Chemical process collaborative optimization method and device based on distributed training model deployment system

Abstract

The embodiment of the invention relates to the field of artificial intelligence and chemical process optimization, and discloses a chemical process collaborative optimization method and device based on a distributed training model deployment system, and the method comprises the steps: generating an initial process scheme through a second chemical process special training model in response to a target compound generation instruction; obtaining an execution result of executing the initial process scheme by the edge device, and generating execution result evaluation data based on the execution result; performing optimization training on the second chemical process special training model based on the execution data of the initial process scheme and the execution result evaluation data to obtain an optimization process scheme meeting preset evaluation data and corresponding optimization model parameters; and when a feedback instruction is received, sending the target process scheme and / or the target model parameters to a cloud end, so as to update the general chemical training model and / or the first chemical process special training model. And the optimization process scheme and the optimization model parameters are selectively sent to the cloud, so that knowledge sharing is realized and privacy controllability is ensured.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of artificial intelligence and chemical process optimization, and in particular to a chemical process collaborative optimization method and device based on a distributed training model deployment system. Background Art

[0002] With the advancement of technology, chemical process equipment has achieved a high degree of automation. For example, existing liquid chromatography separation technology and its instrumentation are highly automated. However, users still need to possess basic knowledge of chromatography theory and conduct a series of experiments to determine essential information for the separation and purification of target compounds, such as solvent systems and Rf values. After separation and purification, the method for the target compound is typically stored on a single liquid chromatograph for future review by that instrument's user, but it cannot be shared with users in other organizations. When different users isolate and purify the same or similar target compounds, they generally need to re-run experiments to determine the essential information due to the isolation and inability to share this information. This results in duplication of effort and waste of resources.

[0003] Chinese patent CN103055541B discloses a method and apparatus for establishing and using a networked separation and purification method database, and provides a chromatograph for performing separation and / or purification of a target compound system. The chromatograph may include: a communication unit for communicating with a terminal device and a networked database server; and a control unit for controlling the communication unit to receive a characteristic information query about the target compound system from the terminal device and retrieve and download a separation and purification method data record matching the characteristic information of the target compound system from the separation and purification method database in the networked database server, wherein the chromatograph performs separation and / or purification of the target compound system based on the separation and purification method data record.

[0004] Chinese patent CN104504152B discloses a device and method for improving chemical process efficiency and promoting the sharing of chemical information, which can guide and motivate scientific researchers and institutions to develop and share more efficient chemical processes. Its technical solution is: by executing and evaluating the relevant chemical processes for analyzing target compounds or target compound systems, and based on Internet technology, providing scientific researchers with an App application and website installed on mobile devices with social and electronic transaction functions, it can achieve the sharing, transaction and evaluation of the relevant chemical processes and chemical information of publicly available compounds. Through the electronic transaction system, it guides and motivates users to share chemical information and experience, thereby developing more efficient chemical processes, reducing resource waste, and promoting R&D efficiency, especially improving the R&D efficiency of unknown innovative chemical processes and compounds.

[0005] Based on the above solution, users can download, execute and verify chemical information related to existing target compounds or target compound systems through the trading and sharing platform and chemical instruments or devices provided by CN103055541B and CN104504152B, and can guide and motivate users to develop and use more efficient chemical processes, thereby improving R&D and production efficiency.

[0006] However, these solutions require sufficient data and cannot achieve truly automated process generation and execution. Therefore, a solution is urgently needed to address these issues, enabling autonomous process generation and execution, further enabling unmanned operation, and making it possible for robots to collaborate with process models to complete / execute chemical processes. Summary of the Invention

[0007] The purpose of the present invention is to at least provide a chemical process collaborative optimization method and device based on a distributed training model deployment system, which can at least solve the problems in the existing method of optimizing chemical processes, such as the inability to adapt to different chemical process equipment, the inability to achieve privacy control, the inability to meet the real-time requirements of offline chemical process equipment, and the insufficient accuracy of the optimized chemical process. At least it can achieve the technical effect of adapting to different chemical process equipment, achieving privacy control, meeting the real-time requirements of offline chemical process equipment, and improving the accuracy of the optimized chemical process.

[0008] To solve the above technical problems, at least one embodiment of the present application provides a chemical process collaborative optimization method based on a distributed training model deployment system, the distributed training model deployment system includes a cloud and at least one edge device, the cloud is provided with a general chemical training model and / or a first chemical process special training model, the edge device is used for chemical process execution, and is provided with a second chemical process special training model, an optional communication connection between the cloud and at least one edge device, the method is applied to the edge device, the method includes: generating a chemical process (including a synthesis process and / or a separation and purification process) instruction in response to a target compound, using the second chemical process special training model to generate an initial process plan; obtaining the execution result of the initial process plan executed by the edge device, and based on the execution result, The execution result generates execution result evaluation data; the second chemical process special training model is optimized and trained based on the execution data of the initial process plan and the execution result evaluation data to obtain the optimized process plan and its corresponding optimized model parameters that meet the preset evaluation data; when a feedback instruction is received, the target process plan and / or target model parameters to be uploaded are determined based on the iterative selection corresponding to the feedback instruction and the optimized process plan and / or optimized model parameters, and the target process plan and / or target model parameters are sent to the cloud to update the general chemical training model and / or the first chemical process special training model, wherein the optimization degree of the target process plan and / or target model parameters is less than or equal to the optimized process plan and / or optimized model parameters.

[0009] At least one embodiment of the present application further provides a distributed training model deployment system, comprising: a cloud, provided with a general chemical training model and / or a first chemical process-specific training model; an edge device, used for chemical process execution, and provided with a second chemical process-specific training model, and an optional communication connection between the cloud and at least one edge device; wherein the edge device is further used to: generate an initial process plan using the second chemical process-specific training model in response to a target compound generation instruction; obtain the execution result of the initial process plan executed by the edge device, and generate execution result evaluation data based on the execution result; and generate execution result evaluation data based on the execution data of the initial process plan and the execution result evaluation data. The second chemical process special training model is optimized and trained based on the estimated data to obtain an optimized process scheme and its corresponding optimized model parameters that meet the preset evaluation data; when a feedback instruction is received, the target process scheme and / or target model parameters to be uploaded are determined based on the iterative selection corresponding to the feedback instruction and the optimized process scheme and / or optimized model parameters, and the target process scheme and / or target model parameters are sent to the cloud to update the general chemical training model and / or the first chemical process special training model, wherein the optimization degree of the target process scheme and / or target model parameters is less than or equal to the optimized process scheme and / or optimization model parameters.

[0010] At least one embodiment of the present application also provides a chemical process collaborative optimization device based on a distributed training model deployment system, which is arranged on an edge device, the edge device is used for chemical process execution, and is provided with a second chemical process special training model. The device includes: a response module, which is used to generate an initial process plan using the second chemical process special training model in response to a target compound generation instruction; an acquisition module, which is used to obtain the execution result of the initial process plan executed by the edge device, and generate execution result evaluation data based on the execution result; a training module, which is used to optimize and train the second chemical process special training model based on the execution data of the initial process plan and the execution result evaluation data to obtain an optimized process plan and its corresponding optimized model parameters that meet the preset evaluation data; a feedback module, which is used to determine the target process plan and / or target model parameters to be uploaded based on the iterative selection corresponding to the feedback instruction and the optimized process plan and / or optimized model parameters when receiving a feedback instruction, and send the target process plan and / or target model parameters to the cloud to update the general chemical training model and / or the first chemical process special training model, wherein the optimization degree of the target process plan and / or target model parameters is less than or equal to the optimized process plan and / or optimized model parameters.

[0011] At least one embodiment of the present application also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the above-mentioned chemical process collaborative optimization method based on the distributed training model deployment system.

[0012] At least one embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned chemical process collaborative optimization method based on a distributed training model deployment system.

[0013] An embodiment of the present application provides a chemical process collaborative optimization method based on a distributed training model deployment system. The distributed training model deployment system includes a cloud and at least one edge device. The cloud is provided with a general chemical training model and / or a first chemical process-specific training model. The edge device is used for chemical process execution and is provided with a second chemical process-specific training model. An optional communication connection is provided between the cloud and at least one edge device. The method is applied to the edge device, and the method includes: in response to a target compound generation instruction, generating an initial process plan using the second chemical process-specific training model; obtaining the execution result of the initial process plan executed by the edge device, and generating execution result evaluation data based on the execution result; optimizing and training the second chemical process-specific training model based on the execution data of the initial process plan and the execution result evaluation data to obtain an optimized process plan and its corresponding optimized model parameters that meet the preset evaluation data; when receiving a feedback instruction, sending the optimized process plan and / or the optimized model parameters to the cloud to update the general chemical training model and / or the first chemical process-specific training model.

