A hierarchical architecture industrial multi-agent collaborative scheduling method and system

Abstract

The application relates to the technical field of multi-agent cooperative scheduling, in particular to a layered architecture industrial multi-agent cooperative scheduling method and system, which comprises the following steps: splitting a demand task into multiple subtasks with time sequence constraint relations; matching special agents for executing the subtasks in an agent candidate pool, and generating an initial execution sequence; importing the initial execution sequence into a digital twin environment to obtain task boundary data and runtime sequence data of each special agent; judging whether there is spatial interference based on the task boundary data and the runtime sequence data; when there is spatial interference, adjusting the execution order and / or the work range of the special agent; generating a cooperative execution sequence according to the adjusted execution order and / or the work range, and delivering the cooperative execution sequence to the corresponding special agent for execution. The application has the effects of eliminating the risk of systematic deadlock and beat cascade stagnation when multiple agents have high concurrency work, and realizing efficient and flexible scheduling of heterogeneous devices.

Description

Technical Field

[0001] This application relates to the field of multi-agent cooperative scheduling technology, specifically to a hierarchical architecture industrial multi-agent cooperative scheduling method and system. Background Technology

[0002] With the evolution of intelligent manufacturing and flexible production, industrial sites are increasingly reliant on multi-agent systems composed of a large number of heterogeneous devices. In this complex and dynamic environment with high spatiotemporal coupling, the coordinated scheduling of multiple agents is a core element in ensuring overall production capacity, process continuity, and physical safety.

[0003] However, existing collaborative scheduling technologies still have significant limitations. First, at the system architecture and task allocation level, existing solutions mostly adopt centralized, rigid assignment logic. As the scale of equipment expands, not only does the computing power of the central node easily face bottlenecks, but it is also difficult to perform fine-grained matching based on the dynamic capabilities of each agent. Due to the lack of effective dimensionality reduction and feature mapping of macro-level task-process correlations, resource mismatch often occurs, with high-capacity equipment executing low-optimal tasks, or workstations running idle due to improper timing connections.

[0004] More importantly, at the physical execution path planning level, existing collaborative scheduling methods generally lack the ability to perform upfront three-dimensional spatiotemporal extrapolation, and their conflict avoidance mainly relies on the passive physical sensing response at the device's underlying level. However, this lagging conflict resolution strategy is prone to causing systemic deadlocks in intensive operation scenarios; at the same time, if the virtual time delay caused by local temporary concessions is not compared and considered with the global task timing network, it often easily leads to cascading halts in downstream related processes, seriously disrupting the overall progress of the industrial production line. Summary of the Invention

[0005] In view of the aforementioned problems, this application is proposed to provide a hierarchical architecture industrial multi-agent cooperative scheduling method and system that overcomes or at least partially solves the aforementioned problems, comprising: The above-mentioned objective of this application is achieved through the following technical solution: A hierarchical architecture industrial multi-agent cooperative scheduling method, the method involving an agent candidate pool, the agent candidate pool including several dedicated agents, the dedicated agents being used to execute specific process operations in industrial manufacturing, the method including the following steps: In response to the requirement task, the requirement task is broken down into multiple sub-tasks with temporal constraints; the determination of the temporal constraints specifically includes: Extract the process dependency logic, state transition delay and resource mutual exclusion conditions of each subtask, instantiate each subtask as an action node, and instantiate the state transition delay as an implicit transition node without an execution subject. Based on the process dependency logic, a temporal connectivity edge is constructed between the action node and the implicit transition node, and based on the resource mutual exclusion condition, a mutual exclusion lock edge is constructed between the action nodes where parallel scheduling is possible. A directed topology graph is generated based on the action nodes, implicit transition nodes, temporally connected edges, and mutually exclusive locked edges. The temporal constraints between the subtasks are determined based on the directed topology graph; the temporal constraints include sequential or parallel relationships. The dedicated agent that executes the subtask is matched in the agent candidate pool, and an initial execution sequence of the dedicated agent is generated according to the temporal constraint relationship; The initial execution sequence is imported into a pre-built digital twin environment for kinematic simulation and deduction, and the task boundary data and runtime sequence data of each dedicated intelligent agent are obtained during the deduction process. Based on the task boundary data and the runtime sequence data, determine whether there is spatial interference in the initial execution sequence; When the determination result indicates the existence of spatial interference, the execution order and / or operating range of the dedicated intelligent agent are adjusted until the spatial interference is eliminated; A collaborative execution sequence is generated based on the adjusted execution order and / or task scope, and the collaborative execution sequence is sent to the corresponding dedicated intelligent agent for execution.

[0006] The second objective of this invention is achieved through the following technical solution: A hierarchical industrial multi-agent collaborative scheduling system includes: The task splitting module is used to split a requirement task into multiple sub-tasks with temporal constraints in response to the requirement task; the determination of the temporal constraints specifically includes: Extract the process dependency logic, state transition delay and resource mutual exclusion conditions of each subtask, instantiate each subtask as an action node, and instantiate the state transition delay as an implicit transition node without an execution subject. Based on the process dependency logic, a temporal connectivity edge is constructed between the action node and the implicit transition node, and based on the resource mutual exclusion condition, a mutual exclusion lock edge is constructed between the action nodes where parallel scheduling is possible. A directed topology graph is generated based on the action nodes, implicit transition nodes, temporally connected edges, and mutually exclusive locked edges. The temporal constraints between the subtasks are determined based on the directed topology graph; the temporal constraints include sequential or parallel relationships. A sequence generation module is used to match the dedicated agent executing the subtask in the agent candidate pool, and generate the initial execution sequence of the dedicated agent according to the temporal constraint relationship; The sequence inference module is used to import the initial execution sequence into a pre-built digital twin environment for kinematic simulation inference, and to obtain the task boundary data and runtime sequence data of each dedicated intelligent agent during the inference process; The data comparison module is used to determine whether there is spatial interference in the initial execution sequence based on the task boundary data and the runtime sequence data; The conflict resolution module is used to adjust the execution order and / or operating range of the dedicated intelligent agent when the judgment result is that spatial interference exists, until the spatial interference is eliminated; The sequence distribution module is used to generate a collaborative execution sequence based on the adjusted execution order and / or job scope, and distribute the collaborative execution sequence to the corresponding dedicated intelligent agent for execution.

