An Industrial Multi-Agent Collaborative System and Method for Complex Processes

Abstract

This invention relates to the field of intelligent agent control technology, specifically to an industrial multi-agent collaborative system and method for complex processes. It generates several agent dispatch strategies by combining a multi-agent knowledge base with the real-time standby status of agents; it generates a process flow feature library based on multi-source operating condition data from the industrial site, thereby delineating dynamic partitions of the process flow; it determines the current agent allocation strategy for each partition based on the agent dispatch strategies and the dynamic partitions of the process flow; it determines the conflict relationships between agents based on the operational status of the partitions, thereby identifying the scheduling characteristic tags of the agents; it compares the process flow execution vacancy characteristics of adjacent partitions with the scheduling characteristic tags of the agents in the partition, and reassigns some agents as shared collaborative agents. By balancing the dedicated and conditional scheduling of agents in the process flow, it fully utilizes the operational capacity of the agent cluster and efficiently meets the agent requirements of all process flows.

Description

Technical Field

[0001] This invention relates to the field of intelligent agent control technology, and in particular to an industrial multi-agent collaborative system and method for complex processes. Background Technology

[0002] Robots and other intelligent agents are widely used in industrial settings. These agents can be deployed across different processes in an industrial environment to manipulate products, improving production efficiency and yield. However, current applications of intelligent agents in industrial settings are limited to dedicated models, where each agent is responsible for only a single process and cannot cross over to other processes. This prevents cross-process coordination, potentially leading to idle capacity in some processes and underutilization of resources in others. Given the strong executability and ability of intelligent agents to perform various tasks, limiting them to single processes reduces the flexibility and scheduling freedom of multi-agent collaboration in industrial settings, failing to efficiently meet the demands of all processes. Summary of the Invention

[0003] Considering that existing multi-agent applications in industrial scenarios all adopt a dedicated mode, the agents can only be restricted to specific work processes and cannot be freely scheduled across different work processes. This leads to problems such as unbalanced energy allocation among agents and inflexible collaboration between agents across different work processes. In view of the above problems, this invention provides an industrial multi-agent collaborative system for complex processes, comprising: The knowledge base generation module is used to obtain the historical work records of the multi-agent cluster and generate a multi-agent knowledge base accordingly. The dispatch strategy generation module is used to generate several agent dispatch strategies based on the multi-agent knowledge base and the agent's standby status. The process feature generation module is used to generate a process feature library based on multi-source operating condition data from the industrial site. The partitioning module is used to define dynamic partitions of the process flow based on the process flow feature library. The allocation strategy generation module is used to determine the current agent allocation strategy for each partition based on the several agent dispatch strategies and the dynamic partitioning of the process flow. The scheduling label determination module is used to determine the conflict relationship between agents based on the actual operation of the partition, and thereby identify the scheduling characteristic label of the agent. The modification module is used to compare the process flow execution gap characteristics of the partition with the scheduling characteristic labels of the agents in the partition, and modify a portion of the agents to become shared collaborative agents.

[0004] Optionally, the knowledge base generation module is used to obtain the historical work records of the multi-agent cluster to generate a multi-agent knowledge base, including: Extract the intrinsic and extrinsic working attributes of each agent from the historical work records of the multi-agent cluster; wherein the intrinsic working attributes include the working procedures set by the agent; and the extrinsic working attributes include the types of external behaviors and actions of the agent during its historical work. Semantic recognition is performed on the intrinsic and extrinsic working attributes to generate intrinsic and extrinsic working labels for each agent, thereby forming a multi-agent knowledge base. The dispatch strategy generation module is used to generate several agent dispatch strategies based on the multi-agent knowledge base and the agents' standby status, including: Based on the real-time charging level and real-time location of each agent, a real-time standby state representation vector is generated for the agent; based on the multi-agent knowledge base and the real-time standby state representation vector, several agent dispatch strategies are generated; wherein the agent dispatch strategy includes the industrial operation scenario attributes matched by the agent.

[0005] Optionally, the process feature generation module is used to generate a process feature library based on multi-source operating condition data from the industrial site, including: Machine vision recognition and hardware equipment inspection are performed on the industrial site to obtain multi-source operating condition data of the industrial site; wherein the multi-source operating condition data includes the structure and motion data of the objects and hardware equipment respectively. The multi-source working condition data is subjected to time-domain change identification to obtain the structural change characteristics and motion change characteristics of objects and hardware equipment during each process flow, thereby generating the process flow feature library of the industrial site. The partitioning module is used to define dynamic partitions of the process flow based on the process flow feature library, including: Calculate the structural change feature similarity and motion change feature similarity among all process flows in the process flow feature library, thereby dividing the industrial site into several process flow dynamic partitions; wherein each process flow dynamic partition corresponds to multiple process flows that meet preset similarity conditions.