[0014] Using the local model deployed locally on the edge device to generate the initial process plan can meet the real-time requirements of process optimization when the edge device is offline. It can also make full use of the learning ability and efficiency of the local model to achieve accurate and efficient generation of the initial process plan, that is, using the locally deployed chemical process special model to autonomously generate the chemical process, without the need for users to carry out complex process plan design, thereby improving the efficiency of edge devices in executing chemical processes; then, using the edge device's own execution capability, the execution result of the initial process plan is obtained, effectively improving the degree of automation of the edge device's process generation and execution. During the application process, the user only needs to input the initial instruction and the edge device can complete the corresponding process generation and execution by itself. Execution; By evaluating and analyzing the execution results of the local model, it can be determined whether the execution result corresponding to the initial process plan is an execution result that meets the preset requirements, so as to evaluate and improve the initial process plan through the execution result evaluation data of the execution result, and to train the local model based on the execution data and the execution result evaluation data of the initial process plan to optimize the local model; through the closed-loop processing process of "generation-execution-evaluation-training", while meeting the real-time requirements of the edge device, the learning ability of the local model is improved, and the optimized local model and the edge device can be matched in real time. It is determined that the process plan optimized based on the local model is highly matched with the current edge device, and the accuracy of the optimized process plan is improved. In addition, by selectively sending the target process plan and / or target model parameters to the cloud, and the optimization degree of the target process plan and / or target model parameters is less than or equal to the optimized process plan and / or optimized model parameters, the cloud and edge devices can collaboratively share knowledge, and the relationship between knowledge openness and knowledge competitiveness can be balanced, that is, promoting the advancement of chemical processes while ensuring the competitiveness of their own technologies, thereby achieving both knowledge sharing and privacy controllability.

[0015] In some optional embodiments, obtaining the execution result of the initial process plan executed by the edge device includes: sending the initial process plan to the cloud so that the cloud optimizes the initial process plan based on the general chemical training model and / or the first chemical process-specific training model to obtain an optimized initial process plan; receiving the optimized initial process plan sent by the cloud, and obtaining the execution result of the edge device executing the optimized initial process plan; and / or receiving process parameter adjustment data corresponding to the process parameter adjustment instruction, optimizing and adjusting the initial process plan based on the process parameter adjustment data to obtain an adjusted and optimized initial process plan, and obtaining the execution result of the edge device executing the adjusted and optimized initial process plan. Sending the initial process plan to the cloud and optimizing the initial process plan using the general chemical training model and the first chemical process-specific training model on the cloud can fully utilize the ability of the general chemical training model and the first chemical process-specific training model on the cloud to have a wider range of knowledge, optimize the initial process plan, fully utilize the knowledge open in the cloud, make the optimized initial process plan more accurate, and achieve the technical effect of collaborative knowledge sharing between the cloud and the edge device. Alternatively, the user can input process parameter adjustment data for optimization through process parameter adjustment instructions to optimize the initial process plan, making full use of the user's professional and technical experience, making the optimized initial process plan more accurate, and providing a basis for forming the user's core competitiveness in subsequent training.

[0016] In some optional embodiments, before the cloud optimizes the initial process plan based on the general chemical training model and / or the first chemical process special training model, it also includes: the cloud obtains the optimization model parameters sent by other edge devices when receiving the feedback instruction; the cloud updates the general chemical training model and / or the first chemical process special training model based on the optimization model parameters; and when receiving the initial process plan, the initial process plan is optimized based on the updated general chemical training model and / or the first chemical process special training model. The feedback instruction is used to selectively optimize and update the cloud model based on the optimization model parameters of other edge devices to improve the learning ability of the cloud model and the real-time performance of the cloud model, so that the initial process plan of the edge device is optimized based on the updated cloud model, thereby improving the real-time performance and accuracy of the initial process plan.

[0017] In some optional embodiments, optimizing and training a second chemical process-specific training model based on the execution data and execution result evaluation data of the initial process plan includes: obtaining first historical execution data and first historical execution result evaluation data stored locally on an edge device, as well as the execution data and execution result evaluation data of the initial process plan; and / or receiving process parameter adjustment data corresponding to a process parameter adjustment instruction; and offline optimizing and training the second chemical process-specific training model based on the first historical execution data, the first historical execution result evaluation data, the execution data of the initial process plan, the execution result evaluation data of the initial process plan, and / or the process parameter adjustment data. Optimizing the training model using the first historical execution data and first historical execution result evaluation data stored locally on the edge device, as well as the execution data and execution result evaluation data of the initial process plan, ensures that model training is based on actual, historical data. This data can more accurately reflect the actual production process, thereby improving the model's predictive accuracy and reliability. Receiving the process parameter adjustment data corresponding to the process parameter adjustment instruction to optimize the training model enables the model to adjust according to real-time production needs and conditions. This dynamic adjustment capability makes the model more flexible and adaptable to changes in the production process, thereby improving the accuracy of the process plan. Offline optimization and training of the model allows for model improvements without disrupting normal production. This reduces the risks associated with online model adjustments while ensuring the model's stability and reliability in real-world applications. Continuous optimization and training of the model allows it to better cope with various uncertainties and changes, enhancing its robustness and adaptability.

[0018] In some optional embodiments, optimization training of a second chemical process-specific training model based on execution data and execution result evaluation data of an initial process plan includes: determining the process complexity corresponding to target compound generation instructions based on the initial process plan and / or execution result evaluation data; generating a distributed training strategy corresponding to the optimization training when the process complexity meets preset conditions; the distributed training strategy including at least one other edge device participating in the training and training content corresponding to each other edge device, the training content including at least an exit condition corresponding to the optimization training; and transmitting the training content to each other edge device via the cloud, so that each other edge device performs optimization training based on the training content. Generating a distributed training strategy based on process complexity enables customized training plans to be developed for different process complexities. This avoids the use of a one-size-fits-all training strategy, allowing each training task to be optimized based on its unique characteristics, thereby improving training efficiency and quality. The distributed training strategy, including at least one other edge device participating in the training and training content corresponding to each other edge device, facilitates the rational and dynamic allocation of computing resources. By distributing training tasks across multiple edge devices, the computing power of each edge device can be fully utilized, accelerating the training process and improving training efficiency. Training content at least includes exit conditions corresponding to optimized training, providing each edge device with clear training objectives and stopping criteria. Edge devices can autonomously train and make decisions based on these exit conditions, ensuring controllable and stable training. Clear exit conditions also prevent unnecessary training, saving computing resources and time. Training content is sent to all other edge devices via the cloud, enabling centralized management and distribution of training data. This ensures consistency and accuracy in training content and facilitates unified monitoring and management of the training process. Edge devices can collaborate on training through the cloud, sharing information and resources to further improve training effectiveness.

[0019] In some optional embodiments, each other edge device performs optimization training based on the training content, including: each other edge device obtains locally stored historical execution results and historical execution result evaluation data; and performs optimization training based on the locally stored historical execution results and historical execution result evaluation data and the training content. It should be understood that the locally stored historical data is closely related to the specific production environment in which the edge device is located and is highly targeted. Different edge devices may vary due to factors such as the process environment and equipment status. Using local historical data for training can enable the local model to better adapt to the actual conditions of the local specific device, thereby improving the training effect and model generalization ability. The edge device performs optimization training based on the locally stored historical execution results and historical execution result evaluation data and training content, which can combine global training content with local actual conditions. The training content may contain some general knowledge and experience, while the local historical data provides personalized information for the local edge device. The combination of the two can enable the local model to have stronger learning capabilities and be highly compatible with the local edge device, improving the accuracy and reliability of the local model.

[0020] In some optional embodiments, the training content also includes one or more of the first historical execution data, the evaluation data of the first historical execution result, the execution data of the initial process plan, the evaluation data of the execution result of the initial process plan, and the process parameter adjustment data stored locally on the edge device. Each other edge device performs optimization training based on the training content, including: other edge devices perform optimization training based on the first historical execution data, the evaluation data of the first historical execution result, the execution data of the initial process plan, the evaluation data of the execution result of the initial process plan, and the process parameter adjustment data stored locally on the edge device. Each other edge device performs optimization training based on the locally stored data, and at the same time combines global information such as the first historical execution data to achieve the fusion of local data and global information. It can fully utilize the actual production data of other edge devices themselves, and draw on the production experience and knowledge of edge devices, so that the training is both targeted and can be optimized from a more macro perspective. It improves the generalization ability of the model in each edge device and the matching ability of the edge device where it is located.

[0021] In some optional embodiments, the method further includes: other edge devices obtain one or more of the first historical execution data uploaded by the edge device, the first historical execution result evaluation data, the execution data of the initial process plan, the execution result evaluation data of the initial process plan, and the process parameter adjustment data through the cloud. Data sharing between different edge devices is achieved through the cloud, breaking the limitations of the data of a single edge device and forming an interconnected data network. Each edge device can obtain the data of other edge devices, so as to better understand the historical execution status and process changes of the entire production system. In addition, "privacy isolation" can also be achieved through the cloud. By pre-processing the data before uploading it to the cloud, other edge devices can ensure normal model training based on the pre-processed data after obtaining the pre-processed data, achieving the expected training effect while ensuring that other edge devices cannot obtain the original data of the edge device.

[0022] In some optional embodiments, the method further includes: each other edge device sends the optimization parameters obtained from the optimization training to the cloud, so that the optimization parameters are sent to the edge device via the cloud; the edge device updates the second chemical process-specific training model based on the optimization parameters. By uniformly scheduling and updating the optimization parameters on the cloud, it is possible to ensure that the second chemical process-specific training models of each edge device converge in the same direction. This avoids model divergence or non-convergence caused by independent training of edge devices, and improves the stability and accuracy of the model. The edge device does not need to perform a large amount of optimization training independently to find suitable optimization parameters, but can directly obtain the optimized optimization parameters from the cloud. This greatly reduces the amount of computation on the edge device, shortens the training time, and improves the training efficiency. And by updating the optimization parameters on the cloud, the expansion of the system becomes easier. When a new edge device is added, it only needs to be connected to the cloud to obtain the latest optimization parameters so that it can be quickly put into the production process without the need to configure and train each device separately.