[0007] This application has the following advantages: In the embodiments of this application, addressing the technical shortcomings of existing technologies where multi-agent collaborative scheduling commonly employs centralized rigid assignment logic, leading to limited computing power and resource mismatch during large-scale heterogeneous device matching, and where the physical execution path planning stage heavily relies on passive sensing obstacle avoidance, resulting in systemic deadlock and global clock cascading stagnation, this application provides a hierarchical architecture industrial multi-agent collaborative scheduling method, comprising the steps of: responding to a demand task, splitting the demand task into multiple sub-tasks with temporal constraints; the temporal constraints including sequential or parallel relationships; matching the dedicated agent executing the sub-task from the agent candidate pool, and according to the temporal constraints... The initial execution sequence of the dedicated intelligent agent is generated by the order constraint relationship; the initial execution sequence is imported into a pre-built digital twin environment for kinematic simulation and deduction to obtain the task boundary data and runtime sequence data of each dedicated intelligent agent during the deduction process; based on the task boundary data and the runtime sequence data, it is determined whether there is spatial interference in the initial execution sequence; when the determination result is that there is spatial interference, the execution order and / or operation range of the dedicated intelligent agent is adjusted until the spatial interference is eliminated; a cooperative execution sequence is generated according to the adjusted execution order and / or operation range, and the cooperative execution sequence is sent to the corresponding dedicated intelligent agent for execution. By responding to demand tasks and breaking them down into multiple sub-tasks with temporal constraints, and then matching dedicated agents from a candidate pool to generate an initial execution sequence, this addresses the problems of centralized, rigid assignment logic commonly used in existing technologies. These issues lead to limited computing power and resource mismatches when matching large-scale heterogeneous devices, as well as idle workstations caused by neglecting process connections. This achieves the technical effect of decoupling macroscopic, complex production intentions into microscopic execution sequences with clear temporal constraints, thus providing guidance for adaptive addressing and dynamic load balancing of underlying heterogeneous devices. Furthermore, by importing the initial execution sequence into a digital twin environment for operation... By using kinetic simulation to acquire task boundary data and runtime sequence data, and jointly adjusting the execution order and / or operating range of dedicated intelligent agents when spatial interference is detected, the existing multi-agent avoidance mechanisms often remain at the level of simple physical sensing response. They ignore the problem that the time delay caused by local temporary retreat can disrupt the originally tight process network and cause local delays to evolve into cascading shutdowns of the entire production line. This achieves the technical effect of transforming lagging physical collision detection into forward virtual data cross-deduction, and directly using spatial envelope and temporal occupancy as intervention anchors with spatiotemporal elasticity, thereby safely guiding intervention retreat.Based on this, this collaborative execution sequence can break through the traditional blind trial and error and rigid obstacle avoidance logic at the underlying control theory level. It enables the control end to detect and resolve interference and temporal crossover in virtual spacetime before the command is issued to the physical execution layer. Thus, without destroying the physical causal dependencies between processes, it can effectively eliminate spatiotemporal conflict deadlocks during high-concurrency multi-agent operations. By integrating the hierarchical dimensionality reduction task matching mechanism with the spatiotemporal reconstruction method based on digital twin feedforward pre-inspection, the collaborative scheduling scheme for large-scale heterogeneous equipment clusters can adapt to the complex evolution trend of industrial site due to process cycle drift or dynamic spatial crossover. It not only solves the technical defects of existing isolated rigid scheduling algorithms that cannot flexibly cope with the propagation of dense spatiotemporal disturbances, but also achieves the technical effect of releasing the effective utilization rate of physical equipment, reducing the hardware reactive power depreciation and wear caused by high-frequency emergency stops of drives, and finally breaking the economic efficiency bottleneck of multi-agent concurrent collaboration and improving the overall throughput of flexible production lines. Attached Figure Description

[0008] To more clearly illustrate the technical solution of this application, the drawings used in the description of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating the steps of a hierarchical architecture industrial multi-agent cooperative scheduling method according to an embodiment of this application; Figure 2 This is a structural block diagram of a hierarchical industrial multi-agent collaborative scheduling system provided in one embodiment of this application; Figure 3 This is a schematic structural block diagram of a computer electronic device provided in one embodiment of this application; 1. Computer electronic device; 2. External device; 3. Processing unit; 4. Bus; 5. Network adapter; 6. I / O interface; 7. Display; 8. Memory; 9. Random access memory; 10. Cache memory; 11. Storage system; 12. Program / utility; 13. Program module. Detailed Implementation

[0010] To make the objectives, features, and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0011] The inventors discovered through analysis of existing technologies that multi-agent collaborative scheduling in existing technologies generally adopts centralized rigid assignment logic, which leads to limited computing power and resource mismatch when matching large-scale heterogeneous devices. Furthermore, the physical execution path planning stage heavily relies on passive sensing obstacle avoidance, which in turn causes systemic deadlock and global clock cascading stagnation.

[0012] Reference Figure 1 This application illustrates a hierarchical architecture industrial multi-agent cooperative scheduling method according to an embodiment of the present application. The method involves an agent candidate pool, which includes several dedicated agents used to execute specific process operations in industrial manufacturing. The method includes the following steps: S110. In response to the requirement task, the requirement task is divided into multiple sub-tasks with temporal constraints; the determination of the temporal constraints specifically includes: Extract the process dependency logic, state transition delay and resource mutual exclusion conditions of each subtask, instantiate each subtask as an action node, and instantiate the state transition delay as an implicit transition node without an execution subject. Based on the process dependency logic, a temporal connectivity edge is constructed between the action node and the implicit transition node, and based on the resource mutual exclusion condition, a mutual exclusion lock edge is constructed between the action nodes where parallel scheduling is possible. A directed topology graph is generated based on the action nodes, implicit transition nodes, temporally connected edges, and mutually exclusive locked edges. The temporal constraints between the subtasks are determined based on the directed topology graph; the temporal constraints include sequential or parallel relationships. S120. Match the dedicated agent that executes the subtask in the agent candidate pool, and generate the initial execution sequence of the dedicated agent according to the timing constraint relationship; S130. The initial execution sequence is imported into a pre-built digital twin environment for kinematic simulation and deduction, and the task boundary data and runtime sequence data of each dedicated intelligent agent are obtained during the deduction process. S140. Based on the task boundary data and the runtime sequence data, determine whether there is spatial interference in the initial execution sequence; S150. When the judgment result indicates that spatial interference exists, adjust the execution order and / or operating range of the dedicated intelligent agent until the spatial interference is eliminated. S160. Generate a collaborative execution sequence based on the adjusted execution order and / or job scope, and send the collaborative execution sequence to the corresponding dedicated intelligent agent for execution.