[0006] Optionally, the allocation strategy generation module is used to determine the current agent allocation strategy for each partition based on the plurality of agent dispatch strategies and the dynamic partitioning of the process flow, including: By comparing the aforementioned agent dispatch strategies with all dynamic partitions of process flows, the completion efficiency of each agent dispatch strategy for all process flows under each dynamic partition of process flows is obtained; based on the completion efficiency, the current agent allocation strategy for each partition is determined; wherein the agent allocation strategy includes the correspondence between each partition and the agent currently allocated to execute its own subordinate process flows. The scheduling label determination module is used to determine the conflict relationships between agents based on the actual operation of the partition, thereby identifying the scheduling characteristic labels of the agents, including: Visual recognition identifies the operational status of all assigned agents within each partition, obtaining the execution status of all assigned agents within the corresponding process flow, thereby determining the conflict relationships among all assigned agents; based on the conflict relationships, the schedulable time characteristics and schedulable work function characteristics of each assigned agent are determined, thereby identifying the agent's scheduling characteristic label.

[0007] Optionally, the modification module is used to compare the process flow execution gap features of adjacent partitions of the partition with the scheduling characteristic labels of the agents in the partition, and modify a portion of the agents as shared collaborative agents, including: Visually identify omissions and / or errors in the process flow execution of adjacent partitions of the partition, thereby determining the missing characteristics of the process flow execution of the other partitions; By comparing the process flow execution gap characteristics with the scheduling characteristic tags of the agents in the partition, a portion of the agents in the partition are changed into shared collaborative agents; wherein the shared collaborative agents are configured to be able to move freely and switch between the partition and the other partitions to execute different process flows.

[0008] As one aspect of the present invention, embodiments of the present invention also provide an industrial multi-agent collaboration method for complex processes, comprising: Step S1: Obtain the historical work records of the multi-agent cluster to generate a multi-agent knowledge base; generate several agent dispatch strategies based on the multi-agent knowledge base and the standby status of the agents. Step S2: Generate a process flow feature library based on multi-source operating condition data from the industrial site; delineate dynamic partitions of the process flow based on the process flow feature library; Step S3: Based on the agent dispatch strategy and the dynamic partitioning of the process flow, determine the current agent allocation strategy for each partition; based on the actual operation of the partition, determine the conflict relationship between agents, thereby identifying the scheduling characteristic label of the agent; Step S4: Compare the process flow execution gap features of the adjacent partitions and the scheduling characteristic labels of the agents in the partition, and change some agents to shared collaborative agents.

[0009] Optionally, in step S1, the historical work records of the multi-agent cluster are obtained to generate a multi-agent knowledge base; based on the multi-agent knowledge base and the standby status of the agents, several agent dispatch strategies are generated, including: Extract the intrinsic and extrinsic working attributes of each agent from the historical work records of the multi-agent cluster; wherein the intrinsic working attributes include the working procedures set by the agent; and the extrinsic working attributes include the types of external behaviors and actions of the agent during its historical work. Semantic recognition is performed on the intrinsic and extrinsic working attributes to generate intrinsic and extrinsic working labels for each agent, thereby forming a multi-agent knowledge base. Based on the real-time charging level and real-time location of each agent, a real-time standby state representation vector is generated for the agent; based on the multi-agent knowledge base and the real-time standby state representation vector, several agent dispatch strategies are generated; wherein the agent dispatch strategy includes the industrial operation scenario attributes matched by the agent.

[0010] Optionally, in step S2, a process flow feature library is generated based on multi-source operating condition data from the industrial site; and dynamic partitioning of the process flow is defined based on the process flow feature library, including: Machine vision recognition and hardware equipment inspection are performed on the industrial site to obtain multi-source operating condition data of the industrial site; wherein the multi-source operating condition data includes the structure and motion data of the objects and hardware equipment respectively. The multi-source working condition data is subjected to time-domain change identification to obtain the structural change characteristics and motion change characteristics of objects and hardware equipment during each process flow, thereby generating the process flow feature library of the industrial site. Calculate the structural change feature similarity and motion change feature similarity among all process flows in the process flow feature library, thereby dividing the industrial site into several process flow dynamic partitions; wherein each process flow dynamic partition corresponds to multiple process flows that meet preset similarity conditions.

