A resource collaborative control method and system based on agent-based AI

Abstract

The present invention discloses a resource collaborative control method and system based on agent-based AI. The method includes: an enterprise management system transmits a production plan to a manufacturing execution system; the manufacturing execution system determines production task requirements based on the production plan and a scheduling model; a production control system determines a production control plan based on equipment occupancy, production task requirements, and a task planning model, and issues the first production planning control of the production control plan to a resource controller and issues the second production planning control of the production control plan to production equipment; the resource controller determines a target resource allocation according to the first production planning control, the current computing and network resource allocation situation, and a resource configuration model to send a resource determination message; the production equipment receives the resource determination message and executes tasks according to the second production planning control; the above technical solution can achieve intelligent collaboration between systems and precise resource allocation, and promote the industrial production line towards intelligence, collaboration, and high efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of resource collaborative control, and particularly to a resource collaborative control method and system based on agent-based AI. Background Art

[0002] Industrial Internet promotes the digital transformation of manufacturing industry and enhances industrial competitiveness. At present, the Industrial Internet has entered a new stage of large-scale development. In this process, as the core link of industrial manufacturing, the intelligent and efficient operation of production lines is crucial. However, with the expansion of production scale and the improvement of product complexity, the traditional production line control mode increasingly exposes many drawbacks, such as rigid resource allocation, slow response, poor coordination, etc., and it is difficult to meet the urgent needs of modern industry for high-precision and high-flexibility production.

[0003] At the same time, it brings new opportunities for change in the industrial field. An artificial intelligence system with the ability of autonomous decision-making and action can autonomously plan tasks and execute operations according to preset goals in a complex and changeable environment, and support real-time adjustment of strategies to cope with uncertainties. However, due to the dynamic and variable demands of various tasks on the production line for computing power and network bandwidth, the unified management of edge, network, and cloud devices is very complex, and there are often phenomena such as idle computing power in some periods and insufficient bandwidth at critical moments. In addition, the lack of deep integration between computing network resource management and industrial application scenarios restricts the maximum exertion of collaborative efficiency. Summary of the Invention

[0004] In view of this, the present invention provides a resource collaborative control method and system based on agent-based AI, which can realize intelligent collaboration between systems and precise allocation of resources, and effectively promote the industrial production line to move towards a higher stage of intelligence, collaboration, and efficiency.

[0005] According to one aspect of the present invention, an embodiment of the present invention provides a resource collaborative control method based on agent-based AI, which is applied to a resource collaborative control system based on agent-based AI. The resource collaborative control system includes: an enterprise management system, a manufacturing execution system, a production control system, a resource controller, and production equipment; wherein, the manufacturing execution system includes a first slave agent-based AI module, and the first slave agent-based AI module includes a production scheduling model; the production control system includes a second slave agent-based AI module, and the second slave agent-based AI module includes a task planning model; the resource controller includes a third slave agent-based AI module, and the third slave agent-based AI module includes a resource allocation model;

[0006] Correspondingly, the resource collaborative control method based on agent-based AI includes:

[0007] The enterprise management system receives a production plan issued by a user and transmits the production plan to the manufacturing execution system;

[0008] The manufacturing execution system determines the production task requirements based on the production plan and the production scheduling model;

[0009] The production control system determines the equipment occupancy of the production equipment, determines the production control plan based on the equipment occupancy, the production task requirements, and the task planning model, and issues the first production planning control of the production control plan to the resource controller and the second production planning control of the production control plan to the production equipment; wherein, the first production planning control is the computing power requirement and network requirement of the production equipment required by the production plan, and the second production planning control is the production operation of the required production line;

[0010] The resource controller obtains the current computing and network resource allocation of the production equipment, determines the target resource allocation corresponding to the production equipment according to the first production planning control, the current computing and network resource allocation, and the resource configuration model, and after the target resource allocation is successful, sends a resource determination message to the production equipment;

[0011] The production equipment receives the resource determination message and executes the task according to the second production planning control.

[0012] According to another aspect of the present invention, an embodiment of the present invention further provides a resource collaboration control system based on an agent-based AI, and the resource collaboration control system includes: an enterprise management system, a manufacturing execution system, a production control system, a resource controller, and production equipment; wherein, the manufacturing execution system includes a first slave agent-based AI module, and the first slave agent-based AI module includes a production scheduling model; the production control system includes a second slave agent-based AI module, and the second slave agent-based AI module includes a task planning model; the resource controller includes a third slave agent-based AI module, and the third slave agent-based AI module includes a resource configuration model;

[0013] Among them, the enterprise management system is used to receive the production plan issued by the user and transmit the production plan to the manufacturing execution system;

[0014] The manufacturing execution system is used to determine the production task requirements based on the production plan and the production scheduling model;

[0015] The production control system is used to determine the equipment occupancy of the production equipment, determine the first production planning control and the second production planning control based on the equipment occupancy, the production task requirements and the task planning model, and send the first production planning control to the resource controller and the second production planning control to the production equipment; wherein, the first production planning control is the computing power requirement and network requirement of the production equipment required by the production plan, and the second production planning control is the production operation of the required production line.

[0016] The resource controller is used to obtain the current computing and network resource allocation of the production equipment, determine the target resource allocation corresponding to the production equipment according to the first production planning control, the current computing and network resource allocation and the resource configuration model, and send a resource determination message to the production equipment after the target resource allocation is successful.

[0017] The production equipment is used to receive the resource determination message and execute the task according to the second production planning control.

[0018] The technical effect of the present invention is that the enterprise management system transmits the production plan to the manufacturing execution system; the manufacturing execution system determines the production task requirements based on the production plan and the production scheduling model, and the production control system determines the production control plan based on the equipment occupancy, the production task requirements and the task planning model, and sends the first production planning control of the production control plan to the resource controller and the second production planning control of the production control plan to the production equipment; the resource controller determines the target resource allocation according to the first production planning control, the current computing and network resource allocation and the resource configuration model to send a resource determination message; the production equipment receives the resource determination message and executes the task according to the second production planning control. Through the mutual cooperation of the above enterprise management system, production control system, resource controller and production equipment, each system is configured with a corresponding slave agent type AI module, and the distributed agent type AI module is used to cooperate and manage the production process and computing and network resources, realizing intelligent cooperation and precise resource allocation, and promoting the industrial production line to move towards intelligence, coordination and high efficiency.

[0019] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. Brief Description of the Drawings

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0021] Figure 1 Flowchart of a resource collaborative control method based on agent-based AI provided by an embodiment of the present invention;

[0022] Figure 2 Structural schematic diagram of a distributed agent-based AI provided by an embodiment of the present invention;

[0023] Figure 3 Flowchart of another resource collaborative control method based on agent-based AI provided by an embodiment of the present invention;

[0024] Figure 4 Flow schematic diagram of the production description string generation process in a manufacturing execution system provided by an embodiment of the present invention;

[0025] Figure 5 Architectural schematic diagram of a resource collaborative control system based on agent-based AI provided by an embodiment of the present invention;

[0026] Figure 6 Architectural schematic diagram of another resource collaborative control system based on agent-based AI provided by an embodiment of the present invention;

[0027] Figure 7 Relationship schematic diagram between a main agent-based AI module and slave agent-based AI modules corresponding to each system provided by an embodiment of the present invention. Detailed implementation manners

[0028] In order to enable those skilled in the art to better understand the solution of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0029] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of the present invention are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present invention described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily limit to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0030] In one embodiment, Figure 1 FIG. is a flowchart of a resource collaborative control method based on agent-based AI provided by an embodiment of the present invention. This embodiment is applicable to the situation of collaborative control of production lines and computing network resources in the industrial field. This method can be executed by a resource collaborative control system based on agent-based AI.