[0023] In some optional embodiments, the method further includes determining that the process complexity meets a preset condition when the number of impurities that cause loss of the target compound is greater than or equal to a preset number; or when the process execution time in the process plan exceeds a preset time. For high-complexity processes, multiple edge devices can collaborate to share the high computing power pressure.

[0024] In some optional embodiments, transmitting the optimized process plan and / or optimized model parameters to the cloud includes encrypting the optimized process plan and / or optimized model parameters according to a preset encryption policy, and transmitting the encrypted optimized model parameters to the cloud. By encrypting the optimized model parameters, the security and privacy of the optimized model parameters uploaded to the cloud are ensured. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] One or more embodiments are exemplarily described by the figures in the corresponding drawings, and these exemplified descriptions do not constitute limitations on the embodiments.

[0026] Figure 1 This is an implementation environment architecture diagram of a chemical process collaborative optimization based on a distributed training model deployment system provided by an embodiment of the present application;

[0027] Figure 2 It is a flow chart of a chemical process collaborative optimization method based on a distributed training model deployment system;

[0028] Figure 3 This is a schematic diagram of the structure of an edge device network in an embodiment of the present application;

[0029] Figure 4 This is a flowchart of the operation content of each processing stage of the edge device provided by an embodiment of the present application;

[0030] Figure 5 This is a schematic diagram of a structure of cloud and edge device collaboration provided by an embodiment of the present application;

[0031] Figure 6 This is a flow chart of a chemical process collaborative optimization method based on a distributed training model deployment system provided by an embodiment of the present application;

[0032] Figure 7 This is a schematic structural diagram of a chemical process collaborative optimization device based on a distributed training model deployment system provided by another embodiment of the present application;

[0033] Figure 8 It is a structural diagram of an electronic device provided by another embodiment of the present application. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, each embodiment of the present application will be described in detail below with reference to the accompanying drawings. However, it will be understood by those skilled in the art that in each embodiment of the present application, many technical details are proposed to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical solutions claimed in the present application can be implemented. The division of the following embodiments is for convenience of description and should not constitute any limitation on the specific implementation of the present application. The various embodiments can be combined and referenced with each other under the premise of no contradiction.

[0035] It should be noted that the acquisition or use of data in the embodiments of this application requires the user's consent. The relevant data can only be obtained after the user's authorization and permission, and the acquisition or use of the data complies with the provisions of relevant laws and regulations.

[0036] To facilitate understanding of the embodiments of the present application, the following describes a method for collaborative optimization of chemical processes based on a distributed training model deployment system.

[0037] With the advancement of technology, chemical process equipment has achieved a high degree of automation. For example, existing liquid chromatography separation technology and its instrumentation are highly automated. However, users still need to possess basic knowledge of chromatography theory and conduct a series of experiments to determine essential information for the separation and purification of target compounds, such as solvent systems and Rf values. After separation and purification, the method for the target compound is typically stored on a single liquid chromatograph for future review by that instrument's user, but it cannot be shared with users in other organizations. When different users isolate and purify the same or similar target compounds, they generally need to re-run experiments to determine the essential information due to the isolation and inability to share this information. This results in duplication of effort and waste of resources.

[0038] Among the related technologies, for example, the trading and sharing platforms and chemical instruments or devices provided by China CN103055541B and CN104504152B can download, execute and verify chemical information related to existing target compounds or target compound systems, and can guide and motivate users to develop and use more efficient chemical processes, thereby improving R&D and production efficiency.

[0039] However, when users optimize chemical processes, they still need to possess relevant chemical knowledge and conduct extensive experiments to determine the optimal information for the separation and purification of target compounds, which severely limits their ability to achieve more efficient chemical processes. Therefore, a solution to the above problems is urgently needed.

[0040] It should be understood that in order to clearly understand the content of this solution, the relationship between at least one edge device, the cloud, the general chemical training model, the first chemical process special training model, and the second chemical process special training model required for the actual application of this solution is explained here. For details, please refer to Figure 1 , Figure 1 An implementation environment architecture diagram of a chemical process collaborative optimization based on a distributed training model deployment system provided in an embodiment of the present application:

[0041] This solution requires at least one edge device 11, cloud 12, general chemical training model 13, first chemical process-specific training model 14, and second chemical process-specific training model 15 in its application;

[0042] Among them, the distributed training model deployment system includes a cloud 12 and at least one edge device 11, the cloud is provided with a general chemical training model 13 and / or a first chemical process-specific training model 14, the edge device 11 is provided with a second chemical process-specific training model 15, and an optional communication connection is provided between the cloud 12 and the at least one edge device 11. In some embodiments, due to the limited computing power of the edge device 11, the edge device 11 can be provided with at least part of the second chemical process-specific training model 15, that is, the second chemical process-specific training model 15 can be distributed on multiple edge devices 11 to reduce the computing power burden of the edge device 11 on the basis of meeting the deployment requirements. This application does not make specific restrictions on this, and the selection can be made based on the size of the second chemical process-specific training model 15 and the computing power of the edge device 11.

[0043] Specifically, the edge device 11 is used to respond to the target compound generation instruction and generate an initial process plan using the second chemical process special training model; obtain the execution result of the initial process plan executed by the edge device, and generate execution result evaluation data based on the execution result; optimize the second chemical process special training model based on the execution data of the initial process plan and the execution result evaluation data to obtain an optimized process plan and its corresponding optimized model parameters that meet the preset evaluation data; when receiving the feedback instruction, the optimized process plan and / or optimized model parameters are sent to the cloud to update the general chemical training model and / or the first chemical process special training model.

[0044] The cloud 12 is used to receive the optimized process plan and / or optimized model parameters sent by the edge device 11, and update the general chemical training model 13 and / or the first chemical process special training model 14 based on the optimized model parameters.

[0045] In order to solve the technical problems in the above-mentioned prior art methods for optimizing chemical processes, such as the inability to adapt to different chemical process equipment, the inability to achieve privacy control, the inability to meet the real-time requirements of offline chemical process equipment, and the insufficient accuracy of the optimized chemical process, the present invention proposes a chemical process collaborative optimization method based on a distributed training model deployment system. The implementation details of the chemical process collaborative optimization method based on a distributed training model deployment system of this embodiment are specifically described below. The following content is only the implementation details provided for easy understanding and is not necessary for the implementation of this solution.

[0046] Example 1:

[0047] The chemical process collaborative optimization method based on the distributed training model deployment system of this embodiment can be applied to electronic devices with communication, computing and data storage capabilities, wherein the distributed training model deployment system includes a cloud and at least one edge device, the cloud is provided with a general chemical training model and / or a first chemical process-specific training model, the edge device is used for chemical process execution, and is provided with at least part of the second chemical process-specific training model, an optional communication connection between the cloud and at least one edge device, the method is applied to the edge device, and its specific process can be as follows Figure 2 Shown, including:

[0048] Step 110 : In response to the target compound generation instruction, an initial process plan is generated using the second chemical process-specific training model.

[0049] Specifically, at least one edge device is at least one instrument that can execute a chemical process. The distributed training model deployment system may include multiple edge devices, that is, multiple instruments that can execute a chemical process. The multiple edge devices can independently deploy a second chemical process-specific training model, so that each edge device can independently use the second chemical process-specific training model to generate an initial process plan. The second chemical process-specific model can also be distributed and deployed on multiple edge devices. Through communication between multiple edge devices, the generated results on multiple edge devices are combined to realize the generation of an initial process plan using the distributed second chemical process-specific model. In order to facilitate the management of multiple instruments that can execute chemical processes, multiple instruments that can execute chemical processes can be networked, and each instrument that can execute chemical processes is a distributed node. The instruments that can execute chemical processes can be liquid chromatographs (medium and low pressure preparative liquid chromatographs, high performance liquid chromatographs, etc.), centrifuges, synthetic reaction instruments, etc.

[0050] For example 1, see Figure 3 , if the distributed training model deployment system includes 3 liquid chromatographs, they are recorded as node A, node B, and node C respectively. Networking is based on node A, node B, and node C. Among them, node A, node B, and node C can have a built-in chromatographic column database, and the built-in chromatographic column database can be associated with the user chromatographic column database. In addition, when the available storage space of node A, node B, and node C is larger than the model volume, for example, 50MB (the model volume is less than 50MB), the lightweight separation and purification process model can be deployed in node A, node B, and node C respectively; when the available storage space of node A, node B, and node C is smaller than the model volume, for example, the available storage space of node B and node C is less than the model volume of 50MB, the second chemical process special model is distributedly deployed in node A, node B, and node C according to the distributed deployment strategy.

[0051] Specifically, the target compound is a compound that needs to be synthesized, separated and purified. The compound can be a compound in multiple fields. That is, this solution can be applied to multiple fields that need to synthesize, separate and purify compounds, and is not limited to a single field. The multiple fields can be the pharmaceutical field, the chemical field, and the material science field. For example, the target compound can be a compound that needs to be synthesized, separated and purified in the pharmaceutical field, a compound that needs to be synthesized, separated and purified in the chemical field, and a compound that needs to be synthesized, separated and purified in the material science field.

[0052] Specifically, the target compound generation instruction refers to an instruction for processing the target compound, including instructions for synthesizing the target compound and separating and purifying the target compound.

[0053] In some examples, the target compound generation instruction also includes characteristic information of the target compound, such as the molecular weight, polarity, solubility, melting point, boiling point, electrochemical point, hydrophobicity and other characteristic information of the target compound, or other characteristic information. This is only an example and is not specifically limited.