[0011] Optionally, in step S3, based on the agent dispatch strategy and the dynamic partitioning of the process flow, the current agent allocation strategy for each partition is determined; based on the operational status of the partition, the conflict relationships between agents are determined, thereby identifying the scheduling characteristic tags of the agents, including: By comparing the aforementioned agent dispatch strategies with all dynamic partitions of process flows, the completion efficiency of each agent dispatch strategy for all process flows under each dynamic partition of process flows is obtained; based on the completion efficiency, the current agent allocation strategy for each partition is determined; wherein the agent allocation strategy includes the correspondence between each partition and the agent currently allocated to execute its own subordinate process flows. Visual recognition identifies the operational status of all assigned agents within each partition, obtaining the execution status of all assigned agents within the corresponding process flow, thereby determining the conflict relationships among all assigned agents; based on the conflict relationships, the schedulable time characteristics and schedulable work function characteristics of each assigned agent are determined, thereby identifying the agent's scheduling characteristic label.

[0012] Optionally, in step S4, by comparing the process flow execution vacancy features of adjacent partitions and the scheduling characteristic labels of the agents in the partition, a portion of the agents are changed to shared collaborative agents, including: Visually identify omissions and / or errors in the process flow execution of adjacent partitions of the partition, thereby determining the missing characteristics of the process flow execution of the other partitions; By comparing the process flow execution gap characteristics with the scheduling characteristic tags of the agents in the partition, a portion of the agents in the partition are changed into shared collaborative agents; wherein the shared collaborative agents are configured to be able to move freely and switch between the partition and the other partitions to execute different process flows.

[0013] The beneficial effects of the above-mentioned technical solutions provided in the embodiments of the present invention include at least the following: This invention provides an industrial multi-agent collaboration system and method for complex processes. It acquires historical work records of a multi-agent cluster to generate a multi-agent knowledge base, and generates several agent dispatch strategies based on the agents' standby status. A process flow feature library is generated based on multi-source operational data from the industrial site. Dynamic partitions of the process flow are defined based on this feature library. The current agent allocation strategy for each partition is determined based on the agent dispatch strategies and the dynamic partitions. Conflict relationships between agents are determined based on the operational status of each partition, thus identifying the agents' scheduling characteristic tags. By comparing the process flow execution vacancy characteristics of adjacent partitions with the scheduling characteristic tags of the agents in the partition, some agents are reassigned as shared collaborative agents. By balancing dedicated and conditional scheduling of agents in the process flow, the system fully utilizes the capacity of the agent cluster to efficiently meet the agent requirements of all process flows.

[0014] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the structure of an industrial multi-agent collaborative system for complex processes provided in an embodiment of the present invention.

[0017] Figure 2 This is a flowchart illustrating an industrial multi-agent collaboration method for complex processes provided in an embodiment of the present invention. Detailed Implementation

[0018] The following disclosure provides many different embodiments or examples for implementing various embodiments of the invention. To simplify the disclosure, specific embodiments are described below. Of course, these are merely examples and are not intended to limit the scope of the invention.

[0019] The embodiments of the invention will now be described in detail with reference to the accompanying drawings.

[0020] Please see Figure 1 As shown, one embodiment of this application provides an industrial multi-agent collaborative system for complex processes. This industrial multi-agent collaborative system for complex processes includes: The knowledge base generation module is used to obtain the historical work records of the multi-agent cluster and generate a multi-agent knowledge base accordingly. The dispatch strategy generation module is used to generate several agent dispatch strategies based on the multi-agent knowledge base and the agent's standby status. The process feature generation module is used to generate a process feature library based on multi-source operating condition data from the industrial site. The partitioning module is used to dynamically partition the process flow based on the process flow feature library; The allocation strategy generation module is used to determine the current agent allocation strategy for each partition based on several agent dispatch strategies and dynamic partitioning of the process flow. The scheduling label determination module is used to determine the conflict relationship between agents based on the actual operation of the partition, and thereby identify the scheduling characteristic label of the agent. The modification module is used to compare the process flow execution gap features of adjacent partitions and the scheduling characteristic labels of the agents in the partition, and modify some agents to become shared collaborative agents.

[0021] The beneficial effects of the above embodiments are that the industrial multi-agent collaborative system for complex processes fully utilizes the operational capacity of the agent cluster by balancing the dedicated and conditional scheduling of agents in the process flow, and efficiently meets the needs of all process flows for agents.