[0031] As Figure 1 shown, the resource collaborative control method based on agent-based AI in this embodiment is applied to a resource collaborative control system based on agent-based AI. The resource collaborative control system includes: an enterprise management system, a manufacturing execution system, a production control system, a resource controller, and production equipment; wherein, the manufacturing execution system includes a first slave agent-based AI module, and the first slave agent-based AI module includes a production scheduling model; the production control system includes a second slave agent-based AI module, and the second slave agent-based AI module includes a task planning model; the resource controller includes a third slave agent-based AI module, and the third slave agent-based AI module includes a resource allocation model.

[0032] In this embodiment, the resource collaborative control method based on agent-based AI specifically includes the following steps:

[0033] S110. The enterprise management system receives the production plan issued by the user and transmits the production plan to the manufacturing execution system.

[0034] Among them, the production plan can be understood as a production plan for producing a certain product on a certain production line in the industrial field.

[0035] The resource controller in this embodiment can be understood as a controller for a network, computing power, or computing power network, etc., and is used to allocate resources such as networks and computing power.

[0036] In this embodiment, the enterprise management system records various information such as production orders, contracts, and materials, which can be used to implement various information management of the enterprise. Of course, the user can formulate a production plan for a certain product according to the order, etc. in the enterprise management system. The enterprise management system receives the production plan issued by the user and then transmits the production plan to the manufacturing execution system for subsequent processing by the manufacturing execution system. In this embodiment, the interaction between the enterprise information management system and the manufacturing execution system is to improve production efficiency and information fluency. This kind of interaction can achieve seamless data circulation, reduce the phenomenon of information silos, thereby providing real-time production data and business analysis, and optimizing the overall business process. For example, in the flexible production and manufacturing scenario, the user can provide various customized product requirements, and the system of this embodiment can automatically adjust the production process or resources according to the new order.

[0037] S120. The manufacturing execution system determines the production task requirements based on the production plan and the scheduling model.

[0038] Among them, the production task requirements may specifically include: production order number, production content description, customized requirement description, and production deadline. Of course, the production task requirements can also be increased according to user needs, and this embodiment does not limit this here.

[0039] In this embodiment, the manufacturing execution system is an integrated software system designed for the manufacturing industry, and its goal is to improve the execution efficiency of the entire production process, coordinate each production stage, monitor production activities in real time, and provide detailed production data. The scheduling model is a local model configured in the manufacturing execution system. This local model can be a pre-trained scheduling model. The scheduling model can be an architecture of a deep neural network and is used to generate production task requirements. The training of this scheduling model can include first collecting data and using the collected data as a training set for training to optimize the model parameters until the model parameters reach the optimal state to obtain a pre-trained scheduling model. Among them, the collected data may include, but is not limited to, historical production plan information corresponding to the production plan, equipment / resource information, process route and process information, and production time data (preparation time, processing time), etc.

[0040] In this embodiment, the manufacturing execution system includes a first slave-agent type AI module. The first slave-agent type AI module includes a production scheduling model. When the manufacturing execution system receives a production plan, it decomposes the production plan into multiple demand statements, and then inputs the multiple demand statements into the production scheduling model in the first slave-agent type AI module, so as to output a description of specific production task requirements, and send the description of production task requirements to the production control system. It can be understood that when the manufacturing execution system receives a production plan, it will optimize and adjust these plans according to factors such as the current production status, equipment capacity, and raw material inventory to ensure the smooth progress of production. Specifically, the first slave-agent type AI deployed on the manufacturing execution system decomposes the production order into a production scheduling plan, takes the old production scheduling plan as input and the new production scheduling plan as output. In some embodiments, in addition to the production scheduling model, the first slave-agent type AI module may further include a supervision model, which can be used to supervise and review the relevant data input into the production scheduling model, so as to optimize the data. Specifically, the manufacturing execution system uses the production plan, the historical production plan, and / or the historical supervision feedback information of the second supervision model in the production control system as the first initial information, then fills it into the requirement template of the production product to obtain the corresponding first production description text, splits the first production description text into at least two first implementation demand statements, adds the first implementation demand statements to the first production queue, and then uses the first supervision model to review the first implementation demand statements in the first production queue. In the case of passing the review, the first implementation demand statements are taken out from the first production queue in turn, input into the production scheduling model to obtain new first implementation demands, use the first supervision model to conduct a secondary review of the new first implementation demands, and combine the new first implementation demands that pass the secondary review to form the production task requirements until all the first implementation demand statements in the first production queue are processed; if not qualified, clear the first production queue and execute again.

[0041] In this embodiment, the manufacturing execution system will send the optimized production plan to the production control system. The production control system is a system close to the actual production process and can convert the production instructions transmitted by the manufacturing execution system into signals that can be understood by the production equipment, so as to control the operation of the equipment.

[0042] S130. The production control system determines the equipment occupancy of the production equipment, determines the production control plan based on the equipment occupancy, the production task requirements, and the task planning model, and sends the first production planning control of the production control plan to the resource controller and the second production planning control of the production control plan to the production equipment.

[0043] Among them, the equipment occupancy can be understood as whether the production equipment is occupied. The production equipment may include, but is not limited to, network controllers, edge computing devices, production line equipment, and cloud computing services, etc.

[0044] In this embodiment, the first production planning control is for the computing power requirements and network requirements of the production equipment required by the production plan, and the second production planning control is for the production operations of the required production lines. It can be understood that the production control scheme includes two parts. One part describes the computing power and network requirements of each device for the production process, and this description is sent to the resource controller; the other part describes the specific production operations of the production line, and this description is sent to the production equipment. It should be noted that the issuance of the first production planning control and the second production planning control in this embodiment is a parallel operation. It can be understood that after obtaining the production control scheme, the first production planning control and the second production planning control are simultaneously issued to the corresponding devices.

[0045] In one embodiment, the first production planning control at least includes: service number, application type, computing power requirements, and network requirements; among them, the computing power requirements at least include the total computing delay, the number of computing units, and the input data size; the network requirements at least include: the total transmission delay, the minimum bandwidth, and the maximum jitter; the second production planning control at least includes: the production equipment address, the application port number, the operation parameters, and the execution time. It should be noted that after the production operation is given by the task planning model, it needs to be executed by applications such as production equipment and edge computing. If these applications communicate remotely, the network is required. In this process, in order to perform resource scheduling more precisely, the application type needs to be clarified. For example, the computing power and network requirements for image recognition are relatively large, while the requirements for sensors are relatively small.

[0046] In this embodiment, the production control system includes a second slave-agent type AI module. The second slave-agent type AI module includes a task planning model, which can be understood as a local model configured in the production control system. This local model can be a pre-trained task planning model for outputting a production control scheme. The task planning model is a model with a deep neural network architecture, and its pre-training can also include first collecting historical relevant data and using the collected historical data as a training set for training to optimize the model parameters until the model parameters reach the optimal value to obtain a pre-trained task planning model. In some embodiments, in addition to including the task planning model, the second slave-agent type AI module can also include a supervision model, which can be used to supervise and review the relevant data input into the task planning model, so as to optimize the data. It can be understood that the agent type AI of the production control system is a module that further decomposes the production plan into production tasks of each device. This module has a high degree of intelligence and automation, and can automatically generate an optimal production task plan according to the complexity of the production plan and the production capacity of the equipment.

