to the actual production line, this invention pre-constructs a digital twin model and a virtual attack and defense test environment, and forms a case package with the failure cases found in the test. Through reinforcement learning model, the policy network parameters of the control logic are iteratively trained and corrected offline, thereby automatically optimizing and solidifying the control strategy. This avoids the problem that serious design defects can only be found and repaired during on-site debugging in the factory, which is a problem of traditional methods. It shortens the engineering debugging cycle and reduces the risk of production interruption, equipment damage or even safety accidents caused by control logic defects. Secondly, this invention integrates equipment physical dynamics, microbial growth and metabolic dynamics, and material and energy flow models to simulate the production process and its dynamic response. At the same time, it automatically generates extreme working conditions and disturbance scenarios through a generative adversarial network architecture to conduct extreme stress tests on the control logic to be applied. Combined with formal verification tools, it proves the safety specifications of the control logic, thereby efficiently and risk-free exposing and locating potential design defects in the robustness, safety, and coordination of the control logic. Thirdly, this invention deploys the verified and solidified post-test control logic to the upper layer of the industrial control system. While interacting with the underlying process control system in real time, it performs simulation and deduction synchronously through a digital twin model. Based on the corresponding virtual verification results, it generates corresponding feedforward compensation instructions, thereby enhancing the adaptability and stability of the production system in the face of uncertainties and internal dynamic changes. This helps to maintain the microbial metabolic process in the optimal state and improve the conversion rate of the target product and batch-to-batch consistency. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the intelligent control method for continuous production of biological pesticides in this invention. Figure 2 This is a flowchart illustrating the virtual environment construction method in this invention; Figure 3 This is a schematic diagram illustrating the construction process of the digital twin model in this invention; Figure 4 This is a schematic diagram of the verification curing treatment method in this invention; Figure 5 This is a flowchart illustrating the coupled deployment and control method in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] refer to Figure 1 This embodiment provides an intelligent control method for continuous production of biopesticides, which specifically includes the following steps: Step S1: Virtual Environment Construction. This involves determining the corresponding equipment physical dynamics, microbial growth and metabolic dynamics, and material and energy flows based on the actual continuous production process of biopesticides. These determined dynamics are then combined to construct a corresponding digital twin model. Simultaneously, a virtual attack and defense testing environment is built using a generative adversarial network architecture to simulate various disturbances and extreme conditions during the production process.

[0021] Step S2: Verification and Consolidation. This involves using the digital twin model with the virtual attack and defense testing environment built in Step S1 to test the application's control logic and obtain the corresponding test results. Simultaneously, the obtained test results are formally verified, and based on the verification results, the test sample data is divided into negative and positive samples.

[0022] Furthermore, based on the identified negative samples, the control logic of the application is adjusted using the configured reinforcement learning model to modify the corresponding policy network parameters. The post-test control logic is then obtained based on the adjusted policy network parameters.

[0023] Step S3: Coupled Deployment and Control. The post-test control logic obtained in Step S2 is deployed to the actual continuous production system of biopesticides to enable system operation. Simultaneously, during actual operation, the virtual verification results corresponding to the digital twin model constructed in Step S1 are collected. Based on the collected virtual verification results, corresponding feedforward compensation instructions are set to compensate and control the optimized instruction setpoints of the post-test control logic according to the set feedforward compensation instructions.

[0024] This embodiment also provides an intelligent control system for continuous production of biological pesticides, which uses the aforementioned intelligent control method for continuous production of biological pesticides.

