Multi-Agent-based workshop management system, cross-workshop migration method and system

Abstract

The invention provides a multi-Agent-based workshop management system and a cross-workshop migration method and system. The multi-Agent-based workshop management system comprises a sensing layer, an intelligent layer, an execution layer and a communication layer, a data acquisition module of the sensing layer is used for acquiring sensing data in a physical scene, and a sensing Agent of the sensing layer generates a corresponding virtual mirror image platform according to the sensing data and outputs an analysis result; the intelligent layer comprises a decision-making Agent and a coordination Agent, the decision-making Agent dynamically generates a global or local decision-making result according to an analysis result, and the coordination Agent monitors and evaluates the decision-making result through a preset mechanism and outputs an optimization strategy; the intelligent layer transmits the decision result and the optimization strategy to the execution layer; an execution Agent of the execution layer deploys a lightweight model through an edge calculation module, and analyzes a decision result and an optimization strategy into an equipment control signal to control equipment operation; and the communication layer provides an information interaction channel for the Agent of each layer. The limitation of a traditional layered architecture is broken through, and a self-organized, self-adaptive and self-optimized distributed intelligent ecological architecture is constructed.

Description

Technical Field

[0001] This application relates to the field of intelligent manufacturing technology, specifically to a multi-agent-based workshop management system, a cross-workshop migration method and system. Background Technology

[0002] With the development of the intelligent manufacturing industry, manufacturing enterprises have gradually achieved flexible, automated, and digital intelligent production within and between workshops. One of the typical characteristics of this intelligence is the division of labor and collaborative operation of various functional modules in the production scenario based on a hierarchical agent architecture. Although the traditional hierarchical architecture (task-workshop-equipment level) has solved the problem of decoupling responsibilities, it has bottlenecks in dynamic adaptability, global optimization efficiency, and heterogeneous equipment collaboration (such as inter-level communication latency, centralized scheduling becoming a bottleneck, and static rules being difficult to cope with flexible production). Summary of the Invention

[0003] The purpose of this application is to provide a multi-agent-based workshop management system, cross-workshop migration method and system, which breaks through the limitations of traditional layered architecture, builds a self-organizing, adaptive and self-optimizing distributed intelligent ecosystem architecture, improves resource utilization and enhances continuous evolution capabilities.

[0004] To address the aforementioned technical problems, this application provides a multi-agent-based workshop management system, comprising: a perception layer, an intelligence layer, an execution layer, and a communication layer. The perception layer includes a data acquisition module and a perception agent. The data acquisition module collects perception data from equipment groups, robot groups, sensor groups, factory-level digital systems, and human-machine input in the physical scene. The perception agent generates a corresponding virtual mirror platform based on the perception data and outputs analysis results based on the virtual mirror platform. The analysis results include at least predictions of equipment failures, assessments of task delay risks, and identification of intent to interrupt tasks. The intelligence layer includes a decision-making agent and a coordination agent. The decision agent dynamically generates global or local decision results based on the analysis results of the perception agent. The coordination agent monitors and evaluates the decision results through a preset mechanism and outputs optimization strategies. The intelligence layer transmits the decision results and optimization strategies to the execution layer. The execution layer includes an execution agent, which deploys a lightweight model through an edge computing module to parse the decision results and optimization strategies into device control signals to control device operation. The communication layer achieves rapid synchronization between physical devices and digital mirrors through a high-speed network, providing an information interaction channel for the agents of the perception layer, the intelligence layer, and the execution layer.

[0005] In one embodiment, the perception agent includes: a data fusion and mirroring module for fusing the perception data and generating a virtual mirror platform of the perception data based on scene map data; an intent understanding module for parsing the perception data based on the virtual mirror platform and outputting tag data that the agent can recognize; and a prediction module for performing prediction calculations based on the fused data and the tag data and outputting the analysis results.

[0006] In one embodiment, the data fusion and mirroring module includes: a data fusion module for fusing the perceived data into unified fused state data to eliminate data redundancy and conflicts; and a virtual scene modeling module for constructing a virtual scene that is one-to-one with the physical scene based on the fused state data, generating a virtual mirror platform that reflects the actual operating status; wherein the map data of the physical scene includes at least: point cloud map data generated based on point cloud registration or visual SLAM, device data, device motion data model, and boundary constraints and rules of the actual physical scene; the prediction module is also used to predict the state of the physical scene within a preset time based on the historical data and real-time online data of the virtual mirror platform, and output the analysis results.