[0054] Continuing with the aforementioned example 1: if the target compound needs to be separated and purified, the characteristic information includes at least one of the CAS number of the target compound, the CAS number of the key raw material compound of the previous operation route in each execution operation, the conversion rate of the previous synthesis process in each execution operation, the sample amount, TLC plate information, etc.; wherein the conversion rate is also called the yield, which may affect the content of impurities and target compounds.

[0055] Specifically, the second chemical process training model refers to the model deployed on the current edge device, known as the local model. This local model is used to generate an initial process plan for synthesizing, separating, and purifying the target compound based on its characteristic information. The second chemical process training model is lightweight, with a size of less than 50MB.

[0056] Specifically, the initial process plan is used to guide the operation process of the edge device so that the edge device generates the target compound. Exemplarily, for the synthesis of the target compound, the initial process plan may include the chemical reaction steps required for the synthesis of the target compound (including the type of each reaction step), reaction conditions (for example, temperature, moderation, reaction time, pressure, solvent, catalyst and dosage, etc.), raw material sources and raw material specifications (for example, raw material purity, water content, etc.) at least one; for the separation and purification of the target compound, the initial process plan may include the crude product pretreatment process, purification method (for example, recrystallization, column chromatography, thin layer chromatography, distillation, rectification, etc.) and purification conditions (for example, temperature, pressure, flow rate, eluent composition, etc.), chromatographic column type, gradient program, gradient elution program at least one.

[0057] Continuing with the above-mentioned example one: the initial process plan generated by the local model is a separation and purification process plan, which may include a gradient program (for example, acetonitrile / water as the mobile phase, the flow rate is set to 10 mL / min, and the injection volume is 50 mg), a chromatographic column, the number of times the sample needs to be repeated (it needs to be judged based on the condition of the chromatographic column after each step of the separation operation, for example, after the first operation is completed and cleaned, the back pressure and baseline reach a certain standard, and the second operation of injection and separation can be repeated), the sample amount each time, and at least one of the detection wavelength.

[0058] In some examples, a user or robot can send instructions for generating a target compound to an edge device. This demonstrates that this solution provides both a human-machine interface and a non-human-machine interface for robot-device interaction, offering multiple possibilities for interaction to meet diverse needs.

[0059] Step 120 , obtaining the execution result of the initial process plan executed by the edge device, and generating execution result evaluation data based on the execution result.

[0060] Specifically, edge devices refer to instruments that can perform chemical processes, such as liquid chromatographs (medium and low pressure preparative liquid chromatographs, high performance liquid chromatography, etc.), centrifuges, synthesis reaction instruments, etc.

[0061] Specifically, the execution result refers to the product result obtained by the edge device when executing the initial process plan. The product result may include information such as product purity, separation efficiency, separation time, synthesis time, solvent consumption, and energy consumption.

[0062] Specifically, execution result evaluation data refers to an evaluation of the execution results to indicate the differences or gaps between the execution results and the target compound. For example, data such as the accuracy of the execution results and the purity of the compound can be used to evaluate the quality of the execution results. This data can be used to indirectly assess the effectiveness of the initial process plan in obtaining the target compound and can also flag abnormal information in the execution results. For example, if the product purity in the execution results is lower than a preset purity threshold, the product purity data can be marked as abnormal data.

[0063] In some examples, there may be multiple methods for obtaining execution result evaluation data. Specifically, in an optional embodiment 1, generating execution result evaluation data based on the execution result includes: obtaining a preset standard execution result, performing a comparative analysis based on the preset standard execution result and the execution result, and generating execution result evaluation data. In an optional embodiment 2, generating execution result evaluation data based on the execution result includes: displaying the execution result in a display module of the edge device; and obtaining execution result evaluation data given by relevant technical personnel for the execution result. It can be seen from this that by generating multiple methods of execution result evaluation data, the method of generating execution result evaluation data can be determined according to needs in different application scenarios to meet the needs of different scenarios. For example, when the edge device cannot perform automated evaluation, the execution result evaluation data can be obtained by manually evaluating the execution results in a reserved manner to ensure the sequential execution of subsequent operations and the sequential execution of the coordinated optimization of the chemical process. For another example, when it is necessary to obtain execution result evaluation data efficiently, the automatic generation method of the edge device can be used to quickly and efficiently generate execution result evaluation data.

[0064] In some examples, obtaining the execution result of the initial process plan executed by the edge device includes: sending the initial process plan to the cloud so that the cloud optimizes the initial process plan based on the general chemical training model and / or the first chemical process special training model to obtain an optimized initial process plan; receiving the optimized initial process plan sent by the cloud, and obtaining the execution result of the edge device executing the optimized initial process plan. It can be seen from this that sending the initial process plan to the cloud and optimizing the initial process plan using the general chemical training model and the first chemical process special training model in the cloud can fully utilize the ability of the general chemical training model and the first chemical process special training model in the cloud to have a wider range of knowledge to optimize the initial process plan, fully utilizing the knowledge open in the cloud, making the optimized initial process plan more accurate, and achieving the technical effect of collaborative knowledge sharing between the cloud and the edge device.

[0065] Specifically, the general chemistry training model is a large general artificial intelligence model. It has more extensive knowledge than the second chemical process training model, which can be understood as having more "world knowledge" than the second chemical process training model.

[0066] Specifically, the difference between the first and second chemical process-specific training models is that the first chemical process-specific training model is deployed in the cloud and, compared to the second chemical process-specific training model, is less compatible with the edge devices on which it is deployed. Compared to the second chemical process-specific training model, the first chemical process-specific training model lacks the ability to generate personalized initial process plans that are highly compatible with edge devices. Furthermore, the second chemical process-specific training model is less complex than the first chemical process-specific training model.

[0067] In some examples, the first chemical process-specific training model can be subjected to knowledge distillation processing using knowledge distillation technology to obtain a second chemical process-specific training model, and the second chemical process-specific training model can be migrated to the edge device, making the second chemical process-specific training model a local model of the edge device.

[0068] In some examples, the first chemical process-specific training model can be obtained from a pre-trained model library, wherein the pre-trained model library contains chemical synthesis process models, chemical separation and purification process models, etc. Among them, the chemical synthesis process model can be used to generate a process scheme for synthesizing a target compound; the chemical separation and purification process model can be used to generate a process scheme for separating and purifying a target compound. It can be seen that a variety of models can be configured through the pre-trained model library to meet a variety of needs. In addition, the models in the pre-trained model library can be optimized offline, and the optimized models can be updated to the pre-trained model library to achieve the isolation of the models used in the cloud and the models in the pre-trained model library. Updating the models in the pre-trained model library will not interfere with the use of the cloud model, thereby ensuring the stability of the cloud model.

[0069] In some examples, before the cloud optimizes the initial process plan based on the general chemical training model and / or the first chemical process special training model, it also includes: the cloud obtains the optimization model parameters sent by other edge devices when receiving the feedback instruction; the cloud updates the general chemical training model and / or the first chemical process special training model based on the optimization model parameters; and when receiving the initial process plan, the initial process plan is optimized based on the updated general chemical training model and / or the first chemical process special training model. It can be seen that the feedback instruction can be used to selectively optimize and update the cloud model based on the optimization model parameters of other edge devices to improve the learning ability of the cloud model and the real-time performance of the cloud model, so that the initial process plan of the edge device is optimized based on the updated cloud model, thereby improving the real-time performance and accuracy of the initial process plan.

[0070] Specifically, optimizing model parameters refers to adjusting parameters of the general chemistry training model and / or the first chemical process specialized training model so as to make the general chemistry training model and / or the first chemical process specialized training model achieve a better result.

[0071] In some examples, after the general chemical training model and / or the first chemical process special training model are updated in the cloud based on the optimization model parameters, the method further includes: the updated general chemical training model and the first chemical process special training model can be identified respectively by a unique version identifier, so as to manage the general chemical training model and the first chemical process special training model through version management. It can be seen that by managing the general chemical training model and the first chemical process special training model based on version management, any general chemical training model and the first chemical process special training model in the historical version can be obtained to meet the different requirements for the general chemical training model and the first chemical process special training model. This avoids the problem that the general chemical training model and the first chemical process special training model of the historical version are lost, resulting in the inability to optimize the process plan of the same historical period based on the historical general chemical training model and the first chemical process special training model, thereby resulting in poor optimization effect.

[0072] In some examples, obtaining the execution result of the initial process plan executed by the edge device includes: receiving process parameter adjustment data corresponding to the process parameter adjustment instruction, adjusting and optimizing the initial process plan based on the process parameter adjustment data to obtain the adjusted and optimized initial process plan, and obtaining the execution result of the edge device executing the adjusted and optimized initial process plan. It can be seen from this that when the initial process plan obtained by the essential oil second chemical process special training model does not meet the user's needs, or the user has a certain amount of relevant knowledge, the user can input the process parameter adjustment instruction to the edge device through the interactive interface according to their own needs or technical requirements, so as to provide the process parameter adjustment data for the initial process plan, thereby obtaining the adjusted and optimized initial process plan for the edge device to execute and obtain the execution result. It should be understood that the process adjustment data entered by the user can inject the user's technical preferences into the initial execution of the current target compound generation instruction by the edge device, provide more user-specific feature information for the subsequent training of the second chemical process special training model, provide technical support from the user for the final training to obtain the optimized process plan and its corresponding optimized model parameters that meet the preset evaluation data, and form core competitiveness with user feature preferences.