[0013] First, some nouns or terms that appear in the description of this embodiment shall be interpreted as follows: A candidate pool of agents refers to a collection of specialized agents of various types and quantities, which possess different capabilities and attributes and can be dynamically selected and matched according to task requirements.

[0014] Specialized intelligent agents refer to automated equipment or software entities used to perform specific technological operations. For example, in industrial manufacturing processes, industrial robots, automated guided vehicles, and CNC machine tools can all serve as specialized intelligent agents.

[0015] A demand task refers to a production instruction or processing request received by an industrial production line that needs to be completed through the collaboration of a multi-agent system. It typically includes one or more sub-goals that need to be decomposed and executed.

[0016] Timing constraints refer to the execution order and / or parallel relationship that must be followed between subtasks after the requirement task is decomposed into multiple subtasks, ensuring the correctness and continuity of the process flow.

[0017] A subtask is a smaller-scale task unit with a clear objective and boundaries, derived from a task. Each subtask typically corresponds to one or a set of specific technological operations.

[0018] The initial execution sequence refers to the preliminary task execution plan generated for each dedicated agent after initial matching of dedicated agents and consideration of the timing constraints of sub-tasks. It has not yet undergone spatial interference detection and adjustment.

[0019] A digital twin environment refers to a pre-built virtual simulation platform that is synchronized in real time with the physical industrial site. This digital twin environment can simulate the motion, state, and interactions of physical intelligent agents, providing a virtual testbed for verifying scheduling schemes.

[0020] Kinematic simulation refers to the process of driving a digital twin entity to simulate motion based on an initial execution sequence in a digital twin environment in order to predict its trajectory, attitude, and potential collisions in three-dimensional space.

[0021] Task boundary data refers to the data obtained by each specialized intelligent agent during kinematic simulation and deduction, representing the maximum spatial range occupied or swept by the agent when performing the task, and is used to determine whether there is spatial interference.

[0022] Runtime sequence data refers to the information recorded synchronously during kinematic simulation, including state machine transitions, time nodes, and durations of each specialized intelligent agent executing a task. Typically, runtime sequence data, combined with task boundary data, can be used to determine spatiotemporal interference.

[0023] Spatial interference refers to the phenomenon where the physical spatial envelopes of two or more specialized intelligent agents overlap during multi-agent collaborative operations within the same time period. Spatial interference can lead to collisions, equipment damage, or mission interruption.

[0024] A cooperative execution sequence refers to a task execution plan that ensures safe and efficient collaborative operation among specialized intelligent agents, generated after the initial execution sequence has undergone spatial interference detection and adjustment. Typically, the cooperative execution sequence is distributed to physical intelligent agents for actual execution.

[0025] First, in response to the task requirements, the task is broken down into multiple subtasks with temporal constraints. In practice, upon receiving the task requirements, a pre-built task parsing engine or a pre-constructed global coordination agent can be invoked to decompose the task into multiple atomic operations. To accurately represent the temporal relationships of physical processes, a directed acyclic graph or process dependency matrix data structure can be used to construct the temporal constraints. In this graph structure, nodes represent subtasks, directed edges represent execution order, and nodes without connections represent parallel relationships that can be executed concurrently.

[0026] Furthermore, after completing the subtask decomposition and setting timing constraints, it is necessary to match a suitable execution agent for each subtask. One implementation approach is to maintain a candidate agent pool that contains real-time dynamic status information of all available equipment in the industrial field, such as current location, load capacity, and kinematic parameters. During matching, the physical attribute requirements of the subtask are optimized across multiple dimensions against the equipment capability matrix in the candidate agent pool, assigning the most suitable dedicated agent to each subtask. After matching, the action instructions of each dedicated agent are assembled along the time axis according to the timing constraints to generate an initial execution sequence that has not undergone interference verification.

[0027] Furthermore, the initial execution sequence is imported into a pre-built digital twin environment for kinematic simulation, acquiring task boundary data and runtime sequence data for each specialized agent during the simulation process. To verify the feasibility of the initial execution sequence, offline simulation software can be used. This offline simulation software can simulate the motion trajectory of each specialized agent during task execution based on a preset agent model and kinematic parameters. During the simulation, the system records the spatial range occupied by each agent at different time points in real time and uses it as task boundary data. Simultaneously, the system also records the start time, completion time, and duration of each agent's task, serving as runtime sequence data to initially reflect the agent's behavior in the virtual environment.

[0028] Furthermore, based on task boundary data and runtime sequence data, it is determined whether spatial interference exists in the initial execution sequence. After obtaining the task boundary data and runtime sequence data, a geometric collision detection algorithm can be used for judgment. Specifically, this algorithm checks whether the task boundary data of two or more specialized intelligent agents overlap at any given time point. For example, if the working areas of two robots intersect in three-dimensional space during the same time period, spatial interference is determined to exist, thus identifying potential physical collision risks.

[0029] Furthermore, when the judgment result indicates the existence of spatial interference, the execution order and / or operating range of the dedicated intelligent agents are adjusted until the spatial interference is eliminated. Specifically, flexible intervention can be performed through preset spatiotemporal decoupling strategies, including: adjusting the execution order, i.e., time axis intervention: without violating the original DAG process topology, by modifying local timing, such as inserting microsecond-level virtual waiting cycles or fine-tuning concurrent startup times, the interfering equipment can be staggered on the time axis; adjusting the operating range, i.e., spatial axis intervention: issuing local path replanning instructions to the interfering dedicated intelligent agents to change their original motion trajectory or attitude to avoid the conflict area. The system can use the above two strategies independently or in combination, and then send the adjusted parameters back into the digital twin environment for cyclical verification until the verification result shows that the spatial interference has been completely eliminated.