[0022] In another embodiment, the knowledge base generation module is used to obtain the historical work records of the multi-agent cluster to generate a multi-agent knowledge base, including: Extract the intrinsic and extrinsic working attributes of each agent from the historical work records of the multi-agent cluster; the intrinsic working attributes include the working procedures set by the agent; the extrinsic working attributes include the types of external behaviors and actions of the agent during its historical work. Semantic recognition is performed on intrinsic and extrinsic job attributes to generate intrinsic and extrinsic job tags for each agent, thereby forming a multi-agent knowledge base. The dispatch strategy generation module is used to generate several agent dispatch strategies based on the multi-agent knowledge base and the agents' standby status, including: Based on the real-time charging level and real-time location of each agent, a real-time standby state representation vector is generated for the agent; based on the multi-agent knowledge base and the real-time standby state representation vector, several agent dispatch strategies are generated; the agent dispatch strategies include the industrial operation scenario attributes matched by the agent.

[0023] In industrial settings such as workshops or assembly lines, multi-agent clusters are deployed. Each agent within the cluster can independently execute tasks and perform different types of actions. Understandably, each agent in a specific industrial setting can cooperate with production equipment to complete different tasks and actions. Furthermore, the tasks and actions performed by the agents differ depending on the type of equipment. For example, agents can smoothly complete tasks and actions without lag for some types of equipment, while for others, they may fail to complete the tasks or actions completely or experience lag. In addition, different agents have different built-in operating programs, resulting in varying task and action adaptation performance within the industrial setting. As the above analysis shows, the industrial tasks and actions suitable for each agent within a multi-agent cluster differ. To ensure that all processes in a real-world industrial setting can be configured with appropriate agents, it is necessary to characterize the performance of each agent in multiple aspects. Specifically, the built-in work program types and content of each agent are extracted from the historical work records of the agent cluster, as well as the types of external behaviors performed by each agent during the historical work period (such as the trajectory, amplitude, and frequency of the performed behaviors). The intrinsic and extrinsic work attributes of an agent directly affect the tasks and actions it can perform. Therefore, semantic recognition is performed on the intrinsic and extrinsic work attributes to generate intrinsic and extrinsic work tags for each agent. These tags characterize the agent's work performance at the levels of internal programs and external behaviors, respectively. Then, using the agent's ID information as a benchmark, the intrinsic and extrinsic work tags of each agent are matched to form a multi-agent knowledge base. This effectively and identifiably represents the work performance of all agents under the multi-agent cluster, providing a basis for subsequent matching of agents to industrial operation scenarios.

[0024] Furthermore, all agents within the agent cluster are in a standby state, charging at their respective locations when not in use. Understandably, each agent's standby state depends on its location and charge level. Generally, the closer an agent is to a specific process flow and the greater its charge level, the higher its priority in being assigned to that process flow. Therefore, an agent's standby state influences its allocation to process flows. To address this, a real-time standby state representation vector is generated for each agent based on its real-time charge level and location. This vector contains elements corresponding to the real-time charge level and location. Then, based on the multi-agent knowledge base and the real-time standby state representation vector for each agent, several agent dispatch strategies are generated to represent the industrial work scenario attributes matched to each agent. These industrial work scenario attributes may include, but are not limited to, the operational content that the matched process flow requires for the product and the construction and production action information of the corresponding production hardware equipment.

[0025] In another embodiment, the process feature generation module is used to generate a process feature library based on multi-source operating condition data from the industrial site, including: Machine vision recognition and hardware equipment inspection are performed on industrial sites to obtain multi-source operating condition data of the industrial site; the multi-source operating condition data includes the structure and motion data of the objects and hardware equipment respectively. By performing time-domain change identification on multi-source operating data, the structural change characteristics and motion change characteristics of objects and hardware equipment during each process are obtained, thereby generating a process feature library of the industrial site. The partitioning module is used to dynamically partition the process flow based on the process flow feature library, including: The similarity of structural change features and motion change features among all process flows in the process flow feature library is calculated to divide the industrial site into several process flow dynamic partitions; each process flow dynamic partition corresponds to multiple process flows that meet the preset similarity conditions.