[0025] In this embodiment, a corresponding virtual environment is constructed by building a digital twin model and a virtual attack and defense testing environment. (Reference) Figure 2 and Figure 3 This embodiment provides a method for constructing a virtual environment, which specifically includes the following steps: Step S1.1: Constructing a digital twin model. This involves constructing corresponding mathematical models based on the physical equipment in the production process (e.g., fermenters, agitators, heat exchangers, pumps, valves, and filtration units) to model the equipment's physical dynamics. Simultaneously, based on the characteristics of the microbial strains required for biopesticides, corresponding kinetic models are established for growth, substrate consumption, synthesis of target products (e.g., active ingredients in biopesticides), and the accumulation of potential byproducts and inhibitors, to model microbial growth and metabolic kinetics. Furthermore, by integrating the constructed mathematical and kinetic models through the flow balance of materials (e.g., culture media, products, waste liquid) and the conservation of energy (e.g., thermal energy, mechanical energy, cooling load), material and energy flow modeling is achieved. In other words, through equipment physical dynamics modeling, microbial growth and metabolic kinetics modeling, and material and energy flow modeling, a corresponding digital twin model is constructed. (It is worth noting that the specific construction equations for each model are not specifically described in this embodiment; they can be set according to actual needs. Therefore, this embodiment only provides illustrative examples.) Specifically, as follows: Step S1.1.1: Equipment Physical Dynamics Modeling. This involves determining the static parameters of each key piece of equipment (e.g., fermenters, pumps, and heat exchangers) based on their design drawings, equipment data sheets, and operation manuals. These parameters include, but are not limited to, geometric dimensions, material properties, rated power, flow-head curve, and heat transfer coefficient. This process is used to collect corresponding physical equipment data. Simultaneously, the continuous biopesticide production line is physically delineated into multiple unit devices (e.g., including culture medium preparation tanks, continuous fermenters, plate heat exchangers, centrifuges, membrane filtration units, and product storage tanks, as well as the pumps, valves, and piping networks connecting these devices). The input, output, and internal states of each unit device are clearly defined.

[0026] Furthermore, based on the decomposed unit devices and the collected physical device data, the mathematical equations corresponding to each unit device are determined to construct the corresponding device physical dynamics model. It is worth noting that this can be specifically configured according to the actual unit devices; therefore, this embodiment does not provide a detailed explanation but only an illustrative one. For example, when the unit device is a fermenter, its corresponding mathematical formulas include, but are not limited to, the transient mass balance equation (e.g., rate of change of cell concentration = growth rate - discharge rate - death rate) and the heat balance equation (e.g., rate of change of temperature inside the tank = heat generated by microbial metabolism + heat energy converted from stirring mechanical energy + heat carried in / out by air intake - heat removed through the jacket or coil - heat loss to the environment).

[0027] Step S1.1.2: Microbial Growth and Metabolic Kinetic Modeling. This involves determining the basic growth and metabolic kinetic parameters corresponding to the target microbial species used in the biopesticide, including but not limited to maximum specific growth rate, substrate yield coefficient, product yield coefficient, and maintenance metabolic coefficient, to collect relevant microbial species and process data. Simultaneously, based on the macroscopic state variables of the target microbial cell population (e.g., total cell concentration, substrate concentration, and product concentration), and combined with the Monod equation and the Luedeking-Piret equation, a corresponding unstructured model is constructed. The concentrations of pseudo-components representing different physiological functions within the target microbial cells are used as state variables to elucidate the changes in key components (e.g., RNA, DNA, proteins, and energy storage substances) and their impact on growth and product synthesis, thereby constructing a corresponding structured model. Finally, based on the reconstruction of the genome-scale metabolic network (including all known metabolic reactions, enzymes, and genes of the target microbial species), flux balance analysis is used to determine the corresponding metabolite fluxes, thus constructing a corresponding metabolic network model. In other words, the collected microbial strains and process data are combined with the constructed unstructured models, structured models, and metabolic network models to construct corresponding microbial growth and metabolic kinetic models.

[0028] Step S1.1.3: Material and Energy Flow Modeling. This involves collecting historical process data, including but not limited to sensor data and control signal data, based on the production plant's distributed control system, monitoring and data acquisition system, and laboratory information management system, to obtain corresponding historical and real-time operational data. Sensor data includes, but is not limited to, temperature, pressure, flow rate, rotational speed, pH, and dissolved oxygen data; control signal data includes, but is not limited to, valve opening data and motor frequency data. Simultaneously, based on the production plant's piping and instrumentation diagrams and process flow diagrams, a corresponding full flow diagram is determined, including process unit nodes and connection / distribution nodes. Process unit nodes are defined as equipment with corresponding physical dynamic models, such as fermenters, heat exchangers, separators, and storage tanks. Connection / distribution nodes include, but are not limited to, pipes, mixers, distributors, and valves.