[0030] Finally, a collaborative execution sequence is generated based on the adjusted execution order and / or work scope, and this sequence is then distributed to the corresponding dedicated agents for execution. After a series of adjustments and confirmation that all spatial interference has been eliminated, the final execution plan is determined as the collaborative execution sequence. Typically, this sequence includes the precise start and end times, execution order, and adjusted work paths for each dedicated agent. Then, through an industrial communication network, it is sent as instructions to the corresponding physical dedicated agents, guiding them to perform collaborative work in the actual production environment.

[0031] Furthermore, process dependency logic defines the sequential, parallel, or mutually exclusive relationships between subtasks in the industrial manufacturing process. State transition delay refers to the time interval required between the completion of one task and the start of the next, such as non-operational time consumption like equipment cooling, material handling, and system response. Resource mutual exclusion refers to the exclusive demand of multiple subtasks for the same finite resource during execution; that is, only one task can occupy this finite resource at any given time. The extraction of state transition delay and resource mutual exclusion can be based on parsing the task intent of the demanding tasks, combined with equipment parameters, process standards, historical operating data, or expert experience. For example, equipment cooling time can be obtained by consulting equipment maintenance manuals, or shared resources can be determined by analyzing the production line configuration list.

[0032] Furthermore, each subtask is instantiated as an action node, and state transition delays are instantiated as implicit transition nodes without an execution agent. Action nodes represent specific executable subtasks, such as "grabbing parts," "welding," and "inspection," and have clearly defined input, output, and execution logic, serving as the basic unit of scheduling and execution. Implicit transition nodes represent time delays or state transition processes between subtasks. These processes themselves do not involve specific agents performing actions, but they have a crucial impact on the timing of the overall task. For example, a "waiting for cooling" step is typically not directly executed by any agent, but must be completed before subsequent tasks begin.

[0033] In defining the specific topological boundaries for the temporal constraints of subtasks, this embodiment considers not only the actions of physical equipment but also the process waiting delays in industrial settings. Specifically, industrial manufacturing often involves physical / chemical reaction times that do not require physical equipment intervention, such as "natural curing time after adhesive application" and "natural cooling time of workpieces after heat treatment." Based on this, such state transition delays are abstracted and instantiated as implicit transition nodes without an execution subject. These nodes do not occupy any dedicated physical agents but rigidly occupy an absolute time window in the directed topology graph. Combined with mutual exclusion edge locking when spatial or resource mutual exclusion exists, the traditional linear scheduling table is upgraded to a directed topology graph with time decay and blocking logic, preventing downstream process rhythm disruptions caused by ignoring process waiting times.

[0034] Furthermore, based on process dependency logic, temporal connectivity edges are constructed between action nodes and implicit transition nodes. These temporal connectivity edges represent the sequential relationship between subtasks or between a subtask and a transition time. For example, subtask B can only begin after subtask A is completed, or a cooldown period is required after subtask A is completed before subtask B can begin. These temporal connectivity edges define the logical flow and temporal dependencies of task execution. Simultaneously, based on resource mutual exclusion conditions, mutex lock edges are constructed between action nodes where parallel scheduling is possible. Mutex lock edges represent constraints between subtasks that theoretically can be executed in parallel but cannot be executed simultaneously due to shared resources. When these subtasks need to exclusively access the same resource, the mutex lock edges ensure that only one subtask can access the corresponding resource at any given time.

[0035] Furthermore, action nodes, implicit transition nodes, temporal connectivity edges, and mutex edges are integrated to generate a directed topology graph. The directed topology graph is a graphical representation where nodes represent action nodes and implicit transition nodes, and edges represent temporal connectivity edges and mutex edges. This directed topology graph clearly shows all subtasks and their temporal and resource dependencies, providing a foundation for subsequent scheduling calculations.

[0036] For example, as a specific implementation, suppose there are multiple specialized intelligent agents such as industrial robots and automated guided vehicles in an industrial manufacturing workshop, which work collaboratively in a shared work area. At a certain moment, the workshop receives a request to "produce a batch of customized parts".

[0037] First, in response to the demand task, it is broken down into multiple subtasks with temporal constraints. Specifically, the demand task can be decomposed into subtasks such as "retrieving materials from the raw material warehouse," "performing rough processing at processing station A," "performing fine processing at processing station B," "transporting semi-finished products to the assembly area," "assembling components in the assembly area," and "transporting finished products to the quality inspection area." In determining the temporal constraints, the system can first extract the process dependency logic, state transition delays, and resource mutual exclusion conditions for each subtask. For example, it extracts the "state transition delay" that the part needs to cool naturally for 3 minutes after rough processing before fine processing; and it extracts the resource mutual exclusion condition that if processing station A and processing station B perform material unloading actions simultaneously, they will compete for the same common transfer station. Based on this, the system instantiates the actual physical operations such as "retrieving materials from the raw material warehouse" and "performing rough processing at processing station A" as action nodes, and instantiates the process of "cooling naturally for 3 minutes" as an implicit transition node without an execution subject. Subsequently, based on the extracted process dependency logic, temporal connectivity edges are constructed between the corresponding action nodes and implicit transition nodes, such as connecting the "rough processing" node to the "natural cooling" node, and then to the "finishing processing" node. Furthermore, based on the aforementioned resource mutual exclusion conditions, mutual exclusion lock edges are constructed between robot A action nodes and robot B action nodes where parallel scheduling is possible, to proactively avoid conflicts at the logical level. Then, based on the generated action nodes, implicit transition nodes, temporal connectivity edges, and mutual exclusion lock edges, a directed topology graph can be generated. Based on the connectivity paths and lock edge constraints of this directed topology graph, the final temporal constraint relationships between each subtask are calculated and determined. Through the above graph theory reconstruction, the temporal constraints between the various subtasks are further clarified: for example, "retrieving materials from the raw material warehouse" must be completed before "rough processing on processing table A"; "rough processing on processing table A" and "finishing on processing table B" can be executed in parallel in the overall process, but their specific material feeding actions are restricted by mutually exclusive edge locking and cannot be in the same absolute time window; and "transporting semi-finished products to the assembly area" can only begin after rough processing, natural cooling and finishing are completed.