[0073] It should be understood that optimizing the initial process plan using the general chemical training model and / or the first chemical process-specific training model and adjusting and optimizing the initial process plan using the process parameter adjustment data corresponding to the process parameter adjustment instruction can be used separately or in combination. For example, after obtaining the initial process plan, the general chemical training model and / or the first chemical process-specific training model are first used to optimize the initial process plan to obtain an optimized initial process plan, and then the process parameter adjustment data corresponding to the process parameter adjustment instruction are used to adjust and optimize the optimized initial process plan to obtain an adjusted and optimized initial process plan. Alternatively, the process parameter adjustment data corresponding to the process parameter adjustment instruction are first used to adjust and optimize the initial process plan to obtain an adjusted and optimized initial process plan, and then the general chemical training model and / or the first chemical process-specific training model are used to optimize the adjusted and optimized initial process plan, or the two are used alternately. This application does not make specific limitations.

[0074] Step 130 , optimizing and training the second chemical process-specific training model based on the execution data of the initial process plan and the evaluation data of the execution result, to obtain an optimized process plan and its corresponding optimized model parameters that meet the preset evaluation data.

[0075] Specifically, execution data refers to information related to the execution process when the edge device executes the initial process plan, such as the execution time of each step in the execution process, the specific execution operation, execution result data, and execution environment data.

[0076] Specifically, the preset evaluation data refers to the standard evaluation data of the execution results corresponding to the process scheme, which is used to represent the threshold of the gap between the execution results and the target compound, or the minimum difference threshold. When this threshold is met, it can be determined whether the execution results meet the preset requirements, that is, it can be determined that the process scheme corresponding to the execution result is a high-quality scheme.

[0077] Specifically, an optimized process plan refers to a plan that better produces the target compound, or can also be understood as a plan that produces a compound closer to the target compound. The optimized process plan may include adjusting the gradient slope, optimizing the injection volume, and other information. Specifically, optimized model parameters refer to the model parameters corresponding to the second chemical process-specific training model that can generate the optimized process plan, i.e., the model parameters corresponding to the second chemical process-specific training model when the process plan optimization conditions are met.

[0078] Continuing with Example 1, if the product purity (the purity of the execution result) of the product obtained by executing the initial process plan is 82%, then by optimizing the initial process plan to obtain an optimized process plan, the product purity of the product obtained by executing the optimized process plan can be increased to 89%. Therefore, it can be seen that after model optimization training, a higher-quality plan can be obtained, which improves the accuracy of the model in obtaining the process plan corresponding to the target compound.

[0079] In some examples, optimizing and training a second chemical process-specific training model based on the execution data and execution result evaluation data of the initial process plan includes: obtaining first historical execution data and first historical execution result evaluation data stored locally on an edge device, as well as the execution data and execution result evaluation data of the initial process plan; and / or receiving process parameter adjustment data corresponding to a process parameter adjustment instruction; and offline optimizing and training the second chemical process-specific training model based on the first historical execution data, the first historical execution result evaluation data, the execution data of the initial process plan, the execution result evaluation data of the initial process plan, and / or the process parameter adjustment data. Optimizing the training model using the first historical execution data and first historical execution result evaluation data stored locally on the edge device, as well as the execution data and execution result evaluation data of the initial process plan, ensures that model training is based on actual, historical data. This data can more accurately reflect the actual production process, thereby improving the model's predictive accuracy and reliability. Receiving the process parameter adjustment data corresponding to the process parameter adjustment instruction to optimize the training model allows the model to adjust according to real-time production needs and conditions. This dynamic adjustment capability makes the model more flexible and adaptable to changes in the production process, thereby improving the accuracy of the process plan. Offline model optimization and training allows for model improvements without disrupting normal production. This reduces the risks associated with online model adjustments while ensuring the model's stability and reliability in real-world applications. Continuous model optimization and training allows the model to better cope with various uncertainties and changes, enhancing its robustness and adaptability. Local models can be fine-tuned by combining local historical data with real-time data to create domain-specific models.

[0080] Specifically, the first historical execution data refers to execution data of a historical process plan corresponding to at least one historical target compound instruction executed by the edge device in a past time period. Specifically, the first historical execution result evaluation data refers to at least one historical execution result evaluation data obtained by evaluating at least one execution result of executing at least one historical process plan in a past time period by the edge device.

[0081] Specifically, the process parameter adjustment instruction refers to an instruction for adjusting the parameters for generating the target compound, and the process parameter adjustment data is used to adjust the parameters for generating the target compound in the initial process plan, wherein the process parameter adjustment data may include the name of the parameter to be adjusted and the amplitude value to be adjusted.

[0082] In some examples, the process parameter adjustment instruction can be sent to the edge device by the relevant technical personnel, or can be sent to the edge device by the cloud. Specifically, optionally in embodiment one, if the edge device displays the initial process plan to the relevant technical personnel, then before receiving the process parameter adjustment data corresponding to the process parameter adjustment instruction, the method also includes: the relevant technical personnel sends the process parameter adjustment data to the edge device through the process parameter adjustment instruction. Optionally in embodiment two, if the edge device sends the initial process plan to the cloud, then before receiving the process parameter adjustment data corresponding to the process parameter adjustment instruction, the method also includes: the cloud compares the initial process plan with the standard process plan to obtain the process parameter adjustment data, and sends the process parameter adjustment data to the edge device through the process parameter adjustment instruction. For example, if the initial process plan includes solvent gradient information, and the cloud or the relevant technical personnel determine that the solvent gradient information is difficult to achieve the target effect, then by dynamically adjusting the solvent gradient, process parameter adjustment data for adjusting the solvent gradient information is generated. According to experimental data, by adjusting the solvent gradient to appropriate gradient data, the consumption of resources or energy can be reduced by 20%-30%. As can be seen, by reserving a "window" for relevant technicians to adjust the parameters in the initial process plan, and also adjusting the parameters in the initial process plan based on the standard parameters in the cloud-based standard process plan, the corresponding method for adjusting process parameters can be selected in different application scenarios, which can meet the needs of different scenarios and reduce resource and energy consumption.

[0083] In some examples, the second chemical process special training model is optimized and trained based on the execution data and execution result evaluation data of the initial process plan, including: determining the process complexity corresponding to the target compound generation instruction based on the initial process plan and / or the execution result evaluation data; generating a distributed training strategy corresponding to the optimization training when the process complexity meets the preset conditions; the distributed training strategy includes at least one other edge device participating in the training and the training content corresponding to each other edge device, and the training content includes at least the exit condition corresponding to the optimization training; the training content is sent to each other edge device through the cloud, so that each other edge device performs optimization training based on the training content. It can be seen that training other edge devices through the training content sent by the edge device can enable the other edge devices to have the characteristics of the edge device, match the edge device, and be used as part of the edge device.

[0084] In some examples, the method further includes: when the number of impurities that cause loss of the target compound is greater than or equal to a preset number; or when the time taken to implement the process in the process plan exceeds a preset time; and determining that the process complexity meets a preset condition.

[0085] As can be seen, generating a distributed training strategy based on process complexity allows for customized training plans tailored to varying process complexities. This avoids a one-size-fits-all training strategy, allowing each training task to be optimized based on its unique characteristics, improving training efficiency and quality. A distributed training strategy includes at least one other edge device participating in the training, as well as the training content for each other edge device. This facilitates the rational and dynamic allocation of computing resources. By distributing training tasks across multiple edge devices based on dynamic computing power allocation, the computing power of each edge device can be fully utilized, accelerating the training process and improving training efficiency. The training content includes at least exit conditions for optimized training, providing each edge device with clear training objectives and stopping criteria. Edge devices can autonomously train and make decisions based on the exit conditions, ensuring controllable and stable training. Furthermore, clear exit conditions avoid unnecessary training, saving computing resources and time. Sending training content to each other edge device via the cloud enables centralized management and distribution of training data. This ensures the consistency and accuracy of training content and facilitates unified monitoring and management of the training process. Edge devices can collaborate on training through the cloud, sharing information and resources, further improving training effectiveness.

[0086] Specifically, process complexity can be used to describe the computational complexity and process difficulty of generating the target compound.

[0087] Specifically, other edge devices can serve as external computing power enhancement devices for edge devices, so that when edge devices are overloaded, other edge devices can dynamically share part of the operating pressure of edge devices.

[0088] Furthermore, if only the computing power of edge devices needs to be enhanced, other edge devices can be equipped with GPU accelerators to enhance their computing power and their ability to handle computational workloads. Increasing the computing power of edge devices allows them to handle more computing power, enhancing their learning capabilities.

[0089] Continuing with the above example 1: When node A is used to generate the target compound, the number of impurities that lose the target compound is greater than or equal to a preset number; or the process implementation time in the process plan exceeds the preset time. The computing power of node A can be increased by configuring a GPU computing power card for node A. Specifically, a chromatographic column database is built into the node A, j device. The GPU computing power card can be associated with the user's real-time inventory data of chromatographic columns;

[0090] It should be understood that if the process complexity meets the preset conditions, that is, the computational complexity and process difficulty of generating the target compound indicated by the process complexity are high, in order to reduce the load on the edge device, part of the generation content of the target compound generated by the edge device can be assigned to other edge devices. In order to enable the other edge devices to better complete the part of the generation content assigned to them by the edge device, the training content corresponding to the generation of the part of the generation content can be sent to the corresponding other edge devices through the edge device, so that the other edge devices, based on their ability to generate the part of the generation content, can assist the edge device in generating the target compound. Therefore, in some examples, after sending the training content to each other edge device via the cloud so that each other edge device performs optimization training based on the training content, the method also includes: when each other edge device performs optimization training based on the training content, if the exit condition corresponding to the optimization training is met, each other edge device stops the optimization training and uses the latest each other edge device as an available other edge device. When there is an available other edge device, the edge device can send the execution content corresponding to the training content corresponding to the available other edge device to the available other edge device via the cloud so that it executes the corresponding execution content. When other available edge devices assist the edge device in generating the target compound, the efficiency of generating the target compound can be improved while meeting the accuracy of generating the target compound.