[0026] Industrial sites, such as workshops or assembly lines, include production hardware and objects to be processed (e.g., raw materials and / or semi-finished products). Under normal operating conditions, both the hardware and objects are in motion, and each possesses different structures (e.g., shape and size). Considering that the entire production process in an industrial setting involves several steps, and that the hardware and objects exhibit different motion and structural states in different steps, the tasks and / or actions required by the intelligent agent vary across these steps. To ensure that each step receives a suitable intelligent agent configuration, it is necessary to pre-distinguish the steps within the industrial site and define the characteristic states of each step throughout the entire production process. Specifically, this is achieved through machine vision recognition and hardware detection in the industrial site, obtaining the structural and motion data of both objects and hardware. This allows for the simultaneous characterization of objects and hardware at both the structural and motion levels, forming multi-source operational data for the industrial site. Then, time change identification is performed on the above multi-source working condition data to obtain the structural change characteristics (such as changes in the shape / size of the object and / or changes in the spatial position of the production components of the hardware) and motion change characteristics (such as changes in the speed / direction of the object and the hardware) of each process flow under the entire production process. This generates a process flow feature library of the industrial site, which includes the structural change characteristics and motion change characteristics of each process flow for the object and hardware.

[0027] In industrial production processes, the sub-processes are not entirely unrelated or significantly different. Understandably, some sub-processes share significant similarities in the structural and motion changes of objects and hardware equipment, while others exhibit less similarity. For sub-processes with significant similarities, configuring the same intelligent agent improves the uniformity of agent configuration across processes, preventing agents from being scattered across different processes and hindering centralized control. Specifically, the similarity of structural and motion change features among all sub-processes in the process feature library is first assessed. This divides the industrial site into several dynamic process partitions. Each dynamic partition covers multiple sub-processes, where the similarity of structural and motion change features exceeds a first similarity threshold and the similarity exceeds a second similarity threshold. This ensures that all sub-processes within each dynamic partition can be completed using the same intelligent agent.

[0028] In another embodiment, the allocation strategy generation module is used to determine the current agent allocation strategy for each partition based on several agent dispatch strategies and dynamic partitioning of the process flow, including: By comparing several agent dispatch strategies and all process flow dynamic partitions, the completion efficiency of each agent dispatch strategy for all process flows under each process flow dynamic partition is obtained; based on the completion efficiency, the current agent allocation strategy for each partition is determined; wherein the agent allocation strategy includes the correspondence between each partition and the agent currently allocated to execute its own process flows. The scheduling label determination module is used to determine the conflict relationships between agents based on the actual operation of the partition, and thereby identify the scheduling characteristic labels of the agents, including: Visual recognition identifies the operational status of all assigned agents within each partition, obtaining the execution status of all assigned agents within the corresponding process flow, thereby determining the conflict relationships among all assigned agents; based on the conflict relationships, the schedulable time characteristics and schedulable work function characteristics of each assigned agent are determined, thereby identifying the scheduling characteristic labels of the agents.

[0029] Different agents have different performance characteristics, resulting in different efficiency in completing the same process (e.g., the number of objects processed per unit time). In order to balance the matching of agents and processes in industrial operation scenarios and the ability of agents to handle processes with high efficiency, several agent dispatch strategies and all process dynamic partitions are compared to obtain the completion efficiency of each agent dispatch strategy for all processes under each process dynamic partition. This determines the current agent allocation strategy for each partition and comprehensively represents the correspondence between each partition and the agents currently assigned to execute all processes under the partition itself. In this way, the corresponding number and type of agents are subsequently allocated to each partition based on the above agent allocation strategy.

[0030] Furthermore, each process flow is dynamically partitioned, covering multiple process flows, and each dynamic partition is configured with multiple agents. Among these agents, at least two agents may perform the same task operation within the same process flow. This can lead to work conflicts between these agents within the same process flow. If these conflicts persist, not only will the process flow malfunction, but different agents will also interfere with each other, reducing the overall operational efficiency of the dynamic partition. Therefore, we first visually identify the operational status of all agents assigned within each partition, obtaining the execution status of all assigned agents within the corresponding process flow. This allows us to determine whether at least two agents are performing the same or similar actions within the same process flow. If so, we determine that at least two agents have a conflict relationship; otherwise, we determine that at least two agents do not have a conflict relationship. Next, identify at least two agents with conflicting relationships, determine the schedulable time characteristics (i.e., the time distribution characteristics of each agent in an idle state) and schedulable work function characteristics (such as the work tasks and behavioral action content that each agent can perform), and use these to identify the schedulable characteristic tags of the corresponding agents, thus identifying and distinguishing agents at the time and function levels, and providing a basis for setting up the collaboration and sharing of agents in the future.