[0029] Furthermore, based on the full flow diagram with process unit nodes and connection and distribution nodes, the corresponding overall material balance equation (e.g., rate of mass accumulation in the control body = total inflow mass flow rate - total outflow mass flow rate), component material balance equation (e.g., accumulation rate of substrate components = inflow rate - outflow rate + reaction generation rate (or - reaction consumption rate)) and energy balance equation (e.g., rate of energy accumulation in the control body = energy flowing in with the material flow - energy flowing out with the material flow + heat input (or - heat output) + work input / output + heat of reaction) are determined to construct the corresponding material and energy flow model.

[0030] Step S1.2: Construct the test environment. This involves building a test framework within the digital twin model obtained in Step S1.1, and generating a corresponding extreme test scenario library using this framework. Specifically, based on the digital twin model constructed in Step S1.1, a corresponding dynamic simulation engine is set up. Simultaneously, a generative adversarial network architecture is used to set up corresponding generators and discriminators. The generators are then used to set the corresponding test scenario parameters, including multi-dimensional extreme test scenario parameter vectors, such as [at time t1, the feed flow rate suddenly increases by 30% and lasts for 5 minutes; at time t2, the pH sensor output signal remains fixed at 6.5; at time t3, the stirrer power coefficient decreases by 20%].

[0031] Furthermore, based on the test scenario parameters set by the generator, the dynamic simulation engine (i.e., the digital twin model) performs simulation to output the evolution trajectory of the corresponding process variables. At the same time, the discriminator outputs the stability score corresponding to the test scenario parameters based on the output evolution trajectory.

[0032] In this embodiment, the control logic of the application is tested using a digital twin model with a test framework set in step S1.2 to obtain the tested control logic. (See reference...) Figure 4 This embodiment provides a method for verifying curing treatment, which specifically includes the following steps: Step S2.1: Virtual Testing. This involves connecting the control logic to be applied (i.e., verified) to the digital twin model with the test framework set up in Step S1.2. Simulation is then performed using the control logic and the digital twin model to obtain the corresponding simulation control commands. In other words, the digital twin model is virtually controlled by the control logic to be applied to perform the corresponding simulation. During the simulation, the test framework within the digital twin model is activated to test the control logic to be applied.

[0033] Furthermore, based on the control logic to be applied, the generator in the test framework sets the corresponding test scenario parameters, which are structured instruction sequences with timestamps. For example, "At the 100th second of the simulation time, increase the 'fermenter feed flow rate' setpoint by 50% within 2 seconds; simultaneously, starting from the 105th second, fix the 'dissolved oxygen sensor' reading output to 80% of the current value." The set test scenario parameters are then used as input to the digital twin model for simulation and deduction, and the control logic to be applied is run synchronously to output the corresponding simulation control commands (such as valve opening and motor speed).

[0034] Step S2.2: Verification Division. The acquired simulation control commands are used as input to the discriminator in the test framework, and the corresponding stability score is output. Simultaneously, based on the control strategy, decision rules, and safety constraints corresponding to the control logic to be applied, the corresponding system state, transition relationships, and attributes to be verified are set. That is, the core state variables corresponding to the control logic to be applied (e.g., "tank temperature zone": {too low, normal, too high}; "pH state": {acid, neutral, alkaline}), the behavioral rules corresponding to the control logic to be applied (e.g., always, if (tank temperature_zone = too high) then in the next cycle, the control logic must (cooling valve opening = increase)), and the functional and safety specifications to be verified corresponding to the control logic to be applied (e.g., safety attribute: "Never ((pH_state = acid) and (ammonia inlet valve = closed))", indicating that it is absolutely forbidden to close the alkali supply valve when in an acidic state). In other words, by using a model checking tool (such as NuSMV or UPPAAL) or a theorem prover, the system state, transformation relationship, and attribute to be verified are combined to construct the corresponding formal verification tool.

[0035] Furthermore, by constructing formal tools, formal verification of the application control logic is performed, that is, testing the functional and security specifications corresponding to the application control logic. In other words, through formal verification of the application control logic, those application control logics that fail formal verification are identified and set as negative examples; conversely, those that pass formal verification are set as positive examples. Simultaneously, the stability score obtained from the discriminator output is compared with a preset scoring threshold to determine stability scores below the preset threshold. In other words, through stability score verification of the application control logic, stability scores below the preset threshold are identified. Based on stability scores below the preset threshold, the corresponding test case data (including detailed data from test rounds, such as attack scripts, simulation data, and evaluation reports) are set as negative samples. Conversely, test case data corresponding to stability scores above the preset threshold are set as positive samples.