[0007] In one embodiment, the perception agent further includes: a synchronization module, used to synchronize the perception data with the data of the virtual mirror platform; and to synchronize and update the data of the perception layer, the intelligence layer and the execution layer with the data of the virtual mirror platform in real time.

[0008] In one embodiment, the intelligent layer further includes a decision engine and a federated learning module; the decision engine trains a strategy generation model based on the perception data output by the perception layer through a trial-and-error reward learning mechanism; the decision agent uses the strategy generation model and dynamically generates global or local decision results based on the analysis results of the perception agent; the coordination agent monitors and evaluates the decision results based on the perception data and the label data, and performs conflict adjustment on conflicting strategies to generate conflict-free task instructions as the optimization strategy; the federated learning module aggregates the overall parameters of the workshop based on the decision results and the optimization strategy to generate a workshop-level global optimization strategy.

[0009] In one embodiment, the method by which the data fusion module fuses the perceived data into unified fused state data includes, but is not limited to: spatiotemporal alignment of data, feature matching and data source association, and deep fusion algorithms.

[0010] In one embodiment, the optimization strategy generated by the intelligent layer carries a priority, and the communication layer allocates bandwidth for communication according to the priority to avoid network congestion.

[0011] This application also provides a multi-agent-based cross-workshop migration method applied to the aforementioned workshop management system. The method includes: directly updating parameters and task interfaces for identical equipment; generating parameter mapping tables and adjusting constraints for heterogeneous equipment based on the original equipment parameter library through dynamic semantic matching; importing point cloud maps of the new workshop to replace the original workshop scene data; adjusting task parameters according to the new workshop constraints and migrating the original workshop strategy set to the new workshop; establishing a virtual mirror platform based on the perception data of the new workshop; wherein the virtual mirror platform performs virtual simulation verification of the new workshop management system based on input data, evaluating the functionality, decision-making and coordination results, and the effectiveness of execution instructions of the virtual mirror platform to complete the deployment of the new workshop.

[0012] In one embodiment, the virtual mirroring platform performs virtual simulation verification of the new workshop management system based on input data, evaluating the effectiveness of the mirroring platform's functions, decision-making and coordination results, and execution instructions, including:

[0013] Based on the input task data, assess whether the virtual image platform functions normally; based on the data from the virtual image platform, assess whether the scheduling strategy generated by the decision agent, the conflict resolution results of the coordination agent, and the instruction conversion logic of the execution agent meet the preset goals.

[0014] This application also provides a multi-agent-based cross-workshop migration system, applied to the aforementioned cross-workshop migration method, comprising: a cross-workshop adaptive module for migrating the multi-agent management system of the original workshop to the new workshop; the cross-workshop adaptive module specifically includes: an equipment mapping module for directly updating equipment parameters and task interfaces for identical equipment; for heterogeneous equipment, a parameter mapping table is generated based on the original equipment parameter library through dynamic semantic matching, and constraints are adjusted; a scene mapping module for importing the point cloud map of the new workshop to replace the scene map of the original workshop; a task and strategy mapping module for adjusting task parameters according to the constraints of the new workshop and migrating the strategy set of the original workshop to the new workshop; and a workshop mirroring module for establishing a virtual mirroring platform based on the perception data of the new workshop; wherein, the virtual mirroring platform performs virtual simulation verification of the new workshop management system based on the input data, evaluates the functionality, decision and coordination results, and effectiveness of the execution instructions of the virtual mirroring platform, so as to complete the deployment of the new workshop.

[0015] The multi-Agent architecture of this application breaks through the inherent framework of traditional architectures, shifting from the traditional centralized task management and centralized task parsing model to a distributed decision-making, autonomous monitoring, and automatic evolutionary multi-Agent architecture. By empowering agents with AI-driven "perception-decision-evolution" closed-loop capabilities, it constructs a self-organizing, adaptive, and self-optimizing distributed intelligent ecosystem, completely breaking through the limitations of traditional layered architectures. Attached Figure Description

[0016] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0017] Figure 1 This is a schematic diagram of the architecture of a multi-agent-based workshop management system according to an optional embodiment of this application;