[0031] In another embodiment, the modification module is used to compare the process flow execution gap characteristics of adjacent partitions and the scheduling characteristic labels of the agents in the partition, and modify a portion of the agents as shared collaborative agents, including: Visually identify the omissions and / or errors in the process flow execution of adjacent partitions in the partition, thereby determining the missing features in the process flow execution of other partitions; By comparing the vacancy characteristics of the process flow execution with the scheduling characteristic labels of the agents in the partition, a portion of the agents in the partition are changed into shared collaborative agents; the shared collaborative agents are configured to be able to move freely between the partition and other partitions and switch to execute different process flows.

[0032] For at least two agents with conflicting relationships within a dynamic partition of a workflow, to ensure that the corresponding workflow can be completed normally and efficiently with the assistance of one agent and to avoid interference from conflicting agents, it is only necessary to retain one agent among all conflicting agents to serve the current partition, while the other agents remain idle in the current partition. Without collaborative scheduling of the other agents, their operational capacity would be wasted. Therefore, we first visually identify the omissions and / or errors in the workflow execution of adjacent partitions to determine the workflow execution gap characteristics of other partitions. These gap characteristics can be, but are not limited to, workflows in adjacent partitions that require additional agent assistance. Then, we compare the workflow execution gap characteristics with the scheduling characteristic tags of the agents in the partition to change some agents within the partition to shared collaborative agents. These changed agents can freely move and switch between partitions to execute different workflows, preventing agents from being idle in their original partitions and ensuring flexible collaboration and free scheduling of agents in other partitions, thus meeting the needs of different partitions for agents.

[0033] Please see Figure 2 As shown, an embodiment of this application provides an industrial multi-agent collaboration method for complex processes. This industrial multi-agent collaboration method for complex processes includes: Step S1: Obtain the historical work records of the multi-agent cluster to generate a multi-agent knowledge base; based on the multi-agent knowledge base and the standby status of the agents, generate several agent dispatch strategies. Step S2: Generate a process flow feature library based on multi-source operating condition data from the industrial site; delineate dynamic partitions of the process flow based on the process flow feature library; Step S3: Based on several agent dispatch strategies and dynamic partitioning of process flows, determine the current agent allocation strategy for each partition; based on the actual operation of the partition, determine the conflict relationship between agents, thereby identifying the scheduling characteristic label of the agents. Step S4: Compare the process flow execution vacancy features of adjacent partitions and the scheduling characteristic labels of the agents in the partition, and change some agents to shared collaborative agents.

[0034] This collaborative approach balances the dedicated and conditional scheduling of agents in the process flow, fully utilizes the operational capacity of the agent cluster, and efficiently meets the needs of all processes for agents.

[0035] In another embodiment, in step S1, the historical work records of the multi-agent cluster are obtained to generate a multi-agent knowledge base; based on the multi-agent knowledge base and the standby status of the agents, several agent dispatch strategies are generated, including: Extract the intrinsic and extrinsic working attributes of each agent from the historical work records of the multi-agent cluster; the intrinsic working attributes include the working procedures set by the agent; the extrinsic working attributes include the types of external behaviors and actions of the agent during its historical work. Semantic recognition is performed on intrinsic and extrinsic job attributes to generate intrinsic and extrinsic job tags for each agent, thereby forming a multi-agent knowledge base. Based on the real-time charging level and real-time location of each agent, a real-time standby state representation vector is generated for the agent; based on the multi-agent knowledge base and the real-time standby state representation vector, several agent dispatch strategies are generated; the agent dispatch strategies include the industrial operation scenario attributes matched by the agent.

[0036] In another embodiment, in step S2, a process flow feature library is generated based on multi-source operating condition data from the industrial site; based on the process flow feature library, dynamic partitions of the process flow are defined, including: Machine vision recognition and hardware equipment inspection are performed on industrial sites to obtain multi-source operating condition data of the industrial site; the multi-source operating condition data includes the structure and motion data of the objects and hardware equipment respectively. By performing time-domain change identification on multi-source operating data, the structural change characteristics and motion change characteristics of objects and hardware equipment during each process are obtained, thereby generating a process feature library of the industrial site. The industrial site is divided into several dynamic partitions based on the similarity of structural change features and motion change features among all process flows in the process flow feature library; each dynamic partition corresponds to multiple process flows that meet the preset similarity conditions.