[0036] Step S2.3: Defect Correction. Based on the counterexamples and negative samples identified in Step S2.2, obtain case data within the corresponding preset time window, and structure and store this case data as a corresponding failure case package. This package includes an environmental snapshot (i.e., the test scenario parameters set by the generator), a process trajectory (i.e., the state evolution sequence of the digital twin model during the test process, such as timestamps, temperature, pressure, concentration, pH, etc.), a decision log (i.e., the sequence of action commands output by the control logic to be applied, such as timestamps, valve opening setpoints, and motor speed setpoints), evaluation results (i.e., the stability score output by the discriminator and the corresponding verification results from the formal tools), and metadata (i.e., the test time, the version number of the control logic to be applied, and the version number of the digital twin model).

[0037] Furthermore, based on the obtained failure case package and reinforcement learning model, the control logic to be applied is trained offline, and the corresponding training reward value is obtained. Simultaneously, through backpropagation and the obtained training reward value, the policy network parameters of the control logic to be applied are updated and corrected to obtain the corresponding corrected policy network parameters. In other words, based on the obtained corrected policy network parameters, the original policy network parameters of the control logic to be applied are corrected to obtain the corrected control logic to be applied, i.e., the corresponding post-test control logic.

[0038] In this embodiment, based on the post-test control logic obtained in step S2.3, the virtual verification result corresponding to the digital twin model is acquired, and the running instructions corresponding to the post-test control logic are compensated and adjusted according to the acquired virtual verification result. (See reference...) Figure 5 This embodiment provides a coupled deployment control method, which specifically includes the following steps: Step S3.1: Logic Integration. This involves engineering-encapsulating the post-test control logic obtained in Step S2.3, and deploying this engineered post-test control logic to the industrial control system of the actual biopesticide production line using a set security protocol. Specifically, the engineered post-test control logic is deployed to the industrial edge server or industrial gateway of the actual biopesticide production line, and is set in the production scheduling and control layer or supervisory control layer of the industrial control system, above the process control layer.

[0039] Furthermore, the industrial control system interacts with the edge server of the post-test control logic in real time via industrial real-time Ethernet. Simultaneously, the post-test control logic determines the corresponding optimized instruction setpoints based on the interacting real-time process data. Then, through a pre-defined communication protocol (such as OPC UA), the determined optimized instruction setpoints are sent to the data blocks or variables in the process control layer of the industrial control system.

[0040] Step S3.2: Synchronous Operation. This involves simulating the digital twin model based on real-time input data from the actual biopesticide production line (including material feed, environmental conditions, and optimized command settings corresponding to the post-test control logic) to collect corresponding virtual verification results. It is worth noting that during the simulation of the digital twin model, the control logic used is the post-test control logic obtained in step S2.3.

[0041] Furthermore, based on the collected virtual verification results, the post-test control logic generates corresponding feedforward compensation instructions, and based on the generated feedforward compensation instructions, it performs feedforward control on the optimized instruction set values ​​of the actual operation in the actual biological pesticide production line.

[0042] 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 embodiments and their equivalents.

Claims

1. A method for intelligent regulation of continuous production of biopesticides, characterized in that, Including: S1: Virtual Environment Construction: Based on the equipment physical dynamics, microbial growth and metabolic dynamics, and material and energy flow of the actual continuous production process of biological pesticides, a digital twin model is constructed, and a virtual attack and defense test environment is constructed in the digital twin model; S2: Verification and solidification process: The control logic to be applied is tested using the digital twin model, and the corresponding test results are obtained and verified. At the same time, based on the verified test results, the policy network parameters of the control logic to be applied are adjusted using a reinforcement learning model to obtain the control logic after testing. S3: Coupled Deployment Control: Based on the post-test control logic, obtain the virtual verification result corresponding to the digital twin model, and based on the virtual verification result, set a feedforward compensation instruction, and based on the feedforward compensation instruction, compensate and control the optimization instruction setting value of the post-test control logic.