[0038] Based on this, the dedicated agent executing the sub-task is matched from the agent candidate pool, and an initial execution sequence for the dedicated agent is generated. For example, the sub-task of "retrieving materials from the raw material warehouse" is assigned to Automated Guided Vehicle A; the sub-task of "performing rough processing at processing table A" is assigned to Robot A; the sub-task of "performing fine processing at processing table B" is assigned to Robot B; the sub-task of "transporting semi-finished products to the assembly area" is assigned to Automated Guided Vehicle A; the sub-task of "assembling components in the assembly area" is assigned to Robot C; and the sub-task of "transporting finished products to the quality inspection area" is assigned to Automated Guided Vehicle A. Based on the temporal constraints of these sub-tasks, an initial execution sequence for Automated Guided Vehicle A, Robot A, Robot B, and Robot C can be preliminarily arranged. This initial execution sequence includes the expected start time, task duration, and approximate work path for each agent.

[0039] Then, the initial execution sequence is imported into a pre-built digital twin environment for kinematic simulation. In the digital twin environment, the twin entities of Automated Guided Vehicle A, Robot A, Robot B, and Robot C will perform virtual movements according to the initial execution sequence. During the simulation, the system acquires the swept volume topology formed by each twin entity in the three-dimensional virtual space in real time, mapping it to generate task boundary data representing the maximum interference boundary of the twin entities. Simultaneously, the system synchronously tracks the state machine transitions of each twin entity during instruction execution, generating runtime sequence data. For example, when Automated Guided Vehicle A moves from the raw material warehouse to the processing table A within a certain time period, its path and occupied space are recorded as task boundary data, and its start and end times are recorded as runtime sequence data.

[0040] Furthermore, based on task boundary data and runtime sequence data, the system determines whether spatial interference exists in the initial execution sequence. Specifically, the dynamic spatiotemporal sequences of Automated Guided Vehicle A, Robot A, Robot B, and Robot C can be timestamped and Boolean intersection operations can be performed using a sliding time window. Suppose that at a certain point in the simulation, the system detects that the reach of Automated Guided Vehicle A on its path to machining table A overlaps with the reach of Robot A during rough machining at machining table A. In this case, the system determines that spatial interference exists in this initial execution sequence.

[0041] Furthermore, when the determination result indicates the existence of spatial interference, the execution order and / or operating range of the dedicated intelligent agent are adjusted until the spatial interference is eliminated. Specifically, regarding spatial interference between the automated guided vehicle A and robot A, the system will make adjustments according to a preset scheduling strategy. For example, the system may try to postpone the departure time of the automated guided vehicle A or adjust its travel path to avoid the working area of ​​robot A. This adjustment process will be repeatedly simulated and verified in the digital twin environment until the automated guided vehicle A and robot A no longer overlap spatially at any point in time.

[0042] Finally, a collaborative execution sequence is generated based on the adjusted execution order and / or work scope, and this sequence is distributed to the corresponding dedicated intelligent agents for execution. After all spatial interferences are eliminated, the final collaborative execution sequence for Automated Guided Vehicle A, Robot A, Robot B, and Robot C is determined. This sequence includes a conflict-free task schedule and movement paths, and is distributed to the physical Automated Guided Vehicle A, Robot A, Robot B, and Robot C to guide these dedicated intelligent agents to safely and efficiently collaborate in completing the required tasks in the actual workshop. In this way, the various dedicated intelligent agents can cooperate effectively, avoiding physical collisions and ensuring the smooth operation of the production process.

[0043] In existing technologies, multi-agent collaborative scheduling often employs centralized, rigid assignment logic and generally lacks forward-looking, three-dimensional spatiotemporal extrapolation capabilities. As illustrated in the example above, if task allocation and sequence generation rely solely on human experience or simple rules, automated guided vehicle A and robot A are likely to enter the same shared area simultaneously, potentially leading to physical collisions or system deadlock. Existing solutions typically only address conflict avoidance at the physical execution level through passive sensing responses. This lagging strategy is inefficient and risky in intensive operation scenarios.

[0044] This embodiment achieves pre-determined three-dimensional spatiotemporal verification of the initial execution sequence by introducing a pre-constructed digital twin environment for kinematic simulation. In the example above, potential spatial interference between the automated guided vehicle A and robot A is identified in the digital twin environment before actual execution. In contrast to the passive conflict avoidance mechanism in existing technologies that relies on physical sensors, this embodiment can proactively predict and avoid conflicts, thereby preventing downtime and losses in actual production. Furthermore, determining the timing constraints based on the process dependency logic between sub-tasks ensures consistency between the task decomposition results and the actual industrial manufacturing process, providing more accurate and reliable basic data for agent matching and scheduling optimization, thus improving the accuracy and effectiveness of the entire collaborative scheduling method.

[0045] Furthermore, when spatial interference is detected, the solution in this embodiment can adjust the execution order and / or working range of the dedicated intelligent agent until the spatial interference is eliminated. In the example above, by adjusting the departure time or travel path of the automated guided vehicle A, the conflict with robot A was effectively resolved. Through this simulation-based iterative adjustment mechanism, the final generated cooperative execution sequence is ensured to be conflict-free and executable. Compared to the potential cascading halts in downstream processes caused by localized temporary concessions in existing technologies, the solution in this embodiment ensures the continuity and stability of the entire production process through global simulation verification and adjustment.

[0046] Therefore, the solution in this embodiment combines processes such as task decomposition, agent matching, digital twin simulation, spatial interference judgment, and dynamic adjustment to form a closed-loop collaborative scheduling process. Based on this, it not only solves the conflict risks caused by the lack of pre-calculation capabilities in existing technologies, but also improves the collaborative efficiency and security of multi-agent systems in complex industrial environments, providing reliable technical support for intelligent manufacturing.

[0047] The following will further describe a hierarchical architecture industrial multi-agent cooperative scheduling method in this exemplary embodiment.

[0048] In one embodiment of this application, the specific process of step S110, "in response to a demand task, splitting the demand task into multiple sub-tasks with temporal constraints," can be further explained in conjunction with the following description.

[0049] As described in the following steps In response to the requested task, extract the global task intent of the requested task; The global task intent is decomposed into multiple sub-tasks.