[0037] In another embodiment, in step S3, based on several agent dispatch strategies and dynamic partitioning of the process flow, the current agent allocation strategy for each partition is determined; based on the actual operation of the partition, the conflict relationships between agents are determined, thereby identifying the scheduling characteristic tags of the agents, including: By comparing several agent dispatch strategies and all dynamic partitions of process flows, the completion efficiency of each agent dispatch strategy for all process flows under each dynamic partition of process flow is obtained; based on the completion efficiency, the current agent allocation strategy for each partition is determined; wherein the agent allocation strategy includes the correspondence between each partition and the agent currently allocated to execute its own process flow. Visual recognition identifies the operational status of all assigned agents within each partition, obtaining the execution status of all assigned agents within the corresponding process flow, thereby determining the conflict relationships among all assigned agents; based on the conflict relationships, the schedulable time characteristics and schedulable work function characteristics of each assigned agent are determined, thereby identifying the scheduling characteristic labels of the agents.

[0038] In another embodiment, in step S4, the process flow execution gap features of adjacent partitions are compared with the scheduling characteristic labels of the agents in the partition, and a portion of the agents are changed to shared collaborative agents, including: Visually identify the omissions and / or errors in the process flow execution of adjacent partitions in the partition, thereby determining the missing features in the process flow execution of other partitions; By comparing the vacancy characteristics of the process flow execution with the scheduling characteristic labels of the agents in the partition, a portion of the agents in the partition are changed into shared collaborative agents; the shared collaborative agents are configured to be able to move freely between the partition and other partitions and switch to execute different process flows.

[0039] The industrial multi-agent collaboration method for complex processes of the present invention corresponds to and has the same operation and effect as the aforementioned industrial multi-agent collaboration system for complex processes. Therefore, the description of the industrial multi-agent collaboration method for complex processes will not be repeated here.

[0040] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. This disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims. Thus, if these modifications and variations of the invention fall within the scope of the claims of the invention and their equivalents, the invention is also intended to include these modifications and variations.

Claims

1. An industrial multi-agent collaborative system for complex processes, characterized in that, include: The knowledge base generation module is used to obtain the historical work records of a multi-agent cluster and generate a multi-agent knowledge base, including: Extract the intrinsic and extrinsic working attributes of each agent from the historical work records of the multi-agent cluster; wherein the intrinsic working attributes include the working procedures set by the agent; and the extrinsic working attributes include the types of external behaviors and actions of the agent during its historical work. Semantic recognition is performed on the intrinsic and extrinsic working attributes to generate intrinsic and extrinsic working labels for each agent, thereby forming a multi-agent knowledge base. The dispatch strategy generation module is used to generate several agent dispatch strategies based on the multi-agent knowledge base and the agents' standby status, including: Based on the real-time charging level and real-time location of each agent, a real-time standby state representation vector is generated for the agent; based on the multi-agent knowledge base and the real-time standby state representation vector, several agent dispatch strategies are generated; wherein the agent dispatch strategy includes the industrial operation scenario attributes matched by the agent; The process feature generation module is used to generate a process feature library based on multi-source operating condition data from the industrial site, including: Machine vision recognition and hardware equipment inspection are performed on the industrial site to obtain multi-source operating condition data of the industrial site; wherein the multi-source operating condition data includes the structure and motion data of the objects and hardware equipment respectively. The multi-source working condition data is subjected to time-domain change identification to obtain the structural change characteristics and motion change characteristics of objects and hardware equipment during each process flow, thereby generating the process flow feature library of the industrial site. The partitioning module is used to define dynamic partitions of the process flow based on the process flow feature library, including: Calculate the structural change feature similarity and motion change feature similarity among all process flows in the process flow feature library, thereby dividing the industrial site into several process flow dynamic partitions; wherein each process flow dynamic partition corresponds to multiple process flows that meet preset similarity conditions; The allocation strategy generation module is used to determine the current agent allocation strategy for each partition based on the plurality of agent dispatch strategies and the dynamic partitioning of the process flow, including: By comparing the aforementioned agent dispatch strategies with all dynamic partitions of process flows, the completion efficiency of each agent dispatch strategy for all process flows under each dynamic partition of process flows is obtained; based on the completion efficiency, the current agent allocation strategy for each partition is determined; wherein the agent allocation strategy includes the correspondence between each partition and the agent currently allocated to execute its own subordinate process flows. The scheduling label determination module is used to determine the conflict relationships between agents based on the actual operation of the partition, thereby identifying the scheduling characteristic labels of the agents, including: Visual recognition identifies the operational status of all assigned agents within each partition, obtaining the execution status of all assigned agents within the corresponding process flow, thereby determining the conflict relationships among all assigned agents; wherein the conflict relationship refers to at least two agents among all assigned agents performing the same or similar actions in the same process flow; based on the conflict relationship, the schedulable time characteristics and schedulable work function characteristics of each assigned agent are determined, thereby identifying the agent's scheduling characteristic label; The modification module is used to compare the process flow execution gap characteristics of adjacent partitions with the scheduling characteristic tags of the agents in the partition, and modify a portion of the agents to become shared collaborative agents, including: Visually identify omissions and / or errors in the execution of processes in adjacent partitions of the partition, thereby determining the process execution gap characteristics of the other partitions; wherein the process execution gap characteristics refer to the process flows in adjacent partitions that require the addition of intelligent agent-assisted services. By comparing the process flow execution gap characteristics with the scheduling characteristic tags of the agents in the partition, a portion of the agents in the partition are changed into shared collaborative agents; wherein the shared collaborative agents are configured to be able to move freely and switch between the partition and the other partitions to execute different process flows.