[0018] Figure 2 This is a schematic diagram of a multi-agent-based workshop management system according to an optional embodiment of this application;

[0019] Figure 3 This is an architecture diagram of the perception layer according to an optional embodiment of this application;

[0020] Figure 4 This is a schematic diagram illustrating the transmission of data between different layers and the virtual image platform according to an optional embodiment of this application;

[0021] Figure 5 This is an architecture diagram of the smart layer according to an optional embodiment of this application;

[0022] Figure 6 This is an architecture diagram of the execution layer according to an optional embodiment of this application;

[0023] Figure 7 This is a flowchart of a multi-agent-based cross-workshop migration method according to an optional embodiment of this application;

[0024] Figure 8 This is a framework diagram of a multi-agent-based cross-workshop migration system according to an optional embodiment of this application;

[0025] Figure 9 This is a simulation architecture diagram of a multi-agent cross-workshop migration system according to an optional embodiment of this application. Detailed Implementation

[0026] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] It should be noted that the following description covers various aspects of embodiments within the scope of the appended claims. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this application, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number and aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0028] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. The drawings only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0029] Additionally, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that practice can be carried out without these specific details.

[0030] With the development of artificial intelligence technology, the traditional layered architecture (task agent - workshop agent - equipment-level agent) is no longer sufficient to meet the demands of more intelligent and adaptive industrial production, a system based on the "functional division of labor" concept from the era of industrial automation. In contrast, the self-organizing and adaptive multi-agent ecosystem based on AI, as a new AI-based architecture, can achieve a self-organizing, adaptive, and self-evolving multi-agent ecosystem, realizing a leap from "passively executing instructions" to "actively optimizing the whole," and can be considered the development direction of the next generation of smart factories. Its core advantages lie in data-driven dynamic decision-making, the high efficiency of distributed collaboration, and the flexible adaptation to heterogeneous equipment, representing the future trend of AI-centric agent deployment in smart workshops.

[0031] Based on this, this application proposes a multi-agent-based workshop management system. This system is an AI-based self-organizing and adaptive multi-agent ecosystem architecture. By empowering agents with a closed-loop capability of "perception-decision-evolution" through AI, it constructs a self-organizing, adaptive, and self-optimizing distributed intelligent ecosystem, completely breaking through the limitations of traditional layered architectures. The architecture of this multi-agent-based workshop management system is as follows: Figure 1 and Figure 2 As shown, it includes: a perception layer 1, an intelligence layer 2, an execution layer 3, and a communication layer 4. The perception layer 1 includes a data acquisition module 11 and a perception agent 12. The data acquisition module 11 is used to collect perception data from equipment groups, robot groups, sensor groups, factory-level digital systems, and human-machine input in the physical scene. The perception agent 12 generates a corresponding virtual image platform based on the perception data and outputs analysis results based on the virtual image platform. The analysis results include at least the prediction of equipment failure, the judgment of task delay risk, and the recognition of the intent of sudden queue-jumping tasks. The intelligence layer 2 includes a decision agent 21 and a coordination agent 22. The decision agent 21, based on the... The analysis results of the perception agent 12 dynamically generate global or local decision results. The coordination agent 22 monitors and evaluates the decision results through a preset mechanism and outputs optimization strategies. The intelligent layer 2 transmits the decision results and optimization strategies to the execution layer 3. The execution layer 3 includes an execution agent 31, which deploys a lightweight model through an edge computing module to parse the decision results and optimization strategies into device control signals to control device operation. The communication layer 4 realizes rapid synchronization between physical devices and digital mirrors through a high-speed network, providing an information interaction channel for the agents of the perception layer 1, the intelligent layer 2, and the execution layer 3.

[0032] For perception layer 1:

[0033] In this embodiment, the data acquisition module 11 in the perception layer 1 can be a multimodal sensor. The perception layer 1 collects perception data from equipment groups, robot clusters, sensor groups, factory-level digital systems, and human-machine input through the multimodal sensor, and processes data of different protocol formats into standardized format data to facilitate subsequent fusion processing. The perception agent 12 generates corresponding digital images of this information to perform tasks such as predicting equipment failures, judging the risk of task delays, and recognizing the intent of sudden queue-jumping tasks, and outputs the results to the intelligence layer 2.