[0047] Specifically, the production control system uses the equipment occupancy situation, production task requirements, historical equipment occupancy situation, and / or the historical supervision feedback information of the third supervision model in the resource controller as the second initial information, determines the second production description text based on the second initial information, splits the second production description text into at least two second implementation requirement statements, adds the second implementation requirement statements to the second production queue, uses the second supervision model to review the second implementation requirement statements in the second production queue. If the review is qualified, the second implementation requirement statements are taken out in sequence, the second implementation requirement statements are input into the task planning model to obtain at least two new second implementation requirements, the second supervision model is used to conduct a secondary review of the new second implementation requirements, the new second implementation requirements that pass the secondary review are combined to form the first production planning control and the second production planning control, and the new second implementation requirements that do not pass the secondary review are reviewed again until all the second implementation requirement statements in the second production queue are processed.

[0048] S140. The resource controller obtains the current computing and network resource allocation situation of the production equipment, determines the target resource allocation corresponding to each production equipment according to the first production planning control, the current computing and network resource allocation situation, and the resource configuration model. After the target resource allocation is successful, a resource determination message is sent to the production equipment.

[0049] Among them, the target resource allocation can be understood as the resource allocation situation finally corresponding to the production equipment.

[0050] In this embodiment, the resource configuration model can be understood as the local model configured in the resource controller. This local model can be a pre-trained resource configuration model, which is used to output the target resource allocation corresponding to each production equipment. The resource configuration model is a model with a deep neural network architecture, and its pre-training can also include first collecting historical relevant data and using the collected historical data as the training set for training to optimize the model parameters until the model parameters reach the optimal value to obtain the pre-trained resource configuration model.

[0051] In this embodiment, the resource controller performs network configuration and computing power configuration. After receiving the computing power and network requirements, the resource controller creates deterministic services for each step of the production process. The deterministic service is a closed-loop business chain composed of multiple computing and communication processes. The slave-agent type AI module in the resource controller determines the target resource allocation corresponding to each production equipment according to the requirements of the first part of the solution and the managed computing and network resource allocation situation, the first production planning control, the current computing and network resource allocation situation, and the resource configuration model, and then generates specific network, edge, and cloud configurations. After the resource allocation is successful, the resource controller sends a resource confirmation message to the production equipment.

[0052] In this embodiment, the resource controller includes a third slave-agent type AI module. The third slave-agent type AI module includes a resource configuration model. Of course, in addition to the resource configuration model, the resource controller may also include a supervision model, which can be used to review the relevant data input into the resource configuration model. It can be understood that the third slave-agent type AI module is a module for adjusting the computing network resource scheduling scheme according to the production plan, and can adjust the computing power, bandwidth, execution order, etc. according to the complexity of the production plan and the production capacity of the device, so that the total execution delay and jitter of all links are controlled within the expected value range. Specifically, the resource controller uses the current computing network resource allocation situation, the first production planning control, and / or the historical computing network resource allocation situation as the third initial information, determines the corresponding third production description text according to the third initial information, splits the third production description text into at least two third implementation requirement statements, uses the third supervision model to review the third implementation requirement statements. In the case of passing the review, the third implementation requirement statements are taken out in sequence, the third implementation requirement statements are input into the task planning model to obtain at least two third new implementation requirements, the third new implementation requirements are secondarily reviewed using the third supervision model, and the third new implementation requirements that pass the secondary review are formed into the target resource allocation until each third implementation requirement statement in the third production queue is processed.

[0053] S150. The production device receives the resource determination message and executes the task according to the second production planning control.

[0054] In this embodiment, when the production device receives the resource determination message and executes the task according to the second production planning control, it can be understood that after receiving these two pieces of information, the production device starts to perform specific production operations and computing network configurations; the network controller performs specific network configurations; the edge computing device and the cloud computing service perform specific supporting applications and computing network configurations. More specifically, the network controller performs network configurations and regularly reports network status information; the production line device performs production operations and regularly reports production status information; the edge computing device runs the edge computing tasks sent by the resource controller and regularly reports the edge computing task status information; the cloud computing service runs the cloud computing tasks sent by the resource controller and regularly reports the cloud computing task status information.

[0055] In the technical solution of the embodiment of the present invention, the enterprise management system transmits the production plan to the manufacturing execution system; the manufacturing execution system determines the production task requirements based on the production plan and the scheduling model, and the production control system determines the production control plan based on the equipment occupancy, production task requirements, and task planning model, and issues the first production planning control of the production control plan to the resource controller, and issues the second production planning control of the production control plan to the production equipment. Then, the resource controller determines the target resource allocation according to the first production planning control, the current computing and network resource allocation situation, and the resource configuration model to send a resource determination message. Finally, the production equipment receives the resource determination message and executes the task according to the second production planning control. Through the mutual cooperation of the enterprise management system, the production control system, the resource controller, and the production equipment, a corresponding slave agent type AI module is configured in each system, and the distributed slave agent type AI module is used to perform collaborative management and control on the production process and computing and network resources, realizing intelligent collaboration and precise resource allocation, and promoting the industrial production line to move towards intelligence, collaboration, and high efficiency.

[0056] In one embodiment, the resource collaborative control system further includes: a master agent type AI module; the master agent type AI module includes a global model; correspondingly, the resource collaborative control method further includes:

[0057] The scheduling model in the first slave agent type AI module, the task planning model in the second slave agent type AI module, and the resource configuration model in the third slave agent type AI module respectively feedback the corresponding model parameters to the global model in the master agent type AI module, so that the global model optimizes the model parameters of the scheduling model, the task planning model, and the resource configuration model respectively according to the corresponding model parameters, and updates the scheduling model, the task planning model, and the resource configuration model with optimized parameters to the corresponding systems.

[0058] Among them, the master agent type AI module can be deployed in other systems, computer devices, or electronic devices other than the systems where the first slave agent type AI module, the second slave agent type AI module, and the third slave agent type AI module are deployed. In this embodiment, the global model aggregates the models of all slave agent type AI modules and performs overall training and optimization, and then issues the trained local models in each slave agent type AI, so that the local models are regularly kept consistent.

[0059] In this embodiment, the resource collaboration control system further includes: a main agent type AI module, which includes a global model. The global model can be an interactive large model and is used to manage the production scheduling model in the first slave agent type AI module, the task planning model in the second slave agent type AI module, and the resource allocation model in the third slave agent type AI module. This management can include, but is not limited to, separately updating the model parameters of the production scheduling model in the first slave agent type AI module, the task planning model in the second slave agent type AI module, and the resource allocation model in the third slave agent type AI module, so as to better optimize the model optimization capabilities of each model. Specifically, the production scheduling model in the first slave agent type AI module can feedback the corresponding model parameters to the global model in the main agent type AI module, and the global model tunes the model parameters to update the production scheduling model with the tuned parameters to the first slave agent type AI module in the manufacturing execution system; similarly, the task planning model in the second slave agent type AI module and the resource allocation model in the third slave agent type AI module will also separately feedback the corresponding model parameters to the global model in the main agent type AI module, so that the global model tunes the model parameters of the task planning model and the resource allocation model according to the corresponding model parameters, and updates the task planning model and the resource allocation model with the tuned parameters to the second slave agent type AI module and the third slave agent type AI module respectively.

[0060] It should be noted that the parameters of the global model mainly include input layer parameters, hidden layer parameters, output layer parameters, activation function parameters, loss function parameters, optimizer parameters, regularization parameters, etc., which are the sum of the association parameters between neurons and the internal parameters of neurons in the neural network.

[0061] In this embodiment, when the global model in the main agent type AI module performs parameter tuning, personalized federated learning or a centralized tuning strategy can be used for parameter tuning. Exemplarily, tuning based on gradient transmission (centralized supervision). In this method, the slave agents need to upload the gradients of the local training data to the main agent. The main agent calculates the global gradient direction, generates the tuned parameters, and then the main agent distributes the adjusted parameters to the slave agents. Of course, it can also be other parameter tuning methods in the prior art, and this embodiment does not limit it here.