2. A multi-agent collaborative method for complex industrial processes, characterized in that, include: Step S1: Obtain the historical work records of the multi-agent cluster to generate a multi-agent knowledge base; Based on the multi-agent knowledge base and the agents' standby status, several agent dispatch strategies are generated, including: Extract the intrinsic and extrinsic working attributes of each agent from the historical work records of the multi-agent cluster; wherein the intrinsic working attributes include the working procedures set by the agent; and the extrinsic working attributes include the types of external behaviors and actions of the agent during its historical work. Semantic recognition is performed on the intrinsic and extrinsic working attributes to generate intrinsic and extrinsic working labels for each agent, thereby forming a multi-agent knowledge base. Based on the real-time charging level and real-time location of each agent, a real-time standby state representation vector is generated for the agent; based on the multi-agent knowledge base and the real-time standby state representation vector, several agent dispatch strategies are generated; wherein the agent dispatch strategy includes the industrial operation scenario attributes matched by the agent; Step S2: Generate a process flow feature library based on multi-source operating condition data from the industrial site; delineate dynamic partitions of the process flow based on the process flow feature library, including: Machine vision recognition and hardware equipment inspection are performed on the industrial site to obtain multi-source operating condition data of the industrial site; wherein the multi-source operating condition data includes the structure and motion data of the objects and hardware equipment respectively. The multi-source working condition data is subjected to time-domain change identification to obtain the structural change characteristics and motion change characteristics of objects and hardware equipment during each process flow, thereby generating the process flow feature library of the industrial site. Calculate the structural change feature similarity and motion change feature similarity among all process flows in the process flow feature library, thereby dividing the industrial site into several process flow dynamic partitions; wherein each process flow dynamic partition corresponds to multiple process flows that meet preset similarity conditions; Step S3: Based on the agent dispatch strategies and the dynamic partitioning of the process flow, determine the current agent allocation strategy for each partition; based on the operational status of the partition, determine the conflict relationships between agents, thereby identifying the scheduling characteristic tags of the agents, including: By comparing the aforementioned agent dispatch strategies with all dynamic partitions of process flows, the completion efficiency of each agent dispatch strategy for all process flows under each dynamic partition of process flows is obtained; based on the completion efficiency, the current agent allocation strategy for each partition is determined; wherein the agent allocation strategy includes the correspondence between each partition and the agent currently allocated to execute its own subordinate process flows. Visual recognition identifies the operational status of all assigned agents within each partition, obtaining the execution status of all assigned agents within the corresponding process flow, thereby determining the conflict relationships among all assigned agents; wherein the conflict relationship refers to at least two agents among all assigned agents performing the same or similar actions in the same process flow; based on the conflict relationship, the schedulable time characteristics and schedulable work function characteristics of each assigned agent are determined, thereby identifying the agent's scheduling characteristic label; Step S4: Compare the process flow execution gap features of adjacent partitions with the scheduling characteristic tags of the agents in the partition, and change a portion of the agents as shared collaborative agents, including: Visually identify omissions and / or errors in the execution of processes in adjacent partitions of the partition, thereby determining the process execution gap characteristics of the other partitions; wherein the process execution gap characteristics refer to the process flows in adjacent partitions that require the addition of intelligent agent-assisted services. By comparing the process flow execution gap characteristics with the scheduling characteristic tags of the agents in the partition, a portion of the agents in the partition are changed into shared collaborative agents; wherein the shared collaborative agents are configured to be able to move freely and switch between the partition and the other partitions to execute different process flows.

Images