[0034] Specifically, the perception agent 12 in this embodiment may include: a data fusion and mirroring module 121, an intent understanding module 122, and a prediction module 123, as shown in the reference. Figure 2 and Figure 3 The data fusion and mirroring module 121 is used to fuse the perceived data and generate a virtual mirroring platform for the perceived data based on the scene map data. This forms a unified mirrored data platform for the status data of equipment, robots, sensors, plant-level information, and human-machine input in the scene. The intent understanding module 122 parses the perceived data based on the virtual mirroring platform, such as parsing plant-level instructions and human-machine input information data, and converts them into data tags that the agent can understand, outputting tag data that the agent can recognize. The output (tag data) of the intent understanding module 122 and the mirroring platform data are used as inputs to the prediction module 123. The prediction module 123 performs prediction calculations based on the fused data and the tag data, such as prediction calculations for equipment status, task priority, and sudden queue interruption, and outputs the analysis results.

[0035] More specifically, the data fusion and mirroring module 121 in this embodiment is a digital virtualization of a physical workshop, mainly including a data fusion module and a virtual scene modeling module. The data fusion module is used to fuse the perceived data into unified fused state data, which has a unified physical state description to eliminate data redundancy and conflicts. Specifically, the data fusion module can generate unified fused state data through spatiotemporal alignment, feature matching and data source association, deep fusion algorithms, etc. Other methods can also be used in practice, and are not limited in this embodiment.

[0036] The virtual scene modeling module constructs a virtual scene that is one-to-one with the physical scene based on the fused state data, and generates a virtual mirror platform that reflects the actual operating conditions and states. The map data of the physical scene includes at least: point cloud map data generated based on point cloud registration or visual SLAM, device data, device motion data models, and boundary constraints and rules of the actual physical scene. In this embodiment, the prediction module 123 is also used to predict the state of the physical scene (including physical equipment and workshop scenes) within a preset time period based on the historical data and real-time online data of the virtual mirror platform, and outputs the analysis results to provide prior information data for the strategy generation of the intelligent layer 2.

[0037] The perception agent 12 also includes a synchronization module 124, for reference. Figure 2 It is used to synchronize the perceived data with the data of the virtual mirror platform; and to synchronize and update the data of the perception layer 1, the intelligence layer 2 and the execution layer 3 with the data of the virtual mirror platform in real time.

[0038] The synchronization module 124 serves two purposes: firstly, it ensures the synchronization of actual data in the physical scene with information data on the virtual mirror platform, guaranteeing real-time performance; secondly, it ensures the synchronization and real-time updates of data at each layer with the virtual mirror platform data, such as... Figure 4 As shown. The significance and function of the virtual mirror platform in this embodiment of the application is as follows: the strategies generated by the intelligent layer 2, as well as conflict detection and coordination strategies, etc., must be verified on the virtual mirror platform. After verification, they are finally sent to the execution layer 3 to be converted into control strategies for the actual physical devices.

[0039] For Smart Layer 2:

[0040] The decision-making agent 21 in the intelligent layer 2 receives task outputs from the perception agent 12, such as predictions or judgments, and dynamically generates global or local decisions, such as task allocation and path planning for robots. Simultaneously, the decision-making agent 21 optimizes its strategies through mechanisms such as trial-and-error rewards, achieving data-driven automatic optimization and autonomous evolution without human intervention to adapt to different workshop requirements, such as layout and equipment variations. Meanwhile, the coordination agent 22 monitors and evaluates the decisions made by the decision-making agent 21, for example, identifying conflicting or resource-constrained decisions and using rules or model algorithms to eliminate conflicts or coordinate resources. For example, an auction algorithm can be used to resolve conflicting decisions. In a sense, the coordination agent 22 essentially assumes the role of the workshop-level agent in a traditional architecture, and the intelligent layer 2 transmits the decision-making and optimization strategies to the execution layer 3.

[0041] In an optional embodiment, the intelligent layer 2 may further include a decision engine 23 and a federated learning module 24, as shown in the reference. Figure 2 and Figure 5 ,like Figure 5 As shown, the input to the decision engine 23 is the perception data output by the perception layer 1, the task queue, and other data. The decision engine 23 trains a strategy generation model based on the data output by the perception layer 1 through a trial-and-error reward learning mechanism (federated learning). The decision agent 21 uses the strategy generation model and dynamically generates global or local decision results based on the analysis results of the perception agent 12. The coordination agent 22 monitors and evaluates the decision results based on the perception data and label data, and adjusts conflicting strategies to generate conflict-free task instructions as the optimization strategy. The federated learning module 24 aggregates the overall parameters of the workshop based on the decision results and the optimization strategy to generate a workshop-level global optimization strategy.