[0062] In one embodiment, the main agent type AI module further includes a global supervision model; correspondingly, the first slave agent type AI module further includes: a first supervision model; the second slave agent type AI module further includes a second supervision model; the third slave agent type AI module includes a third supervision model; correspondingly, the resource collaboration control method further includes:

[0063] The first supervision model in the first slave-agent AI module, the second supervision model in the second slave-agent AI module, and the third supervision model in the third slave-agent AI module respectively feedback the corresponding supervision parameters to the global supervision model in the master-agent AI module, so that the global supervision model optimizes the corresponding supervision model according to the corresponding supervision parameters.

[0064] Among them, the first supervision model, the second supervision model, and the third supervision model are all models with a neural network architecture. Exemplarily, it can be Hidden Markov, Support Vector Machine, etc. In this embodiment, the supervision parameters corresponding to each supervision model may include, but are not limited to, demand templates, demand parameters, supervision algorithms, heuristic function thresholds, etc., and can be stored in the form of JSON or XML files.

[0065] In this embodiment, the master-agent AI module manages each slave-agent AI module, and there is an order execution relationship among the slave-agent AI modules. The output of the predecessor agent AI module is the input of the successor agent AI module. The model of the master-agent AI module is used to train and update the models of multiple slave-agent AI modules. At the same time, the feedback results of the supervision module of the successor agent AI are used to adjust and optimize the supervision module of the predecessor agent AI. Exemplarily, the first slave-agent AI module is the predecessor node of the second slave-agent AI module, and the second slave-agent AI module is the predecessor node of the third slave-agent AI module; the second slave-agent AI module is the successor node of the first slave-agent AI module.

[0066] In this embodiment, the main task of supervised learning is to learn a mapping function from existing labeled data so that for a given input, the corresponding output can be obtained. In supervised learning, a labeled dataset is usually used for training, and each sample in the training set has an input and a corresponding output. During the training process, the model learns the mapping relationship between the input and the output, so as to learn a function that can map new inputs to the correct outputs.

[0067] In this embodiment, the first supervision model in the first slave agent type AI module, the second supervision model in the second slave agent type AI module, and the third supervision model in the third slave agent type AI module respectively check the rationality of the corresponding production description string requirements, and evaluate whether the existing resources can meet the requirements of the production task. If the required computing power and network exceed the existing resources, the insufficient resource situation will be fed back to the predecessor node supervision module. The predecessor node supervision module adds the feedback information to the context information, which affects the subsequent generation of the production description string. At the same time, the production description string of this service is regenerated. It can be understood that the second slave agent type AI module makes a first evaluation based on the review result of the second supervision model to determine whether the computing power resources and network resources can meet the requirements of the production task. If the requirements of the production task are not met, the result of the first evaluation is used as the first feedback information and fed back to the first supervision model in the first slave agent type AI module, so that the first slave agent type AI module uses the first feedback information as the historical supervision feedback information of the first supervision model; similarly, the third slave agent type AI module makes a second evaluation based on the review result of the third supervision model to determine whether the computing power resources and network resources can meet the requirements of the production task. If the requirements of the production task are not met, the result of the second evaluation is used as the second feedback information and fed back to the historical supervision feedback information of the second supervision model.

[0068] In this embodiment, the master agent type AI module further includes a global supervision model. The first slave agent type AI module further includes: a first supervision model, the second slave agent type AI module further includes a second supervision model, and the third slave agent type AI module includes a third supervision model. The first supervision model in the first slave agent type AI module, the second supervision model in the second slave agent type AI module, and the third supervision model in the third slave agent type AI module in this embodiment respectively feed the corresponding supervision parameters back to the global supervision model in the master agent type AI module, so that the global supervision model optimizes the corresponding supervision model according to the corresponding supervision parameters. Specifically, the method for optimizing the supervision parameters can be gradient correction supervision.

[0069] It can be understood that the master agent type AI module is responsible for the management, training, and update of the supervision model and the global model, and completes the unified planning logic through the master agent type AI module. The resource scheduling plan may include the allocation and scheduling of resources such as computing power and network. Through reasonable resource scheduling, the continuity, efficiency, and certainty of the production process can be ensured. At the same time, the master agent type AI module also has the global optimization ability, and can dynamically adjust and optimize the global model according to the supervision results of the slave agent type AI modules on each device.

[0070] In this embodiment, the main agent-based AI module maintains the entire architecture system, regularly evaluates the operation of the entire production line, and updates the execution templates and supervision parameters of each subordinate agent-based AI module in each system. The subordinate agent-based AI takes the requirements of the predecessor node agent-based AI as input, and after multiple rounds of review and optimization by the local supervision module, outputs the local configuration and the requirements of the successor node. In addition, the main agent-based AI module is also responsible for coordinating and managing the communication and computing tasks between each subordinate agent-based AI module, ensuring that the strategies of the subordinate agent-based AI modules on multiple devices tend to be consistent. Its main functions include model aggregation, model update and distribution, parameter transfer and encryption, task management and scheduling, visualization and monitoring, etc.

[0071] In one embodiment, for better understanding of the structure of the distributed agent-based AI, Figure 2 FIG. is a schematic structural diagram of a distributed agent-based AI provided by an embodiment of the present invention. Figure 2 The subordinate Agerntic AI in it is the subordinate agent-based AI model, representing the first subordinate agent-based AI module, the second subordinate agent-based AI module, and the third subordinate agent-based AI module. Each subordinate agent-based AI module includes a corresponding local supervision model and a local model, that is, in the above embodiment, the first subordinate agent-based AI module includes a first supervision model and a production scheduling model; the second subordinate agent-based AI module includes a second supervision model and a task planning model; the third subordinate agent-based AI module includes a third supervision model and a resource allocation model. In this embodiment, the main agent-based AI module is responsible for the management, training, and update of the supervision models and local models in each other system. The resource scheduling plan may include the allocation and scheduling of resources such as computing power and network. Through reasonable resource scheduling, the continuity, efficiency, and certainty of the production process can be ensured. At the same time, the main agent-based AI also has the ability of global optimization, and can dynamically adjust and optimize the global model according to the supervision results of the subordinate agent-based AI on each device.

[0072] In one embodiment, Figure 3 FIG. is a flowchart of another resource collaborative control method based on agent-based AI provided by an embodiment of the present invention. On the basis of the above embodiments, this embodiment further details that the manufacturing execution system determines the production task requirements based on the production plan and the production scheduling model; the production control system determines the equipment occupancy of the production equipment, and determines the production control plan based on the equipment occupancy, the production task requirements, and the task planning model; the resource controller obtains the current computing and network resource allocation of the production equipment, and determines the target resource allocation corresponding to each production equipment according to the first production planning control, the current computing and network resource allocation, and the resource allocation model, and when the production equipment receives the resource determination message, executes the task according to the second production planning control.

[0073] As Figure 3As shown in the figure, the resource collaborative control method based on the agent - type AI in this embodiment may specifically include the following steps:

[0074] S310. The enterprise management system receives the production plan issued by the user and transmits the production plan to the manufacturing execution system.

[0075] S320. The manufacturing execution system determines the production task requirements based on the production plan and the scheduling model.

[0076] In one embodiment, S320 may specifically include: S3201 - S3206. Specifically:

[0077] S3201. The manufacturing execution system takes the production plan and the first context information as the first initial information.

[0078] In this embodiment, the manufacturing execution system receives the production plan, and then takes the production plan and the first context information as the first initial information. Among them, the first context information includes: the historical production plan and / or the historical supervision feedback information of the second supervision model in the production control system. Among them, the historical supervision feedback information can be understood as the feedback of the historical review results of the second supervision model.