[0091] It should be understood that to ensure the security and privacy of edge device data, in some examples, before sending training content to each other edge device via the cloud so that each other edge device performs optimized training based on the training content, the method further includes: encrypting the training content using a preset encryption method to obtain encrypted training content, and using the encrypted training content as the training content, wherein the preset encryption method is at least one of homomorphic encryption, federated learning, multi-party computation, and differential privacy. The preset encryption method is characterized by being able to encrypt the original data of the training content, while ensuring that the model trained based on the encrypted training content has the same performance and accuracy as the model trained based on the pre-encrypted training content. Thus, achieving "privacy isolation" through the cloud can pre-process data before uploading to the cloud, so that other edge devices, after obtaining the pre-processed data, can ensure normal model training based on the pre-processed data, achieving the expected training effect, while also ensuring that other edge devices cannot access the edge device's original data. The preset encryption method ensures the privacy of the original data while also ensuring the effectiveness of the trained model.

[0092] In some examples, each other edge device performs optimization training based on the training content, including: each other edge device obtains locally stored historical execution results and historical execution result evaluation data; and performs optimization training based on the locally stored historical execution results and historical execution result evaluation data and training content. Different edge devices may differ due to factors such as the process environment and equipment status. Using local historical data for training can make the local model better adapt to the actual conditions of local specific equipment, thereby improving the training effect and the generalization ability of the model. The edge device performs optimization training based on the locally stored historical execution results and historical execution result evaluation data and training content, which can combine the global training content with the local actual conditions. The training content may contain some general knowledge and experience, while the local historical data provides personalized information for the local edge device. The combination of the two can make the local model highly match the local edge device on the basis of having stronger learning ability, thereby improving the accuracy and reliability of the local model.

[0093] Specifically, the locally stored historical execution results refer to the historical execution results stored locally by each other edge device, wherein the historical execution results refer to at least one execution result of at least one historical process solution executed by each other edge device in the past time period. Specifically, the historical execution result evaluation data refers to the historical execution result evaluation data stored locally by each other edge device, wherein the historical execution result evaluation data refers to at least one historical execution result evaluation data obtained by evaluating at least one execution result of executing at least one historical process solution by each other edge device in the past time period.

[0094] In some examples, the training content also includes one or more of the first historical execution data, the evaluation data of the first historical execution result, the execution data of the initial process plan, the evaluation data of the execution result of the initial process plan, and the process parameter adjustment data stored locally on the edge device. Each other edge device performs optimization training based on the training content, including: other edge devices perform optimization training based on the first historical execution data, the evaluation data of the first historical execution result, the execution data of the initial process plan, the evaluation data of the execution result of the initial process plan, and the process parameter adjustment data stored locally on the edge device. Each other edge device performs optimization training based on the locally stored data, and at the same time combines the global information in the edge device (such as the first historical execution data, the evaluation data of the first historical execution result, the execution data of the initial process plan, the evaluation data of the execution result of the initial process plan, and the process parameter adjustment data stored locally on the edge device), thereby realizing the fusion of local data and global information. It can fully utilize the actual production data of other edge devices themselves, and draw on the production experience and knowledge of edge devices, so that the training is both targeted and can be optimized from a more macro perspective. It improves the generalization ability of the model in each edge device and the matching ability of the edge device where it is located.

[0095] In some examples, the method further includes: other edge devices obtaining, via the cloud, one or more of the following: first historical execution data uploaded by the edge device, evaluation data of the first historical execution results, execution data of the initial process plan, evaluation data of the execution results of the initial process plan, and process parameter adjustment data. Sharing data between different edge devices via the cloud overcomes the limitations of individual edge device data and forms an interconnected data network. Each edge device can obtain data from other edge devices, thereby better understanding the historical execution status and process changes of the entire production system.

[0096] In some examples, the method further includes: each other edge device sends the optimization parameters obtained from the optimization training to the cloud, so that the optimization parameters are sent to the edge device via the cloud; and the edge device updates the second chemical process-specific training model based on the optimization parameters. By uniformly scheduling and updating the optimization parameters on the cloud, the second chemical process-specific training models of each edge device can be ensured to converge in the same direction. This avoids model divergence or non-convergence caused by independent training of edge devices, thereby improving the stability and accuracy of the model. Instead of performing extensive optimization training to find appropriate optimization parameters, edge devices can directly obtain already optimized optimization parameters from the cloud. This significantly reduces the computational workload of edge devices, shortens training time, and improves training efficiency. Updating optimization parameters through the cloud also makes system expansion easier. When a new edge device is added, it only needs to be connected to the cloud to obtain the latest optimization parameters for rapid integration into the production process, eliminating the need for individual configuration and training of each device.

[0097] Step 140, when receiving the feedback instruction, based on the iterative selection corresponding to the feedback instruction and the optimized process scheme and / or optimized model parameters, determine the target process scheme and / or target model parameters to be uploaded, and send the target process scheme and / or target model parameters to the cloud to update the general chemical training model and / or the first chemical process special training model, wherein the optimization degree of the target process scheme and / or target model parameters is less than or equal to the optimized process scheme and / or optimized model parameters.

[0098] Specifically, the feedback instruction refers to an instruction to optimize the process plan and / or model parameters corresponding to the optimization training results (such as process accuracy, compound purity improvement, etc.) of the local second chemical process special training model on the cloud.

[0099] In some examples, based on the iterative selection corresponding to the feedback instruction and the optimized process plan and / or optimized model parameters, the target process plan and / or target model parameters to be uploaded are determined, and the target process plan and / or target model parameters are sent to the cloud.

[0100] That is to say, when the user chooses to use the data improved by local optimization to feed back the general chemical training model and / or the first chemical special training model deployed in the cloud, the feedback instruction may include an iterative selection specifically used to select the feedback data. Among them, the iterative selection may include the selection of iteration accuracy. For example, the accuracy of the optimized process after training can reach 90%, and the iteration accuracy selected by the user is 60%. Then, the model parameters with an accuracy of 60% can be obtained from the historical version of the second chemical process special training model and sent to the cloud. The iterative selection may also include an iteration step (number of iterations). For example, the iteration step corresponding to the optimized model after training is 90, and the iteration step selected by the user is 60. Then, the model parameters with an iteration step of 60 can be obtained from the historical version of the second chemical process special training model and sent to the cloud. Thus, local users can send the optimized model parameters or process data to the cloud through selective feedback instructions to benefit more users, and can distinguish their own achievements from shared achievements, thereby maintaining the core competitiveness of the local user end.

[0101] In some examples, sending the target process solution and / or target model parameters to the cloud includes encrypting the target process solution and / or target model parameters according to a preset encryption policy and sending the encrypted target model parameters to the cloud. Encrypting the target model parameters ensures the security and privacy of the target model parameters uploaded to the cloud. Furthermore, the target process solution for generating the target compound can be directly accessed through the cloud, thereby improving the utilization rate of the target process solution, reducing the resources required to generate the target process solution, and thus conserving resources.

[0102] In some examples, in order to promote technological development, edge devices that upload target process plans and / or target model parameters to the cloud can be rewarded through a corresponding platform points reward system on the cloud.

[0103] In some examples, when a user obtains a process plan for generating a target compound through the cloud, the target process plan corresponding to the target compound stored in the cloud can be displayed to the user, and the user can obtain the target process plan from the cloud by downloading.

[0104] In some cases, users can obtain encrypted target process plans and / or target model parameters from the cloud for a fee or free of charge. This shows that sending target model parameters to the cloud can improve the model capabilities of the first chemical process-specific training model deployed on the cloud, thereby promoting technological development.

[0105] In some cases, when no feedback instructions are received, the target process plan and / or target model parameters are stored locally on the edge device. This shows that the second chemical process-specific training model corresponding to the target model parameters is a private model and is only stored locally on the edge device. The target process plan is also a private plan and is only stored locally on the edge device. By selectively uploading the target model parameters and target process plan to the cloud, the technical uniqueness of the edge device itself is guaranteed, forming a certain technical barrier.

[0106] It should be understood that in order to make the second chemical process special training model as a private model accessible to other devices and to ensure the security of the private model. In some examples, the method also includes: identifying the second chemical process special training model stored locally on the edge device through a unique model identifier, and sending the unique model identifier to the corresponding authorized edge device through authorization, so that the authorized edge device with the unique model identifier can access the second chemical process special training model corresponding to the unique model identifier locally on the edge device or deploy the second chemical process special training model on the authorized edge device. It can be seen that the unique model identifier allows the authorized edge device to access the corresponding model on the edge device, prohibits the unauthorized edge device that does not have the unique model identifier from accessing the corresponding model on the edge device, and ensures the security of the private model. At the same time, it also provides an interface for other edge devices to access or utilize the private model to improve the utilization rate of the private model.

[0107] In some examples, the security of the private model can be improved by hardware encryption and chip storage of the private model, thereby preventing unauthorized devices from copying the private model.