[0042] Specifically, the decision engine 23 uses a reinforcement learning algorithm to dynamically generate a policy generation model that conforms to the global objective (such as efficiency) based on real-time data received from the perception layer 1 (such as device status, task queues, and other perception data). The decision agent 21 then uses the trained policy generation model to generate the optimal scheduling instructions or policies.

[0043] Based on the perception data output by perception agent 12 and the results of task intent analysis and prediction, coordination agent 22 checks the initial strategy generated by decision agent 21, verifies whether there are conflicts, whether optimization is needed, and adjusts the conflict through certain methods (such as game theory model algorithms), such as reallocating resources or adjusting task parameters, to generate conflict-free task instructions.

[0044] Coordination Agent 22 and Decision Agent 21 transmit the adjusted and optimized strategy to Federated Learning Module 24. Federated Learning Module 24 aggregates the overall parameters of the workshop to generate a workshop-level global optimization strategy. Its output can be transmitted to the intelligent layer 2 of other workshops (shared), and when necessary, other workshops or similar scenarios can directly adopt the optimization strategy.

[0045] In this embodiment, Decision Agent 21 is responsible for generating policies, while Coordination Agent 22 is responsible for coordinating conflicts. Decision Agent 21 needs to handle complex dynamic scenarios and perception data, and it typically deploys deep learning model algorithms (such as policy generation models) internally to ultimately generate the optimal scheduling policy. Its internal process generally follows: Perception Layer 1 state perception data → Preprocessing → Optimal policy generation → Instruction output. Coordination Agent 22 mainly resolves policy conflicts and needs to quickly generate adjustment methods. Therefore, it often does not deploy deep learning models but directly uses more efficient intelligent algorithms, such as heuristic optimization algorithms, to achieve faster computation speed and consume fewer computing resources. Its internal process generally follows: Policy conflict detection → Policy coordination calculation → Policy / constraint verification → Policy output.

[0046] Therefore, this application achieves real-time task and resource matching through the collaboration of decision-making agents and coordination agents, dynamically adjusting production strategies to adapt to unexpected situations such as order insertions and equipment failures. This improves overall efficiency, reduces communication latency in traditional centralized scheduling, and optimizes resource utilization. By aggregating multi-workshop strategies through a federated learning module, cross-workshop experience reuse is achieved, optimizing global production strategies (e.g., an efficient model from workshop A can be directly transferred to workshop B). Furthermore, each workshop agent continuously optimizes itself through shared strategies, enhancing its continuous evolution capabilities.

[0047] For execution layer 3:

[0048] Execution layer 3 internally deploys execution agent 31, which controls or adjusts device behavior based on the results of intelligence layer 2. Execution agent 31 enhances the device's intelligent decision-making capabilities and strengthens its intelligent control or behavior optimization capabilities based on its embedded lightweight model.

[0049] Specifically, Agent31 translates the optimized strategy from Intelligence Layer 2 into executable instructions. A lightweight model is deployed based on the edge computing module to achieve intelligent decision-making capabilities at the edge and device cluster levels, and to control the device cluster to execute tasks, such as... Figure 6 As shown in the diagram, Agent31 receives instructions from the intelligent layer Agent, parses them, and obtains the core parameters of the instructions. Then, it sends control signals to the corresponding physical devices through communication with the devices. The lightweight model acts as the intelligent brain of Agent31, primarily through online learning or pre-training of the parsed data and device status data, outputting adaptive decisions to optimize the control and operation of the devices.

[0050] For communication layer 4:

[0051] The communication layer 4 does not include an agent, enabling rapid synchronization between physical devices and digital mirrors via high-speed networks; and it integrates a multi-protocol model for broader compatibility and support for digital connections with heterogeneous devices.

[0052] Communication Layer 4 is the nerve center of this architecture, ensuring efficient and reliable information transmission between layers. Communication Layer 4 is a general term for various communication protocols, compatible with different types of protocols. It also includes a protocol conversion module used to convert different protocols into a format recognizable by Intelligent Layer 2. For example, it converts the perception data and status data from Perception Layer 1 into a unified ISON / XML format and transmits it to Intelligent Layer 2.