[0079] S3202. Fill the first initial information into the requirement template of the production product to obtain the corresponding first production description text.

[0080] Among them, the first production description text is the production description text information corresponding to the first initial information. The requirement template can also be called the demand template, and this requirement template can supplement more necessary information, such as the size, weight, and other information of the production product.

[0081] In this embodiment, first set the corresponding keywords according to the user requirements, and then fill the initial information into the requirement template of the production product through the keywords to form an executable first production description text.

[0082] S3203. Split the first production description text into at least two first implementation requirement statements, and add the first implementation requirement statements to the first production queue.

[0083] In this embodiment, the first production description text can be split into at least two first implementation requirement statements according to various delimiters (for example, line break characters, html tags, etc.), and the first implementation requirement statements are added to the first production queue to wait for the review of the requirement statements. Among them, the first implementation requirement statements are multiple implementation requirement statements corresponding to the splitting of the first production description text.

[0084] S3204. Use the first supervision model to review the first implementation requirement statement in the first production queue, and determine whether the first implementation requirement statement is qualified. If it is qualified, execute S3205; if it is unqualified, execute S3206.

[0085] In this embodiment, the first supervision model in the first slave agent type AI module is used to review the first implementation requirement statement in the first production queue. Specifically, the review may include, but is not limited to, reviewing the syntax, semantics, and rationality of the requirement statements in the generation queue, and then determining whether the first implementation requirement statement is qualified. In the case of qualification, the first slave agent type AI module processes the statements in the generation queue in a loop, inputs each statement into the production scheduling model, and writes each input implementation requirement into the production description string to be output. The production scheduling model will feedback multiple new implementation requirements, and the first supervision model reviews these implementation requirements.

[0086] S3205. When the first production queue is not empty, sequentially take out the first implementation requirement statement from the first production queue, input the first implementation requirement statement into the production scheduling model to obtain a new first implementation requirement, use the first supervision model to conduct a secondary review of the new first implementation requirement, mark the new first implementation requirement that passes the secondary review as the first leaf node, form production task requirements with each first leaf node, and re-add the new first implementation requirement that fails the secondary review to the first production queue, and return to the step of using the first supervision model to review the first implementation requirement statement in the production queue until all the first implementation requirement statements in the first production queue are processed.

[0087] Among them, the first leaf node is a string, and all the first leaf nodes are concatenated together to form production task requirements, and then sent to the production description string of the next slave agent module (the second slave agent type AI module).

[0088] In this embodiment, in the case of being qualified, the first implementation requirement statement is sequentially taken out from the first production queue, the first implementation requirement statement is input into the production scheduling model to obtain a new first implementation requirement, the new first implementation requirement is secondarily reviewed using the first supervision model, the new first implementation requirement that passes the secondary review is marked as the first leaf node, each first leaf node forms a production task requirement, and the new first implementation requirement that fails the secondary review is re-added to the first production queue, and the step of reviewing the first implementation requirement statement in the production queue using the first supervision model is returned until all the first implementation requirement statements in the first production queue are processed; in the case of being unqualified, the first production queue is emptied, the step of taking the production plan and the first context information as the first initial information is returned, and the process is re-executed. Exemplarily, the input from the agent module in the manufacturing execution system is an order content of the enterprise information system. The agent AI supervision module forms a more complete production description according to the requirement template of the production product of model XX. For example, which parts are needed, processing steps, etc. Then the production description is input into the large model to obtain production task requirements, such as when to start production, in which manufacturing system to produce, raw material inventory processing, etc.

[0089] S3206. Empty the first production queue, return the step of taking the production plan and the first context information as the first initial information, and re-execute.

[0090] S330. The production control system determines the equipment occupancy of the production equipment, and determines the first production planning control and the second production planning control based on the equipment occupancy, the production task requirements, and the task planning model.

[0091] In one embodiment, S330 may specifically include: S3301 - S3305. Specifically:

[0092] S3301. The production control system takes the equipment occupancy and the second context information as the second initial information.

[0093] In this embodiment, the production control system takes the equipment occupancy and the second context information as the second initial information to perform subsequent operations. Among them, the second context information includes: production task requirements, historical equipment occupancy, and / or historical supervision feedback information of the third supervision model in the resource controller.

[0094] It should be noted that the function of the second supervision model included in the second slave agent type AI module in the production control system is similar to that of the first supervision model included in the first slave agent type AI module in the manufacturing execution system. This embodiment will not provide specific introduction. Similarly, the same applies to the third supervision model included in the third slave agent type AI module in the resource controller in the following text. It can be understood that the processes of the production control system, the manufacturing execution system, and the resource controller for processing information are the same, except for the content of the information processed.

[0095] S3302. Determine the second production description text based on the second initial information, split the second production description text into at least two second implementation requirement statements, and add the second implementation requirement statements to the second production queue.

[0096] In this embodiment, the second production description text is obtained by filling the requirements template for the production product with the second initial information to obtain the production description text.

[0097] In this embodiment, corresponding keywords are set according to the user requirements first, and then the second initial information is filled into the requirements template for the production product through the keywords to form an executable second production description text. Then, the second production description text is split into multiple second implementation requirement statements according to certain requirements, and the second implementation requirement statements are added to the second production queue.

[0098] S3303. Use the second supervision model to review the second implementation requirement statements in the second production queue and determine whether the second implementation requirement statements are qualified. If they are qualified, execute S3304; if they are unqualified, execute S3305.

[0099] In this embodiment, the second supervision model in the second slave agent type AI module is used to review the second implementation requirement statements in the second production queue, and determine whether the second implementation requirement statements are qualified. If they are qualified and the second production queue is not empty, the second implementation requirement statements are sequentially taken out from the second production queue, and the second implementation requirement statements are input into the task planning model to obtain at least two new second implementation requirements. The second supervision model is used to conduct a secondary review of the new second implementation requirements. The new second implementation requirements that pass the secondary review are marked as second leaf nodes, and each second leaf node forms a first production planning control and a second production planning control. The new second implementation requirements that do not pass the secondary review are re-added to the second production queue, and the step of using the second supervision model to review the second implementation requirement statements in the second production queue is returned until all the second implementation requirement statements in the second production queue are processed. Exemplarily, the input from the agent module of the production manufacturing system is a production scheduling plan. The supervision model will form the production steps composed of production tasks according to the production scheduling plan and the requirement template of the product. This requirement template for producing the product should be designed according to the factory production line. Each production task corresponds to a specific link in the production line. The content of the output text is the production task and its detailed requirements. For example, for the equipment of visual inspection, a camera is required to monitor, the picture is transmitted to the edge computing, and after the edge computing makes a decision, the result is transmitted to the robotic arm.

[0100] S3304. When the second production queue is not empty, the second implementation requirement statements are sequentially taken out from the second production queue, and the second implementation requirement statements are input into the task planning model to obtain at least two new second implementation requirements. The second supervision model is used to conduct a secondary review of the new second implementation requirements. The new second implementation requirements that pass the secondary review are marked as second leaf nodes, and each second leaf node forms a first production planning control and a second production planning control. The new second implementation requirements that do not pass the secondary review are re-added to the second production queue, and the step of using the second supervision model to review the second implementation requirement statements in the second production queue is returned until all the second implementation requirement statements in the second production queue are processed.

[0101] In this embodiment, the second slave agent type AI module makes a first evaluation based on the review result of the second supervision model to determine whether the computing power resources and network resources can meet the requirements of the production tasks. If they do not meet the requirements of the production tasks, the result of the first evaluation is used as the first feedback information and fed back to the first supervision model in the first slave agent type AI module, so that the first slave agent type AI module uses the first feedback information as the historical supervision feedback information of the first supervision model.