[0108] In some examples, the methods of obtaining a unique model identifier include a paid method and a gift method.

[0109] In some examples, a temporary access key may be generated for an authorized edge device having a unique model identifier, so that the authorized edge device can access the private model corresponding to the unique model identifier through the temporary access key.

[0110] Continuing with the previous example, if the private model in node A is authorized to nodes B and C in a network with node A, node A can allow nodes B and C access by generating a temporary access key. As can be seen, the temporary access key has a certain time limit. While the temporary access key approach allows authorized devices to access the private model, it also limits the length of time that authorized devices can access the private model.

[0111] For example, in order to more clearly understand the various processing stages of the edge device in this solution, please refer to Figure 4: The edge device includes four processing stages, namely the process generation stage, the execution stage, the evaluation stage, and the model iteration stage. Among them, in the process generation stage, the characteristic information of the target compound is first obtained, and the characteristic information is analyzed and processed using the local model to generate an initial process plan for generating the target compound, wherein the initial process plan includes process parameters, such as chromatographic flow rate gradient and eluent ratio. In the execution stage, the initial process plan is executed to obtain the execution results. In the evaluation stage, key indicators such as purity, yield, energy consumption, etc. are collected from the execution results, and compared with the standard expectations based on the collected key indicators to determine the gap between the execution results and the target compound, so as to obtain the execution result evaluation data. In the model iteration stage, the execution result evaluation data is used to optimize the local model for training to obtain a better local model. It can be seen that using the local model deployed locally on the edge device to generate the initial process plan can meet the real-time requirements of process optimization when the edge device is offline. It can also make full use of the learning ability and efficiency of the local model to achieve accurate and efficient generation of the initial process plan, that is, using the locally deployed chemical process special model to autonomously generate the chemical process, without the need for users to carry out complex process plan design, thereby improving the efficiency of the edge device in executing the chemical process; then, using the execution capability of the edge device's own local model, the execution result of the initial process plan is obtained, effectively improving the degree of automation of the edge device's process generation and execution. During the application process, the user only needs to input the initial instruction and the edge device can complete the corresponding Process generation and execution; by evaluating and analyzing the execution results through local model, it can be determined whether the execution results corresponding to the initial process plan meet the preset requirements, so that the initial process plan can be evaluated and improved through the execution result evaluation data of the execution result, and the local model can be trained based on the execution data of the initial process plan and the execution result evaluation data to optimize the local model; through the closed-loop processing process of "generation-execution-evaluation-training", the real-time requirements of the edge device are met, the learning ability of the local model is improved, and the optimized local model and the edge device can be matched in real time. It is determined that the process plan optimized based on the local model is highly matched with the current edge device, and the accuracy of the optimized process plan is improved.

[0112] For example, to better understand the collaboration between cloud and edge devices in this solution, please refer to Figure 5 , local models on edge devices can be privatized by storing them in a private domain, or they can be shared with cloud devices to achieve knowledge sharing. Furthermore, cloud models can be used to assist local models in generating high-quality process solutions. This demonstrates that cloud-edge device collaboration enables flexible knowledge sharing and technology isolation.

[0113] In a specific embodiment, Figure 6As shown, the user can input the characteristic information of the target compound system that needs to be separated and purified through the interactive interface on the edge device. The edge device generates an initial process plan through the locally deployed second chemical process special training model, and determines whether the user chooses to access the cloud. If accessed, the initial process plan is optimized through the general chemical training model and / or the first chemical process special training model deployed in the cloud to obtain an optimized initial process plan, and the optimized initial process plan is executed to obtain an execution result. If not accessed, the initial process plan generated by the second chemical process special training model is directly executed to obtain an execution result.

[0114] The execution results of the initial process plan or the execution results of the optimized initial process plan are evaluated to generate execution result evaluation data to determine whether the process plan needs to be further optimized. If necessary, the second chemical process special training model is optimized based on the execution results and the execution result evaluation data. During the optimization process, the process parameter adjustment data input by the user can be received to correct and optimize the process to obtain an optimized process plan that meets the preset evaluation data and the optimized model parameters corresponding to the optimized process plan. If not, the initial process plan and its model parameters or the optimized initial process plan and its model parameters are directly obtained.

[0115] Send a prompt message to the user, prompting the user to upload the process plan and model parameters and conduct the transaction. Determine whether the feedback instruction input by the user is received. If the feedback instruction is received, determine the target process plan and / or target model parameters to be uploaded based on the iterative selection corresponding to the feedback instruction and the optimized process plan and / or optimized model parameters, and upload the target process plan and / or target model parameters. If the feedback instruction is not received, save the process plan and model parameters locally for local users to use.

[0116] In summary, the embodiments of the present application provide a chemical process collaborative optimization method based on a distributed training model deployment system. The distributed training model deployment system includes a cloud and at least one edge device. The cloud is provided with a general chemical training model and / or a first chemical process special training model. The edge device is used for chemical process execution and is provided with a second chemical process special training model. An optional communication connection is provided between the cloud and at least one edge device. The method is applied to the edge device, and the method includes: in response to a target compound generation instruction, generating an initial process plan using the second chemical process special training model; obtaining the execution result of the initial process plan executed by the edge device, and generating execution result evaluation data based on the execution result; optimizing and training the second chemical process special training model based on the execution data of the initial process plan and the execution result evaluation data to obtain an optimized process plan and its corresponding optimized model parameters that meet the preset evaluation data; when receiving a feedback instruction, sending the optimized process plan and / or the optimized model parameters to the cloud to update the general chemical training model and / or the first chemical process special training model.

[0117] The use of local models deployed locally on edge devices to generate initial process plans can meet the real-time requirements of process optimization when edge devices are offline, and can also make full use of the learning ability and efficiency of local models to achieve accurate and efficient generation of initial process plans; through the execution ability of local models, the execution results of the initial process plan can be executed, and the execution results corresponding to the initial process plan can be obtained, so as to quickly obtain the execution results and avoid the high cost and efficiency problems caused by manual repeated experiments, thereby improving the efficiency of process optimization and reducing the cost of process optimization; through the evaluation and analysis of the execution results by local models, it can be determined whether the execution results corresponding to the initial process plan meet the preset requirements, so as to evaluate the execution results of the initial process plan. The process plan is evaluated and improved, and the local model is trained based on the execution data of the initial process plan and the evaluation data of the execution results to optimize the local model, so as to ensure that the optimized local model can well match the current edge device, so as to ensure that the process plan optimized based on the local model can well match the current edge device; through the closed-loop processing process of "generation-execution-evaluation-training", while meeting the real-time requirements of the edge device, the learning ability of the local model is improved, and the optimized local model and the edge device can be matched in real time. It is determined that the process plan optimized based on the local model is highly matched with the current edge device, which improves the accuracy of the optimized process plan, and based on the accurate process plan, the process development efficiency can be significantly improved and the resource consumption of process development can be reduced.

[0118] Furthermore, by selectively sending optimized process plans and model parameters to the cloud, collaborative knowledge sharing between the cloud and edge devices is achieved, while also balancing the relationship between knowledge openness and knowledge competitiveness. This promotes chemical process advancement while safeguarding the competitiveness of its own technology, thereby achieving both knowledge sharing and controllable privacy. Initial process plans are automatically generated through models (a general chemical training model, a first chemical process-specific training model, and a second chemical process-specific training model), reducing manual trial and error costs. Data is also dynamically adjusted to reduce consumption. Technological competitiveness is formed through privatized models, and feedback is used to promote technological development and industry collaboration, achieving controllable privacy. Knowledge sharing is achieved, balancing privacy and knowledge sharing.

[0119] Example 2:

[0120] Another embodiment of the present application relates to a distributed training model deployment system. The following describes the implementation details of the distributed training model deployment system of this embodiment. The following content is only for the convenience of understanding the implementation details and is not necessary for the implementation of this solution. The edge devices in the system of this solution can be used to execute the chemical process collaborative optimization method based on the distributed training model deployment system in the above embodiments. The distributed training model deployment system of this embodiment includes:

[0121] In the cloud, a general chemical training model and / or a first chemical process-specific training model are provided;

[0122] An edge device for chemical process execution, provided with a second chemical process-specific training model, and a selectable communication connection between the cloud and at least one edge device; wherein the edge device is further configured to:

[0123] In response to the target compound generation instruction, an initial process plan is generated using the second chemical process-specific training model; an execution result of the initial process plan executed by the edge device is obtained, and execution result evaluation data is generated based on the execution result;

[0124] Optimize and train the second chemical process special training model based on the execution data of the initial process plan and the execution result evaluation data to obtain an optimized process plan and its corresponding optimized model parameters that meet the preset evaluation data;

[0125] When a feedback instruction is received, the target process scheme and / or target model parameters to be uploaded are determined based on the iterative selection corresponding to the feedback instruction and the optimized process scheme and / or optimized model parameters, and the target process scheme and / or target model parameters are sent to the cloud to update the general chemical training model and / or the first chemical process special training model, wherein the optimization degree of the target process scheme and / or target model parameters is less than or equal to the optimized process scheme and / or optimized model parameters.

[0126] Example 3:

[0127] Another embodiment of the present application relates to a chemical process collaborative optimization device based on a distributed training model deployment system, which is arranged on an edge device, which is used for chemical process execution and is provided with a second chemical process-specific training model. The following is a detailed description of the implementation details of the chemical process collaborative optimization device based on a distributed training model deployment system of this embodiment. The following content is only for the convenience of understanding the implementation details, and is not necessary for the implementation of this solution. The device of this solution can be used to execute the chemical process collaborative optimization method based on a distributed training model deployment system in the above embodiments. The schematic diagram of the chemical process collaborative optimization device 60 based on a distributed training model deployment system of this embodiment can be as follows Figure 7 As shown, it includes a response module 601, an acquisition module 602, a training module 603 and a feedback module 604.