[0050] It's important to note that the global task intent refers to the overall goal or core purpose of the task, while boundary constraints specify the restrictive requirements that must be met during task execution, such as time, cost, quality, and resources. Specifically, natural language processing (NLP) techniques can be used to semantically analyze the textual description of the task, identifying core verbs, nouns, and their modifiers to deduce the global task intent. Simultaneously, qualifiers such as time, space, and quality can be parsed to extract boundary constraints. Alternatively, a preset task template and parameterized configuration interface can be used. When submitting a task, users can select a template and fill in key parameters, and the system will automatically identify the global task intent and boundary constraints.

[0051] Furthermore, the global task intent is decomposed into multiple subtasks, aiming to break down the complex overall task into the smallest independently executable units and clarify the execution order and parallel relationships between these smallest units. Subtasks are indivisible execution units with clearly defined inputs and outputs.

[0052] Through the above technical solution, this application can systematically decompose and describe complex industrial demand tasks to ensure a comprehensive understanding of the demand tasks, avoid subsequent scheduling problems caused by vague task definitions or improper decomposition, and enable the decomposed sub-tasks to closely revolve around the overall task objectives.

[0053] In one embodiment of this application, the specific process of "determining the temporal constraint relationship between each of the subtasks based on the directed topology graph" can be further explained in conjunction with the following description.

[0054] As described in the following steps Based on the earliest trigger time of the action node in the directed topology graph, the decay time of the implicit transition node, and the blocking condition of the mutual exclusion edge locking, calculate the absolute time window of each subtask. The absolute time window is used to determine the sequential or parallel relationships between the subtasks.

[0055] It should be noted that, based on this directed topology graph, the absolute time window for each subtask is calculated according to the earliest trigger time of the action nodes, the decay time of implicit transition nodes, and the blocking conditions of mutex edges. The earliest trigger time refers to the earliest time an action node can begin execution after satisfying all prerequisite timing dependencies and resource availability conditions. The decay time refers to the duration of the delay represented by the implicit transition node. The blocking condition refers to the resource occupancy constraint imposed by mutex edges; that is, when a resource is occupied, other action nodes that need this resource must wait. The absolute time window refers to the allowed start and end time range for each subtask in the entire task flow, defined by the earliest start time and the latest end time, ensuring that the task is completed while satisfying all constraints. Finally, the absolute time window determines the sequential or parallel relationships between subtasks. Specifically, by comparing the absolute time windows of different subtasks, it can be accurately determined whether these subtasks must be executed sequentially or can be executed in parallel.

[0056] Through the above technical solution, this application can accurately model the complex temporal dependencies between subtasks in industrial manufacturing processes. Specifically, by integrating state transition delays and resource mutual exclusion conditions into a directed topology graph, and calculating the absolute time window of each subtask based on this, the determined temporal constraints can fully reflect various explicit and implicit constraints in the actual production environment. Based on this, the solution in this embodiment can improve the accuracy and reliability of subtask temporal constraints, avoiding unreasonable scheduling schemes or execution conflicts caused by ignoring critical delays or resource contention. Therefore, the solution in this embodiment can effectively improve the overall efficiency of the entire hierarchical architecture industrial multi-agent collaborative scheduling method.

[0057] In one embodiment of this application, the specific process of step S120, "matching the dedicated agent that executes the sub-task in the agent candidate pool and generating the initial execution sequence of the dedicated agent according to the timing constraint relationship," can be further described in conjunction with the following description.

[0058] As described in the following steps Obtain the operating parameters and static capability attributes of each dedicated intelligent agent in the candidate pool of intelligent agents, and generate the capability feature vector of each dedicated intelligent agent based on the operating parameters and static capability attributes; Extract the capability requirements and accuracy indicators corresponding to each subtask, and perform feature space mapping on the capability requirements and accuracy indicators to generate execution feature labels; Calculate the spatial distance between each of the execution feature labels and each of the capability feature vectors, and generate a bipartite graph representing the resource allocation relationship; In the bipartite graph, an optimization objective is set and a global optimization allocation is performed to determine the target dedicated intelligent agent that performs each of the sub-tasks; Based on the aforementioned temporal constraints, the execution process of each target-specific intelligent agent is time-mapped to generate the initial execution sequence.

[0059] It should be noted that acquiring the operational parameters of a dedicated intelligent agent refers to obtaining data on its actual working status at the current moment or in the near future, such as the robot's joint angles, end effector speed, load, energy consumption, temperature, wear level, and failure rate. This data reflects the agent's real-time performance and health status. The acquisition of operational parameters can be achieved through internal sensors, external monitoring equipment, or an industrial IoT platform. Static capability attributes refer to the inherent capabilities of a dedicated intelligent agent determined during design or manufacturing, such as the robot's maximum load, maximum working range, repeatability accuracy, kinematic model, supported process types, and tool interface types. These are typically pre-stored in the agent's database or configuration file. A capability feature vector is a multi-dimensional data representation that integrates the dedicated intelligent agent's real-time operational parameters and static capability attributes, quantitatively describing the set of capabilities currently available to the agent. For example, a capability feature vector may include dimensions such as "current maximum load capacity," "current available workspace," "current accuracy level," and "current health index." The generation of capability feature vectors can be achieved through data fusion, feature extraction, or machine learning models, aiming to accurately reflect the agent's actual execution capabilities.

[0060] Furthermore, capability requirements refer to the functional requirements of the dedicated intelligent agent during the execution of the subtask, such as the size of the materials to be processed, the processing method, and the required torque. Accuracy indicators refer to the precision requirements of the subtask's execution results, such as machining tolerances, assembly clearances, and positioning error ranges. Feature space mapping transforms the capability requirements and accuracy indicators of the subtask into the same mathematical space as the capability feature vector of the dedicated intelligent agent. This can be achieved through normalization, encoding, or feature extraction, ensuring that different types of data can be effectively compared and distance calculated. Execution feature labels, after feature space mapping, are identifiers used to characterize the subtask's position in the feature space. As a vector or data structure similar to the capability feature vector structure, it can intuitively reflect the specific requirements of the subtask for the dedicated intelligent agent.