[0102] S3305. Clear the second production queue, return to the step of using the equipment occupancy situation and the second context information as the second initial information, and execute again.

[0103] S340. The resource controller obtains the current computing and network resource allocation situation of the production equipment, and determines the target resource allocation corresponding to the production equipment according to the first production plan control, the current computing and network resource allocation situation, and the resource configuration model.

[0104] In one embodiment, S340 may specifically include: S3401 - S3405. Specifically:

[0105] S3401. The resource controller uses the current computing and network resource allocation situation and the third context information as the third initial information.

[0106] In this embodiment, the resource controller uses the current computing and network resource allocation situation and the third context information as the third initial information. Among them, the third context information includes: the first production plan control and / or the historical computing and network resource allocation situation.

[0107] S3402. Determine the corresponding third production description text based on the third initial information, split the third production description text into at least two third implementation requirement statements, and add the third implementation requirement statements to the third production queue.

[0108] In this embodiment, first set the corresponding keywords according to the user requirements, then fill the third initial information into the requirement template of the production product through the keywords to form an executable third production description text, and then split the third production description text into multiple third implementation requirement statements according to certain requirements, and add the third implementation requirement statements to the second production queue.

[0109] S3403. Use the third supervision model to review the third implementation requirement statements in the third production queue, and determine whether the third implementation requirement statements are qualified. If they are qualified, execute S3404; if they are unqualified, execute S3405.

[0110] In this embodiment, the third supervision model in the third slave agent type AI module reviews the third implementation requirement statement in the third production queue and determines whether the third implementation requirement statement is qualified. If it is qualified, and when the third production queue is not empty, the third implementation requirement statement is sequentially taken out from the third production queue, the third implementation requirement statement is input into the task planning model to obtain at least two third new implementation requirements, the third supervision model is used to conduct a secondary review of the third new implementation requirements, the third new implementation requirements that pass the secondary review are marked as third leaf nodes, the third leaf nodes form the target resource allocation, and the third new implementation requirements that do not pass the secondary review are re-added to the third production queue, and the step of using the third supervision model to review the third implementation requirement statement in the third production queue is returned until each third implementation requirement statement in the third production queue is processed; if it is unqualified, the third production queue is cleared, the step of using the current computing and network resource allocation situation and the third context information as the third initial information is returned, and the process is re-executed. Exemplarily, the input of the slave agent module of the resource controller is a production task. For example, after the robotic arm receives what signal, it performs an assembly operation. The slave agent model outputs the required network configuration and computing power configuration according to the local model, and the output is the resource configuration information.

[0111] S3404. When the third production queue is not empty, the third implementation requirement statement is sequentially taken out from the third production queue, the third implementation requirement statement is input into the task planning model to obtain at least two third new implementation requirements, the third supervision model is used to conduct a secondary review of the third new implementation requirements, the third new implementation requirements that pass the secondary review are marked as third leaf nodes, the third leaf nodes form the target resource allocation, and the third new implementation requirements that do not pass the secondary review are re-added to the third production queue, and the step of using the third supervision model to review the third implementation requirement statement in the third production queue is returned until each third implementation requirement statement in the third production queue is processed.

[0112] In this embodiment, the third slave agent type AI module conducts a second evaluation based on the review result of the third supervision model to determine whether the computing power resources and network resources can meet the requirements of the production task. If they do not meet the requirements of the production task, the result of the second evaluation is used as the second feedback information and fed back to the historical supervision feedback information of the second supervision model.

[0113] S3405. Clear the third production queue, return to the step of using the current computing and network resource allocation situation and the third context information as the third initial information, and re-execute.

[0114] Exemplarily, for better understanding of the production description string generation process in each process Figure 4A flowchart showing the process of generating a production description string in a manufacturing execution system provided by an embodiment of the present invention. In this embodiment, an example is given where the manufacturing execution system determines production task requirements based on a production plan and a scheduling model. The production description string in this embodiment can be understood as the string corresponding to the production task requirements in the above embodiment. The local supervision model in this embodiment can be understood as the first supervision model in the above embodiment. The next slave-agent AI module in this embodiment can be understood as the second slave-agent AI module in the above embodiment.

[0115] a1. Construct initial information from the input and context information.

[0116] a2. Split the production description text into multiple implementation requirements and input the implementation requirements into the generation queue.

[0117] a3. Use the first supervision model to review the implementation requirements in the generation queue.

[0118] a4. Determine whether the implementation requirements are qualified. If not, execute a5; if qualified, execute a6.

[0119] a5. Clear the generation queue and return to a1.

[0120] a6. Determine whether the generation queue is empty. If so, execute a7; if not, execute a8.

[0121] a7. Send the production description string to the next slave-agent AI module.

[0122] a8. The first slave-agent AI module retrieves the first requirement statement in the generation queue, inputs the requirement statement into the scheduling model, and then concatenates it to the production description string to be output.

[0123] a9. The first supervision model reviews multiple new implementation requirements fed back by the scheduling model.

[0124] a10. Determine whether the new implementation requirements have been processed. If so, return to a6; if not, execute a11.

[0125] a11. Select an implementation requirement and determine whether it meets the implementation conditions. If it does, execute a12; if not, execute a13.

[0126] a12. Mark it as a leaf node and return to a10 until all leaf nodes are obtained, and use each leaf node as a production task requirement.

[0127] a13. Re-add it to the generation queue and return to a10.

[0128] The S350 network controller performs network configuration and regularly reports network status information.

[0129] In this embodiment, the network controller is software or a device that manages a deterministic network, supports the local deployment of the Agentic AI module, manages network devices, maintains the network topology, and regularly reports network status information.

[0130] The S360 production line equipment performs production operations and regularly reports production status information.

[0131] In this embodiment, the input of the production equipment from the agent module is the production task, and the output is the operation instruction.

[0132] The S370 edge computing device runs the edge computing tasks sent by the resource controller and regularly reports the edge computing task status information.

[0133] In this embodiment, the edge computing device is an edge computing power device that executes deterministic applications, including but not limited to edge servers, edge controllers, edge all-in-ones, etc., supports the deployment of Agentic AI on this device, runs the computing tasks of deterministic execution, and regularly reports the computing task status information.

[0134] The S380 cloud computing service runs the cloud computing tasks sent by the resource controller and regularly reports the cloud computing task status information.

[0135] In this embodiment, the cloud computing service is a cloud service that executes deterministic applications, supports the deployment of Agentic AI in this service, runs the computing tasks of deterministic execution, and regularly reports the computing task status information. The inputs of the edge computing and cloud services from the agent module are production tasks, and the outputs are operation instructions and computing power configuration information.

[0136] The technical solution of the embodiment of the present invention, through a distributed structure, can deploy a dedicated agent-based AI module on different systems, form a consistent management strategy through centralized control, and finally form an intelligent management and operation and dynamic optimization solution, while taking into account the reduction of management and operation costs and agile control optimization, so as to further realize intelligent cooperation between devices and precise allocation of resources, and promote the industrial production line towards intelligence, collaboration, and high efficiency.

[0137] In one embodiment, Figure 5 It is a schematic diagram of the architecture of a resource collaborative control system based on agent-based AI provided by an embodiment of the present invention.