[0053] In addition, the optimization strategy generated by the intelligent layer 2 needs to be transmitted to the execution layer 3. In this embodiment, the optimization strategy generated by the intelligent layer 2 carries a priority. The communication layer 4 allocates bandwidth communication according to the priority and assigns communication priority to different strategies to avoid abnormal situations such as network congestion. For example, the priority of emergency queue jumping and order jumping tasks is higher than that of ordinary tasks, so dedicated bandwidth communication is allocated.

[0054] Therefore, in this embodiment, the perception agent 12 is responsible for data mapping and mirroring from the physical world to the digital world, and makes predictions based on the mirrored data. In the intelligent layer 2, the decision agent 21 is the decision-maker, and the coordination agent 22 is the coordination agent, jointly realizing the generation and optimization of AI-driven dynamic strategies. In the execution layer 3, the execution agent 31 is the execution agent, transforming digital strategies into physical actions. The communication layer 4 does not have a dedicated agent, but provides an interaction channel for all agents, ensuring high-speed information flow. Agents at different levels can collaborate hierarchically to achieve a leap from passively executing instructions to actively optimizing the overall system, ultimately building a self-organizing and adaptive workshop ecosystem.

[0055] The multi-Agent architecture of this application breaks through the inherent framework of traditional architectures, shifting from the traditional centralized task management and centralized task parsing model to a distributed decision-making, autonomous monitoring, and automatic evolutionary multi-Agent architecture. By empowering agents with AI-driven "perception-decision-evolution" closed-loop capabilities, it constructs a self-organizing, adaptive, and self-optimizing distributed intelligent ecosystem, completely breaking through the limitations of traditional layered architectures.

[0056] Another aspect of this application provides a multi-agent-based cross-workshop migration method, applied to the aforementioned workshop management system, such as... Figure 7 As shown, it includes the following steps:

[0057] In step 101, parameters and task interfaces are directly updated for identical devices, and parameter mapping tables are generated and constraints are adjusted based on the original device parameter library through dynamic semantic matching for heterogeneous devices.

[0058] In step 102, the point cloud map of the new workshop is imported to replace the original workshop scene data;

[0059] In step 103, the task parameters are adjusted according to the constraints of the new workshop, and the original workshop strategy set is migrated to the new workshop;

[0060] In step 104, a virtual mirror platform is established based on the perception data of the new workshop. This virtual mirror platform performs virtual simulation verification of the new workshop management system based on the input data, evaluating the platform's functionality, decision-making and coordination results, and the effectiveness of execution instructions to complete the new workshop deployment. Specifically, based on the input task data, the platform's functionality is evaluated; and based on the data from the virtual mirror platform, the scheduling strategy generated by the decision agent, the conflict resolution results of the coordination agent, and the instruction conversion logic of the execution agent are evaluated to determine whether they meet the preset goals.

[0061] This embodiment maps the equipment, scenarios, and strategies of the original workshop to the new workshop, automatically ensuring compatibility with heterogeneous equipment. It generates a parameter mapping table based on dynamic semantic mapping to achieve system migration, quickly adapting the existing multi-agent management system of the workshop to the new workshop, enabling cross-workshop experience reuse, and improving the efficiency of workshop management.

[0062] This application also provides a multi-agent-based cross-workshop migration system, such as... Figure 8 As shown, it includes:

[0063] Cross-workshop adaptive module: used to migrate the multi-agent management system of the original workshop to the new workshop; the cross-workshop adaptive module specifically includes:

[0064] Device mapping module: used to directly update device parameters and task interfaces for the same device; for heterogeneous devices, it generates a parameter mapping table based on the original device parameter library through dynamic semantic matching and adjusts the constraints.

[0065] Scene mapping module: Used to import the new workshop point cloud map and replace the original workshop scene map;

[0066] Task and strategy mapping module: used to adjust task parameters according to the constraints of the new workshop and migrate the original workshop strategy set to the new workshop;

[0067] The workshop mirroring module is used to establish a virtual mirroring platform based on the perception data of the new workshop. The virtual mirroring platform performs virtual simulation verification on the new workshop management system based on the input data, evaluates the functionality, decision-making and coordination results, and the effectiveness of execution instructions of the virtual mirroring platform, so as to complete the deployment of the new workshop.