[0061] Furthermore, spatial distance is a quantitative indicator that measures the similarity or matching degree between the capability feature vector of a specialized agent and the execution feature label of a subtask. Commonly used calculation methods include Euclidean distance, Manhattan distance, or cosine similarity. A smaller distance generally indicates a higher matching degree. A bipartite graph is a special graph structure where the nodes can be divided into two independent sets, with each edge connecting a node in one set to a node in the other set. In this context, the nodes in one set represent the specialized agent, and the nodes in the other set represent the subtask. The edges represent the potential allocation relationship between the agent and the subtask, and the weight of the edges can be determined by spatial distance or matching degree.

[0062] Furthermore, the optimization objective is the goal to be achieved during global optimization allocation in a bipartite graph. For example, it could be set to minimize the total spatial distance between all assigned agents and subtasks, maximize overall matching efficiency, or achieve balanced workload distribution among specialized agents. Global optimization allocation refers to finding the optimal combination of agents and subtasks on the bipartite graph based on the set optimization objective using an algorithm. This can be achieved through various methods such as the Hungarian algorithm, the Kuhn-Munkres algorithm, linear programming, or heuristic algorithms to ensure that each subtask is assigned to the most suitable specialized agent. The target specialized agent refers to the specialized agent ultimately determined after global optimization allocation to be responsible for executing a specific subtask.

[0063] Finally, timeline mapping rationally arranges and plans the execution of assigned subtasks by each target agent in the time dimension based on the temporal constraints between subtasks. Specifically, it includes determining the start time, end time, and duration of each subtask, while considering the dependencies between tasks. The initial execution sequence, generated after timeline mapping, is a preliminary execution plan for all target agents. It typically includes information such as the subtasks to be executed by each agent, the execution order, the expected start time, and the expected end time, serving as input for subsequent digital twin simulations.

[0064] In this embodiment, to overcome the common resource mismatch problem in existing technologies when assigning target-specific intelligent agents in bipartite graph matching, a resource dimensionality reduction impedance penalty mechanism as described above is introduced. For example, when the system attempts to assign the subtask of "carrying 10kg of light materials" to a "heavy-duty AGV with a rated load of 10 tons" in the candidate pool, a large feature overflow component will be generated after the capability features of the two are mapped. When this feature overflow component is greater than a preset efficiency redundancy threshold, the system will convert it into an impedance value through a preset monotonically increasing function, which will be added as a penalty term to the weight of the associated edges in the bipartite graph. Through this mechanism, the reactive power loss caused by over-configuration of resources can be physically simulated, forcing the weighted bipartite graph matching algorithm to actively abandon this high overflow scheme during optimization and instead look for a combination of devices that better matches energy consumption and load, thereby achieving a global resource occupancy balance of heterogeneous device clusters at the physical space and computing power levels.

[0065] Through the above technical solution, this application can achieve precise matching between dedicated intelligent agents and subtasks in industrial manufacturing processes, and generate an optimized and reasonable initial execution sequence. Specifically, by comprehensively quantifying the operating parameters and static capability attributes of dedicated intelligent agents, and performing feature space mapping and distance calculation with the capability requirements and accuracy indicators of subtasks, it can ensure that each subtask is assigned to the most suitable dedicated intelligent agent, thereby improving the accuracy and efficiency of resource allocation. Based on this, the solution of this embodiment provides this optimized matching mechanism. Combined with time axis mapping based on temporal constraints, the generated initial execution sequence already possesses high rationality and feasibility before entering the digital twin environment for simulation and deduction. This can effectively reduce potential spatial interference and temporal conflicts caused by improper initial allocation, reduce the complexity and iteration number of subsequent interference detection and adjustment, not only improving the overall efficiency of the entire hierarchical architecture industrial multi-agent collaborative scheduling method, but also providing a foundation for the smooth operation of industrial manufacturing processes.

[0066] In one embodiment of this application, the specific process of "setting an optimization target and performing global optimization allocation in the bipartite graph to determine the target dedicated intelligent agent for executing each of the sub-tasks" can be further explained in conjunction with the following description.

[0067] As described in the following steps The optimization objective is to set global resource utilization balancing. Based on the bipartite graph, the feature overflow component of the capability feature vector of the dedicated agent relative to the execution feature label of the associated subtask is extracted; When the feature overflow component is greater than a preset performance redundancy threshold, the resource dimensionality reduction impedance is calculated based on the feature overflow component; the resource dimensionality reduction impedance is used to simulate the resistance caused by resource over-configuration. The resource dimensionality reduction impedance is superimposed as a penalty term onto the bipartite graph, and the global maximum weighted matching combination that meets the optimization objective is calculated by a weighted bipartite graph matching algorithm. Based on the global maximum weight matching combination, each of the subtasks is assigned to the corresponding dedicated agent to determine the target dedicated agent.

[0068] It should be noted that global resource balancing refers to ensuring that the load or utilization rate of all available resources is as consistent as possible or fluctuates within an acceptable range during resource allocation, avoiding situations where some resources are overloaded while others are idle. Specifically, this can be achieved by defining an objective function that measures the standard deviation or variance of the load of each specialized agent and minimizing this standard deviation or variance; or by setting a maximum load threshold for each specialized agent and ensuring that the load of all agents does not exceed the maximum load threshold during allocation, while maximizing overall utilization.

[0069] Furthermore, the feature overflow component refers to the portion of a dedicated agent's capability feature vector that exceeds the capability required by the execution feature label of its associated subtask. It quantifies the excess capability or redundant resources possessed by the agent when performing a specific task. Specifically, the excess portion can be calculated by subtracting or comparing the dedicated agent's capability feature vector with the execution feature label of the subtask to obtain the feature overflow component; or a function can be defined that quantifies this feature overflow component in a multi-dimensional feature space based on the difference between the agent's capability and the task requirements.