[0138] As Figure 5As shown in the figure, the resource collaboration control system includes: an enterprise management system 510, a manufacturing execution system 520, a production control system 530, a resource controller 540, and production equipment 550; among them, the manufacturing execution system 520 includes a first slave-agent type AI module, and the first slave-agent type AI module includes a production scheduling model; the production control system 530 includes a second slave-agent type AI module, and the second slave-agent type AI module includes a task planning model; the resource controller 540 includes a third slave-agent type AI module, and the third slave-agent type AI module includes a resource allocation model;

[0139] Among them, the enterprise management system 510 is used to receive the production plan issued by the user and transmit the production plan to the manufacturing execution system 520;

[0140] The manufacturing execution system 520 is used to determine the production task requirements based on the production plan and the production scheduling model;

[0141] The production control system 530 is used to determine the equipment occupancy of the production equipment, and determine the first production planning control and the second production planning control based on the equipment occupancy, the production task requirements, and the task planning model, and send the first production planning control to the resource controller 540 and send the second production planning control to the production equipment 550; among them, the first production planning control is the computing power requirement and network requirement of the production equipment required by the production plan, and the second production planning control is the production operation of the required production line;

[0142] The resource controller 540 is used to obtain the current computing and network resource allocation situation of the production equipment, determine the target resource allocation corresponding to the production equipment according to the first production planning control, the current computing and network resource allocation situation, and the resource allocation model, and after the target resource allocation is successful, send a resource determination message to the production equipment;

[0143] The production equipment 550 is used to receive the resource determination message and execute the task according to the second production planning control.

[0144] In an embodiment, Figure 6 This is a schematic diagram of the architecture of another resource collaboration control system based on the agent type AI provided by an embodiment of the present invention. As Figure 6As shown in the figure, the resource collaboration control system includes: an enterprise management system 610, a manufacturing execution system 620, a production control system 630, a resource controller 640, production equipment 650, and a master agent-based AI module 660; among them, the manufacturing execution system 620 includes a first slave agent-based AI module, and the first slave agent-based AI module includes a production scheduling model and a first supervision model; the production control system 630 includes a second slave agent-based AI module, and the second slave agent-based AI module includes a task planning model and a second supervision model; the resource controller 640 includes a third slave agent-based AI module, and the third slave agent-based AI module includes a resource allocation model and a third supervision model; the master agent-based AI module 660 includes a global model and a global supervision model.

[0145] 1) Interaction between the enterprise information management system and the MES (Manufacturing Execution System): When the MES receives a production plan, it will optimize and adjust these plans based on factors such as the current production status, equipment capabilities, and raw material inventory to ensure the smooth progress of production. The slave agent-based AI deployed on the MES decomposes the order into a production scheduling plan, taking the old production scheduling plan as the input and the new production scheduling plan as the output.

[0146] 2) Interaction between the MES and the production control system: The MES will send the optimized production plan to the production control system. The agent-based AI of the production control system is a module that further decomposes the production plan into production tasks for each device. This module has a high degree of intelligence and automation capabilities and can automatically generate the optimal production task plan according to the complexity of the production plan and the production capabilities of the devices.

[0147] 3) Resource scheduling of the resource controller: The slave agent-based AI of the resource controller is a module used to adjust the computing network resource scheduling plan according to the production plan. It can adjust the computing power, bandwidth, and execution order according to the complexity of the production plan and the production capabilities of the devices, so that the total execution delay and jitter of all links are controlled within the expected value range.

[0148] 4) Control of edge, network, and cloud devices: Edge, network, and cloud devices directly execute remote control instructions. Optionally, the slave agent-based AI translates the requirements and executes them.

[0149] 5) Global control of the distributed agent-based AI: The master agent-based AI is responsible for the management, training, and update of the supervision modules and models, and completes the unified planning logic through the master agent. The resource scheduling plan may include the allocation and scheduling of resources such as computing power and network. Through reasonable resource scheduling, the continuity, efficiency, and determinacy of the production process can be ensured. At the same time, the master agent-based AI also has global optimization capabilities and can dynamically adjust and optimize the global model according to the supervision results of the slave agent-based AI on each device.

[0150] In this embodiment, to facilitate a better understanding of the relationship between the master-agent type AI module and the slave-agent type AI modules corresponding to each system, Figure 7 FIG. is a schematic diagram showing the relationship between a master-agent type AI module and slave-agent type AI modules corresponding to each system according to an embodiment of the present invention.

[0151] It should be understood that various forms of the processes shown above can be used, steps can be reordered, added or deleted. For example, the steps described in the present invention can be executed in parallel, sequentially, or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and no limitations are imposed herein.

[0152] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A resource collaborative control method based on agent-based AI, characterized in that: Applied to a resource collaborative control system based on agent-type AI, the resource collaborative control system includes: an enterprise management system, a manufacturing execution system, a production control system, a resource controller and production equipment; wherein the manufacturing execution system includes a first slave agent-type AI module, the first slave agent-type AI module includes a production scheduling model; the production control system includes a second slave agent-type AI module, the second slave agent-type AI module includes a task planning model; the resource controller includes a third slave agent-type AI module, the third slave agent-type AI module includes a resource configuration model; Accordingly, the resource collaborative control method based on agent-based AI includes: The enterprise management system receives a production plan issued by a user, and transmits the production plan to the manufacturing execution system; The manufacturing execution system determines the production task demand based on the production plan and the scheduling model; The production control system determines the equipment occupancy of the production equipment, determines the production control scheme based on the equipment occupancy, the production task requirements and the task planning model, and sends the first production planning control of the production control scheme to the resource controller, and sends the second production planning control of the production control scheme to the production equipment; wherein the first production planning control is the computing power requirement and network requirement of the production equipment required for the production plan, and the second production planning control is the production operation of the required production line; The resource controller obtains the current computing network resource allocation status of the production equipment, determines the target resource allocation corresponding to the production equipment respectively according to the first production planning control, the current computing network resource allocation status and the resource configuration model, and sends a resource confirmation message to the production equipment after the target resource allocation is successful; The production equipment receives the resource confirmation message and controls the execution of tasks according to the second production plan; Wherein, the first slave agent AI module further includes: a first supervisory model; The manufacturing execution system determines the production task requirements based on the production plan and the production scheduling model, including: The manufacturing execution system uses the production plan and the first context information as the first initial information; wherein the first context information includes: historical production plans and / or historical supervisory feedback information of the second supervisory model in the production control system; Filling the first initial information into a production product requirement template to obtain a corresponding first production description text; Splitting the first production description text into at least two first implementation requirement statements, and adding the first implementation requirement statements to a first production queue; Using a first supervisory model to review a first implementation requirement statement in the first production queue, and determining whether the first implementation requirement statement is qualified; If qualified, then when the first production queue is not empty, take out the first implementation requirement statements from the first production queue in turn, input the first implementation requirement statements into the production scheduling model to obtain new first implementation requirements, use the first supervision model to conduct a second review of the new first implementation requirements, mark the new first implementation requirements that pass the second review as first leaf nodes, each of the first leaf nodes forms the production task requirements, and re-add the new first implementation requirements that fail the second review to the first production queue, and return to the step of using the first supervision model to review the first implementation requirement statements in the production queue until each of the first implementation requirement statements in the first production queue is processed; If unqualified, the first production queue is cleared, and the process returns to the step of using the production plan and the first context information as the first initial information and is executed again.

2. The resource collaborative control method according to claim 1, characterized in that: The resource collaborative control system further includes: a master-agent AI module; the master-agent AI module includes a global model; Accordingly, the resource collaborative control method further includes: The production scheduling model in the first slave agent AI module, the task planning model in the second slave agent AI module, and the resource allocation model in the third slave agent AI module respectively feed back corresponding model parameters to the global model in the master agent AI module, so that the global model optimizes the model parameters of the production scheduling model, the task planning model, and the resource allocation model according to the corresponding model parameters, so as to update the production scheduling model, task planning model, and resource allocation model after parameter optimization to the corresponding systems.