2. The method according to claim 1, wherein, A virtual attack and defense testing environment is constructed within the digital twin model, including: S1.1: Constructing a digital twin model: Constructing a digital twin model through equipment physical dynamics modeling, microbial growth and metabolic dynamics modeling, and material and energy flow modeling; S1.2: Constructing the test environment: Based on the digital twin model, a dynamic simulation engine is set up, and a generator and discriminator are set up through a generative adversarial network architecture. At the same time, the test scenario parameters are set through the generator, and the stability score corresponding to the test scenario parameters is output through the discriminator.

3. The method according to claim 2, wherein, Building a digital twin model includes: S1.1.1: Equipment physical dynamics modeling: Based on the physical boundaries of the continuous production line for biopesticides, unit equipment is set up, and a physical dynamics model of the equipment is constructed based on the collected physical equipment data; S1.1.2: Microbial growth and metabolic kinetics modeling: Based on the target microbial species corresponding to the biopesticide, the corresponding microbial species and process data are collected and obtained, and the microbial species and process data are combined with the unstructured model, structured model and metabolic network model corresponding to the target microbial species to construct a microbial growth and metabolic kinetics model; S1.1.3: Material and Energy Flow Modeling: Based on the full flow diagram with process unit nodes and connection and distribution nodes, set the corresponding overall material balance equation, component material balance equation and energy balance equation, and combine the collected historical process data with the overall material balance equation, component material balance equation and energy balance equation to construct the material and energy flow model.

4. The method according to claim 1, wherein, The control logic obtained after the test includes: S2.1: Virtual Test: Based on the control logic to be applied, the generator sets the corresponding test scenario parameters, uses the test scenario parameters as input to the digital twin model, and outputs the corresponding simulation control commands. S2.2: Verification and Division: The simulation control command is used as the input of the discriminator, and the corresponding stability score is obtained. At the same time, the formal verification of the control logic to be applied is performed through the constructed formal tool, and the corresponding negative samples and counterexamples are determined according to the stability score and the formal verification results. S2.3: Defect Correction: Based on the negative samples and counterexamples, set up a corresponding failure case package, and use the failure case package and reinforcement learning model to train the control logic to be applied offline to obtain the corresponding training reward value. At the same time, use the training reward value to update and correct the policy network parameters of the control logic to be applied to obtain the control logic after testing.

5. The method according to claim 4, wherein, Based on the control strategy, decision rules, and security constraints corresponding to the control logic to be applied, the corresponding system state, transformation relationship, and attribute to be verified are set, and the system state, transformation relationship, and attribute to be verified are combined through model checking tools or theorem provers to construct a formal tool.

6. The intelligent regulation method for continuous production of a biopesticide according to claim 4 or 5, characterized in that, The formalization tool is used to perform formal verification on the control logic to be applied, and the control logic to be applied that fails formal verification is set as a negative example, while the control logic to be applied that passes formal verification is set as a positive example.

7. The method according to claim 4, wherein, The stability score is compared with a preset score threshold. Test case data corresponding to stability scores below the preset score threshold are set as negative samples, and test case data corresponding to stability scores above the preset score threshold are set as positive samples.

8. The method of claim 1, wherein the method is characterized by, The optimized instruction settings for compensating and regulating the control logic after the test include: S3.1: Logic Interconnection: Through industrial real-time Ethernet, the industrial control system and the edge server of the control logic after testing interact in real time with process data. At the same time, based on the real-time process data, the corresponding optimization instruction setting value is determined, and the optimization instruction setting value is sent to the process control layer in the industrial control system through the communication protocol. S3.2: Synchronous Operation: Based on the real-time input data of the actual biopesticide production line, the digital twin model is simulated and deduced, and virtual verification results are collected. At the same time, based on the virtual verification results, the post-test control logic generates corresponding feedforward compensation instructions, and feedforward adjustment is performed on the optimization instruction set value of the actual biopesticide production line according to the feedforward compensation instructions.

9. The method according to claim 8, wherein, The engineered and packaged post-test control logic is set in the production scheduling and control layer or the monitoring control layer of the industrial control system, and is located above the process control layer.

10. A biological pesticide continuous production intelligent regulation system, characterized in that, The method for intelligent control of continuous production of biological pesticides according to any one of claims 1-9 was used.

Images