[0128] A response module 601 is configured to generate an initial process plan using a second chemical process-specific training model in response to a target compound generation instruction; an acquisition module 602 is configured to obtain the execution result of the initial process plan executed by the edge device and generate execution result evaluation data based on the execution result; a training module 603 is configured to optimize the second chemical process-specific training model based on the execution data of the initial process plan and the execution result evaluation data to obtain an optimized process plan and its corresponding optimized model parameters that meet the preset evaluation data;

[0129] The feedback module 604 is used to determine the target process scheme and / or target model parameters to be uploaded based on the iterative selection corresponding to the feedback instruction and the optimized process scheme and / or optimized model parameters when receiving the feedback instruction, and send the target process scheme and / or target model parameters to the cloud to update the general chemical training model and / or the first chemical process special training model, wherein the optimization degree of the target process scheme and / or target model parameters is less than or equal to the optimized process scheme and / or optimized model parameters.

[0130] It is worth mentioning that all modules involved in this embodiment are logical modules. In actual applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, to highlight the innovation of this application, this embodiment does not include units that are not closely related to solving the technical problem proposed by this application. However, this does not mean that other units do not exist in this embodiment.

[0131] Example 4:

[0132] Another embodiment of the present application relates to an electronic device, such as Figure 8 As shown, it includes: at least one processor 901; and a memory 902 communicatively connected to the at least one processor 901; wherein the memory 902 stores instructions that can be executed by the at least one processor 901, and the instructions are executed by the at least one processor 901 so that the at least one processor 901 can execute the chemical process collaborative optimization method based on the distributed training model deployment system in the above-mentioned embodiments.

[0133] The memory and processor are connected using a bus, which can include any number of interconnected buses and bridges. The bus connects various circuits of one or more processors and memories. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits. These are all well known in the art and are therefore not described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single component or multiple components, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over a wireless medium via an antenna. Furthermore, the antenna receives data and transmits it to the processor.

[0134] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory can be used to store data used by the processor when performing operations.

[0135] Embodiment 5:

[0136] Another embodiment of the present application relates to a computer-readable storage medium storing a computer program, which implements the above method embodiment when executed by a processor.

[0137] That is, those skilled in the art will understand that all or part of the steps in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a program, which is stored in a storage medium and includes a number of instructions for causing a device (which may be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc., various media that can store program code.

[0138] Those skilled in the art will appreciate that the above embodiments are specific embodiments for implementing the present application, and that in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present application.

Claims

1. A chemical process collaborative optimization method based on a distributed training model deployment system, characterized in that: The distributed training model deployment system includes a cloud and at least one edge device, wherein the cloud is provided with a general chemical training model and / or a first chemical process-specific training model, and the edge device is used for chemical process execution and is provided with at least a portion of a second chemical process-specific training model. The cloud and at least one edge device have an optional communication connection, and the method is applied to the edge device, the method comprising: In response to the target compound generation instruction, generating an initial process plan using the second chemical process-specific training model; Obtaining an execution result of the initial process plan executed by the edge device, and generating the execution result evaluation data based on the execution result; Optimizing and training the second chemical process-specific training model based on the execution data of the initial process plan and the execution result evaluation data to obtain an optimized process plan and its corresponding optimized model parameters that meet the preset evaluation data; When a feedback instruction is received, the target process scheme and / or target model parameters to be uploaded are determined based on the iterative selection corresponding to the feedback instruction and the optimized process scheme and / or optimized model parameters, and the target process scheme and / or target model parameters are sent to the cloud to update the general chemical training model and / or the first chemical process special training model, wherein the optimization degree of the target process scheme and / or target model parameters is less than or equal to the optimized process scheme and / or optimized model parameters.

2. The chemical process collaborative optimization method based on the distributed training model deployment system according to claim 1 is characterized in that: Obtaining an execution result of the initial process plan executed by the edge device, including: Sending the initial process plan to the cloud, so that the cloud optimizes the initial process plan based on the general chemical training model and / or the first chemical process-specific training model to obtain an optimized initial process plan; Receiving the optimized initial process plan sent by the cloud, and obtaining the execution result of the optimized initial process plan executed by the edge device; and / or Receive process parameter adjustment data corresponding to the process parameter adjustment instruction, optimize and adjust the initial process plan based on the process parameter adjustment data to obtain the adjusted and optimized initial process plan, and obtain the execution result of the edge device executing the adjusted and optimized initial process plan.

3. The chemical process collaborative optimization method based on the distributed training model deployment system according to claim 2 is characterized in that: Before the cloud optimizes the initial process plan based on the general chemical training model and / or the first chemical process-specific training model, the method further includes: The cloud obtains the optimization model parameters sent by the other edge devices when receiving the feedback instruction; The cloud updates the general chemical training model and / or the first chemical process-specific training model based on the optimized model parameters; and When the initial process plan is received, the initial process plan is optimized based on the updated general chemical training model and / or the first chemical process-specific training model.

4. The chemical process collaborative optimization method based on a distributed training model deployment system according to claim 1, characterized in that: The optimizing training of the second chemical process-specific training model based on the execution data of the initial process plan and the execution result evaluation data includes: Obtaining first historical execution data and first historical execution result evaluation data stored locally on the edge device, as well as the execution data of the initial process plan and the execution result evaluation data; and / or receiving process parameter adjustment data corresponding to the process parameter adjustment instruction; Based on the first historical execution data, the evaluation data of the first historical execution results, the execution data of the initial process plan, the evaluation data of the execution results of the initial process plan and / or the process parameter adjustment data, the second chemical process special training model is subjected to offline optimization training.

5. The chemical process collaborative optimization method based on the distributed training model deployment system according to 1 is characterized in that: The optimizing training of the second chemical process-specific training model based on the execution data of the initial process plan and the execution result evaluation data includes: Determining a process complexity corresponding to the target compound generation instruction based on the initial process plan and / or the execution result evaluation data; When the process complexity meets the preset conditions, a distributed training strategy corresponding to the optimization training is generated; the distributed training strategy includes at least one other edge device participating in the training and training content corresponding to each of the other edge devices, and the training content at least includes an exit condition corresponding to the optimization training; The training content is sent to each of the other edge devices through the cloud, so that each of the other edge devices performs optimization training based on the training content.

6. The chemical process collaborative optimization method based on a distributed training model deployment system according to claim 5, characterized in that: Also includes: Each of the other edge devices sends the optimization parameters obtained by the optimization training to the cloud, so that the optimization parameters are sent to the edge device through the cloud; The edge device updates the second chemical process-specific training model based on the optimization parameters.

7. The chemical process collaborative optimization method based on a distributed training model deployment system according to claim 1, characterized in that: The sending of the optimized process plan and / or optimized model parameters to the cloud comprises: The optimization process scheme and / or optimization model parameters are encrypted according to a preset encryption strategy, and the encrypted optimization model parameters are sent to the cloud.

8. A distributed training model deployment system, characterized in that: include: In the cloud, a general chemical training model and / or a first chemical process-specific training model are provided; an edge device for chemical process execution and provided with at least a portion of a second chemical process-specific training model, and a selectable communication connection between the cloud and at least one of the edge devices; Among them, edge devices are also used for: In response to the target compound generation instruction, generating an initial process plan using the second chemical process-specific training model; Obtaining an execution result of the initial process plan executed by the edge device, and generating the execution result evaluation data based on the execution result; Optimizing and training the second chemical process-specific training model based on the execution data of the initial process plan and the execution result evaluation data to obtain an optimized process plan and its corresponding optimized model parameters that meet the preset evaluation data; When a feedback instruction is received, the target process scheme and / or target model parameters to be uploaded are determined based on the iterative selection corresponding to the feedback instruction and the optimized process scheme and / or optimized model parameters, and the target process scheme and / or target model parameters are sent to the cloud to update the general chemical training model and / or the first chemical process special training model, wherein the optimization degree of the target process scheme and / or target model parameters is less than or equal to the optimized process scheme and / or optimized model parameters.

9. A chemical process collaborative optimization device based on a distributed training model deployment system, characterized in that: The device is provided on an edge device, the edge device being used for chemical process execution and provided with a second chemical process-specific training model, and the device includes: a response module, configured to generate an initial process plan using the second chemical process-specific training model in response to the target compound generation instruction; an acquisition module, configured to acquire an execution result of the initial process plan executed by the edge device, and generate the execution result evaluation data based on the execution result; A training module, configured to perform optimization training on the second chemical process-specific training model based on the execution data of the initial process plan and the execution result evaluation data, to obtain an optimized process plan and its corresponding optimized model parameters that meet the preset evaluation data; A feedback module is used to determine the target process scheme and / or target model parameters to be uploaded based on the iterative selection corresponding to the feedback instruction and the optimized process scheme and / or optimized model parameters when receiving a feedback instruction, and send the target process scheme and / or target model parameters to the cloud to update the general chemical training model and / or the first chemical process special training model, wherein the optimization degree of the target process scheme and / or target model parameters is less than or equal to the optimized process scheme and / or optimized model parameters.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it implements the chemical process collaborative optimization method based on the distributed training model deployment system as described in any one of claims 1-11.

Images