3. The resource collaborative control method according to claim 2, characterized in that: The master agent AI module also includes a global supervision model; correspondingly, the second slave agent AI module also includes a second supervision model; the third slave agent AI module includes a third supervision model; Accordingly, the resource collaborative control method further includes: The first supervisory model in the first slave-agent AI module, the second supervisory model in the second slave-agent AI module, and the third supervisory model in the third slave-agent AI module respectively feed back corresponding supervisory parameters to the global supervisory model in the master-agent AI module, so that the global supervisory model optimizes the corresponding supervisory model according to the corresponding supervisory parameters.

4. The resource collaborative control method according to claim 1, characterized in that: The production task requirements include at least: production order number, production content description, customization requirement description and production deadline.

5. The resource collaborative control method according to claim 1, characterized in that: The production control system determines the equipment occupancy of the production equipment, and determines the first production planning control and the second production planning control based on the equipment occupancy, the production task requirements and the task planning model, including: The production control system uses the equipment occupancy and the second context information as the second initial information; wherein the second context information includes: production task requirements, historical equipment occupancy, and / or historical supervision feedback information of the third supervision model in the resource controller; Determine a second production description text according to the second initial information, divide the second production description text into at least two second implementation requirement statements, and add the second implementation requirement statements to a second production queue; Using a second supervisory model to review the second implementation requirement statement in the second production queue, and determining whether the second implementation requirement statement is qualified; If qualified, then when the second production queue is not empty, take out the second implementation requirement statements from the second production queue in turn, input the second implementation requirement statements into the task planning model to obtain at least two new second implementation requirements, use the second supervision model to conduct a secondary review of the new second implementation requirements, mark the new second implementation requirements that pass the secondary review as second leaf nodes, form the first production planning control and the second production planning control with each second leaf node, and re-add the new second implementation requirements that fail the secondary review to the second production queue, and return to the step of using the second supervision model to review the second implementation requirement statements in the second production queue until the processing of each second implementation requirement statement in the second production queue is completed; If unqualified, the second production queue is cleared, and the process returns to the step of using the device occupancy status and the second context information as the second initial information and is executed again.

6. The resource collaborative control method according to claim 1, characterized in that: The resource controller obtains the current network computing resource allocation of the production equipment, and determines the target resource allocation corresponding to each of the production equipment according to the first production planning control, the current network computing resource allocation and the resource configuration model, including: The resource controller uses the current computing network resource allocation situation and the third context information as the third initial information; wherein the third context information includes: the first production planning control and / or the historical computing network resource allocation situation; Determine a corresponding third production description text according to the third initial information, divide the third production description text into at least two third implementation requirement statements, and add the third implementation requirement statements to a third production queue; Using a third supervisory model to review a third implementation requirement statement in the third production queue, and determining whether the third implementation requirement statement is qualified; If qualified, then if the third production queue is not empty, take out the third implementation requirement statements from the third production queue in turn, input the third implementation requirement statements into the task planning model to obtain at least two third new implementation requirements, use the third supervision model to conduct a secondary review of the third new implementation requirements, mark the third new implementation requirements that pass the secondary review as third leaf nodes, form target resource allocation with the third leaf nodes, and re-add the third new implementation requirements that fail the secondary review to the third production queue, and return to the step of using the third supervision model to review the third implementation requirement statements in the third production queue until each third implementation requirement statement in the third production queue is processed; If unqualified, the third production queue is cleared, and the process returns to the step of using the current computing network resource allocation status and the third context information as the third initial information and is executed again.

7. The resource collaborative control method according to any one of claims 4 to 6, characterized in that: The second slave agent AI module performs a first evaluation based on the review result of the second supervisory model to determine whether the computing resources and network resources can meet the requirements of the production task. If the requirements of the production task are not met, the result of the first evaluation is fed back to the first supervisory model in the first slave agent AI module as first feedback information, so that the first slave agent AI module uses the first feedback information as historical supervisory feedback information of the first supervisory model. The third slave agent AI module performs a second evaluation based on the review results of the third supervisory model to determine whether the computing power resources and network resources can meet the needs of the production task. If the needs of the production task are not met, the result of the second evaluation is fed back as the second feedback information to the historical supervisory feedback information of the second supervisory model.

8. The resource collaborative control method according to claim 1, characterized in that: The first production planning control includes at least: service number, application type, computing power requirement and network requirement; wherein the computing power requirement includes at least total calculation delay, number of computing units and input data size; the network requirement includes at least total transmission delay, minimum bandwidth and maximum jitter; The second production planning control includes at least: production equipment address, application port number, operation parameters and execution time.

9. The resource collaborative control method according to claim 1, characterized in that: The production equipment at least includes a network controller, a production line device, an edge computing device, and a cloud computing service; accordingly, the production equipment receives the resource determination message and controls the execution of tasks according to the second production plan, including: The network controller performs network configuration and regularly reports network status information; The production line equipment performs production operations and regularly reports production status information; The edge computing device runs the edge computing tasks issued by the resource controller and regularly reports edge computing task status information; The cloud computing service runs the cloud computing tasks issued by the resource controller and regularly reports cloud computing task status information.

10. A resource collaborative control system based on agent-based AI, characterized in that: The resource collaborative control system includes: an enterprise management system, a manufacturing execution system, a production control system, a resource controller and production equipment; wherein the manufacturing execution system includes a first slave agent AI module, the first slave agent AI module includes a production scheduling model; the production control system includes a second slave agent AI module, the second slave agent AI module includes a task planning model; the resource controller includes a third slave agent AI module, the third slave agent AI module includes a resource configuration model; Wherein, the enterprise management system is used to receive the production plan issued by the user and transmit the production plan to the manufacturing execution system; The manufacturing execution system is used to determine the production task requirements based on the production plan and the scheduling model; The production control system is used to determine the equipment occupancy of the production equipment, determine the first production planning control and the second production planning control based on the equipment occupancy, the production task requirements and the task planning model, and send the first production planning control to the resource controller, and send the second production planning control to the production equipment; wherein the first production planning control is the computing power requirement and network requirement of the production equipment required for the production plan, and the second production planning control is the production operation of the required production line; The resource controller is used to obtain the current network resource allocation status of the production equipment, determine the target resource allocation corresponding to the production equipment respectively according to the first production planning control, the current network resource allocation status and the resource configuration model, and send a resource confirmation message to the production equipment after the target resource allocation is successful; The production equipment is used to receive the resource confirmation message and control the execution of tasks according to the second production plan; Wherein, the first slave agent AI module further includes: a first supervisory model; The manufacturing execution system determines the production task requirements based on the production plan and the production scheduling model, including: The manufacturing execution system uses the production plan and the first context information as the first initial information; wherein the first context information includes: historical production plans and / or historical supervisory feedback information of the second supervisory model in the production control system; Filling the first initial information into a production product requirement template to obtain a corresponding first production description text; Splitting the first production description text into at least two first implementation requirement statements, and adding the first implementation requirement statements to a first production queue; Using a first supervisory model to review a first implementation requirement statement in the first production queue, and determining whether the first implementation requirement statement is qualified; If qualified, then when the first production queue is not empty, take out the first implementation requirement statements from the first production queue in turn, input the first implementation requirement statements into the production scheduling model to obtain new first implementation requirements, use the first supervision model to conduct a second review of the new first implementation requirements, mark the new first implementation requirements that pass the second review as first leaf nodes, each of the first leaf nodes forms the production task requirements, and re-add the new first implementation requirements that fail the second review to the first production queue, and return to the step of using the first supervision model to review the first implementation requirement statements in the production queue until each of the first implementation requirement statements in the first production queue is processed; If unqualified, the first production queue is cleared, and the process returns to the step of using the production plan and the first context information as the first initial information and is executed again.

Images