[0068] In this embodiment, data such as new workshop equipment, scenarios, and workshop strategies are matched and mapped with the equipment mapping module, scenario mapping module, and task / strategy mapping module in the cross-workshop adaptive module. For identical equipment, equipment parameters and task interfaces can be directly updated. For different types of equipment, based on dynamic semantic mapping, a corresponding mapping relationship is established from the generated "parameter mapping table" according to the original equipment parameter library, and the corresponding constraints are adjusted.

[0069] For example: The original semantic description of a certain piece of equipment in the workshop was: AMR standard load capacity 300kg. The new semantic description of the equipment is: Manipulator-1 load capacity 500kg# lifting height 1.8m; Strategy adjustment: The operation model has been changed from the original "movement only" to "movement + palletizing", and it has been marked as re-debugging and recalibrating.

[0070] In addition, the scene data in this embodiment can be directly imported into the new scene point cloud map through the scene mapping module, replacing the original workshop map.

[0071] In this embodiment, after the basic module update and mirror platform mapping are established, the system needs to undergo virtual simulation and verification of the basic processes based on the new scenario data platform. Once the basic processes are normal, the prerequisites for actual deployment in the new workshop are met, and actual deployment and debugging can begin.

[0072] The virtual processes established in this system are related to the real management system as follows: Figure 9 As shown.

[0073] The simulation system requires data input. One method is to manually set and edit the data to generate sequence task data, which can then be used as input for the simulation system. Another method is to extract typical task instruction flow data from the original workshop operation database and convert it into input data for the simulation system through "typical operation and task data extraction / conversion."

[0074] The simulation is actually based on a mirror platform built from the perception data of the new workshop. Based on the input task data, it evaluates whether the mirror function is normal, and whether the results generated by Decision Agent 21 and Collaboration Agent based on the mirror platform and task data meet expectations or logic, and whether the overall system is operating normally. Execution Agent 31 evaluates whether it generates reasonable and normal execution instructions based on the data from Intelligent Layer 2, and modifies each component of the system or retrains the agents based on the evaluation results during the simulation (system simulation) operation.

[0075] In other words, the simulation input data in this application embodiment can be either existing typical data or actual operational data of the new workshop. When migrating the multi-agent management system from the original workshop to the new workshop, existing typical data can be input first, and the output results can be evaluated to determine whether the settings and mappings of the new workshop management system are correct or need correction. If the evaluation result is satisfactory, actual operational data of the new workshop can then be input to execute real workshop management.

[0076] The updating of the perception layer mirror platform based on data such as physical scene, equipment, and workshop strategy is the key foundation for the migration of the workshop management system. The execution of other layers is based on the data of the mirror platform.

[0077] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A multi-agent-based workshop management system, characterized in that, include: The layers are: perception layer, intelligence layer, execution layer, and communication layer. The perception layer includes a data acquisition module and a perception agent. The data acquisition module is used to collect perception data from equipment groups, robot groups, sensor groups, factory-level digital systems, and human-machine input in the physical scene. The perception agent generates a corresponding virtual mirror platform based on the perception data and outputs analysis results based on the virtual mirror platform. The analysis results include at least the prediction of equipment failure, the judgment of task delay risk, and the identification of intent to interrupt tasks. The intelligent layer includes a decision agent and a coordination agent. The decision agent dynamically generates global or local decision results based on the analysis results of the perception agent. The coordination agent monitors and evaluates the decision results through a preset mechanism and outputs optimization strategies. The intelligent layer transmits the decision results and optimization strategies to the execution layer; The execution layer includes an execution agent, which deploys a lightweight model through an edge computing module to parse the decision results and optimization strategies into device control signals to control device operation; The communication layer enables rapid synchronization between physical devices and digital mirrors via a high-speed network, providing an information interaction channel for the agents of the perception layer, the intelligence layer, and the execution layer.

2. The workshop management system according to claim 1, characterized in that, The perception agent includes: Data fusion and mirroring module: used to fuse the perceived data and generate a virtual mirroring platform for the perceived data based on scene map data; Intent understanding module: Based on the virtual image platform, it parses the perceived data and outputs tag data that the Agent can recognize; Prediction module: Performs prediction calculations based on the fused data and the label data, and outputs the analysis results.

3. The workshop management system according to claim 2, characterized in that, The data fusion and mirroring module includes: The data fusion module is used to fuse the perceived data into unified fused state data, eliminating data redundancy and conflicts; The virtual scene modeling module constructs a virtual scene that is one-to-one with the physical scene based on the fused state data, and generates a virtual mirror platform that reflects the actual operating status; wherein, the map data of the physical scene includes at least: point cloud map data generated based on point cloud registration or visual SLAM, device data, device motion data model, and boundary constraints and rules of the actual physical scene; The prediction module is also used to predict the state of the physical scene within a preset time based on the historical data and real-time online data of the virtual mirror platform, and output the analysis results.

4. The workshop management system according to claim 2, characterized in that, The perception agent also includes: The synchronization module is used to synchronize the perceived data with the data of the virtual mirror platform; and to synchronize and update the data of the perception layer, the intelligence layer and the execution layer with the data of the virtual mirror platform in real time.

5. The workshop management system according to claim 2, characterized in that, The intelligent layer also includes a decision engine and a federated learning module; The decision engine trains a strategy generation model based on the perception data output by the perception layer through a trial-and-error reward learning mechanism. The decision agent uses the strategy to generate the model and dynamically generates global or local decision results based on the analysis results of the perception agent. The coordinating agent monitors and evaluates the decision results based on the perception data and the tag data, and adjusts conflicting strategies to generate conflict-free task instructions as the optimization strategy. The federated learning module aggregates the overall parameters of the workshop based on the decision results and the optimization strategy to generate a workshop-level global optimization strategy.

6. The workshop management system according to claim 3, characterized in that, The method by which the data fusion module fuses the perceived data into unified fused state data includes, but is not limited to: Spatiotemporal alignment of data, feature matching and data source association, and deep fusion algorithms.

7. The workshop management system according to claim 1, characterized in that, The optimization strategy generated by the intelligent layer carries a priority, and the communication layer allocates bandwidth for communication according to the priority to avoid network congestion.

8. A multi-agent-based cross-workshop migration method, applied to a workshop management system as described in any one of claims 1-7, characterized in that, include: For identical devices, parameters and task interfaces are updated directly; for heterogeneous devices, a parameter mapping table is generated based on the original device parameter library through dynamic semantic matching, and constraints are adjusted. Import the point cloud map of the new workshop to replace the original workshop scene data; Adjust task parameters according to the constraints of the new workshop, and migrate the original workshop strategy set to the new workshop; A virtual mirror platform was established based on the sensor data from the new workshop; The virtual mirror platform performs virtual simulation verification of the new workshop management system based on the input data, evaluates the functionality, decision-making and coordination results, and effectiveness of the execution instructions of the virtual mirror platform, so as to complete the deployment of the new workshop.

9. The inter-workshop migration method according to claim 8, characterized in that, The virtual mirror platform performs virtual simulation verification of the new workshop management system based on the input data, evaluating the platform's functions, decision-making and coordination results, and the effectiveness of executed instructions, including: Based on the input task data, assess whether the virtual image platform functions normally; based on the data from the virtual image platform, assess whether the scheduling strategy generated by the decision agent, the conflict resolution results of the coordination agent, and the instruction conversion logic of the execution agent meet the preset goals.

10. A multi-agent-based cross-workshop migration system, applied to the cross-workshop migration method as described in claim 8 or 9, characterized in that, include: Cross-workshop adaptive module: used to migrate the original multi-agent management system of the workshop to the new workshop; The cross-workshop adaptive module specifically includes: Device mapping module: used to directly update device parameters and task interfaces for the same device; for heterogeneous devices, it generates a parameter mapping table based on the original device parameter library through dynamic semantic matching and adjusts the constraints. Scene mapping module: Used to import the new workshop point cloud map and replace the original workshop scene map; Task and strategy mapping module: used to adjust task parameters according to the constraints of the new workshop and migrate the original workshop strategy set to the new workshop; The workshop mirroring module is used to build a virtual mirroring platform based on the perception data of the new workshop. The virtual mirror platform performs virtual simulation verification of the new workshop management system based on the input data, evaluates the functionality, decision-making and coordination results, and effectiveness of the execution instructions of the virtual mirror platform, so as to complete the deployment of the new workshop.

Images