[0070] Furthermore, the performance redundancy threshold is a preset value used to determine whether the feature overflow component of a dedicated agent reaches a level requiring resource dimensionality reduction impedance calculation, defining the acceptable range of resource redundancy. This performance redundancy threshold can be set based on empirical values ​​or system design requirements. For example, when the agent's capabilities exceed task requirements by a certain percentage, significant redundancy is considered to exist. It can also be dynamically adjusted through historical data analysis or machine learning models to adapt to different production scenarios and resource conditions. Resource dimensionality reduction impedance is a penalty mechanism. When the capabilities of a dedicated agent far exceed task requirements, resource dimensionality reduction impedance is introduced to reduce its weight in the matching process, thereby guiding the system to prioritize agents with more matched capabilities or more balanced resource utilization. Resource dimensionality reduction impedance simulates the resistance caused by resource over-allocation. It can be a function proportional to the feature overflow component; the larger the overflow, the greater the impedance. Alternatively, it can be a step function; when the overflow exceeds a certain threshold, the impedance value increases significantly. Adding the penalty term to the corresponding matching weight in the bipartite graph means adding the calculated resource dimensionality reduction impedance as a negative value or subtraction term to the weight of the edge connecting the subtask node and the dedicated agent node in the bipartite graph. Based on this, in the weighted bipartite graph matching algorithm, the weight of matching combinations with greater resource dimensionality reduction resistance will be reduced, thereby reducing their probability of being selected.

[0071] Finally, the weighted bipartite graph matching algorithm is a graph theory algorithm used to find a matching in a weighted bipartite graph that maximizes or minimizes the sum of the weights of all matching edges. Common algorithms include the Hungarian algorithm and the KM algorithm. The global maximum weight matching combination refers to the matching scheme calculated by the weighted bipartite graph matching algorithm that maximizes the sum of the weights between all matched subtasks and the dedicated agent.

[0072] By setting global resource occupancy balancing as the optimization objective and introducing feature overflow components, performance redundancy thresholds, and resource dimensionality reduction impedance as penalty terms, this application can effectively avoid over-matching or resource waste between the capabilities of dedicated intelligent agents and the requirements of subtasks. Guided by the weighted bipartite graph matching algorithm, the system can calculate the global maximum weight matching combination, thereby achieving a reasonable and balanced allocation of dedicated intelligent agent resources. Based on this, the solution in this embodiment not only improves the overall utilization and scheduling efficiency of dedicated intelligent agents in industrial manufacturing processes but also reduces potential conflicts and production bottlenecks caused by improper resource allocation, ensuring the smooth and efficient operation of industrial manufacturing processes.

[0073] In one embodiment of this application, the specific process of step S130, "importing the initial execution sequence into a pre-built digital twin environment for kinematic simulation and deduction, and obtaining the task boundary data and runtime sequence data of each dedicated intelligent agent during the deduction process," can be further explained in conjunction with the following description.

[0074] As described in the following steps Generate twin-drive instructions based on the initial execution sequence; Based on the twin driving instructions, the twin entities corresponding to each of the dedicated intelligent agents are driven to perform kinematic simulation in the digital twin environment; During the kinematic simulation, a multidimensional spatial envelope is constructed based on the real-time motion parameters of each twin entity and a preset safety margin. The swept volume topology and resource occupation range formed by the multidimensional spatial envelope in the three-dimensional virtual space are collected and mapped to generate the task boundary data used to characterize the maximum interference boundary of the twin entity; The state machine transition process of each twin entity is tracked when executing the twin-driven instructions to generate the runtime sequence data.

[0075] It should be noted that twin-driven instructions are a set of specific operational instructions used to guide the kinematic simulation of twin entities in a digital twin environment. Their function is to transform the abstract initial execution sequence into low-level control commands that the digital twin system can recognize and execute. Twin-driven instructions can be generated by parsing information such as task type, target position, motion path, and velocity from the initial execution sequence and converting it into robot programming language or simulation software API call instructions. Another approach is to use preset mapping rules or machine learning models to convert high-level task descriptions into multiple low-level motion control instructions, ensuring that the twin entity can accurately reproduce the expected behavior of the physical agent. The digital twin environment is a virtual simulation platform that synchronizes and interacts with industrial manufacturing processes in the physical world in real time. The twin entity is a virtual model in the digital twin environment that corresponds one-to-one with the physical agent, possessing the same geometric, physical, and behavioral attributes as the physical agent. Kinematic simulation refers to the simulation of the motion trajectory, posture changes, and interaction with the environment of a twin entity in a three-dimensional space based on twin-driven commands within a digital twin environment. Its purpose is to pre-enact the actual operation of the physical intelligent agent in virtual space, thereby identifying potential problems without affecting actual production. One approach to kinematic simulation is to load the twin entity model into specialized robot simulation software such as RoboDK, ProcessSimulate, and Gazebo, and input twin-driven commands for simulation. Another approach is to build a customized simulation engine that integrates CAD models, a physics engine, and control logic to simulate the kinematic and dynamic behavior of a dedicated intelligent agent with high precision.

[0076] Furthermore, real-time motion parameters refer to the dynamic data of the twin entity, such as position, attitude, velocity, and acceleration, at each time step during kinematic simulation. The preset safety margin refers to the additional space or time buffer reserved during design and operation to ensure system safety and avoid collisions. The multidimensional spatial envelope is a geometric model describing the three-dimensional spatial range occupied by the twin entity at a specific moment, considering the safety margin. Its function is to accurately define the actual space occupied by the agent during motion, including its own body and the surrounding area it may influence. One method for constructing the multidimensional spatial envelope is to extend the geometric model of the twin entity outward by a preset safety distance to form an expanded collision body, such as a sphere, capsule, or convex hull; this expanded collision body then constitutes the multidimensional spatial envelope. Another method is to dynamically adjust the safety margin based on the kinematic characteristics and task requirements of the twin entity, and combine it with real-time motion parameters to generate a more adaptable irregular multidimensional spatial envelope, for example, based on point cloud data or voxel meshes.

[0077] Furthermore, swept volume topology refers to the geometric representation of the entire three-dimensional spatial region covered by the motion trajectory of a multidimensional spatial envelope over a period of time. Resource occupancy range refers to the operational space, tool extension space, etc., that a twin entity may require during task execution, in addition to its own spatial occupancy. Task boundary data is a set of data used to accurately describe the maximum possible spatial occupancy and temporal range of a twin entity during the execution of a specific task. Its role is to provide accurate geometric and temporal basis for spatial interference judgment. Acquisition and mapping can be achieved by continuously recording the position and shape of the multidimensional spatial envelope at each time step during kinematic simulation, accumulating and fusing these data to form a swept volume containing all motion paths. The geometric information and timestamp information of this swept volume constitute the task boundary data. Another approach is to use spatial discretization techniques, such as octrees and voxel meshes, to quantize the swept volume topology and resource occupancy range into a discrete set of spatial units, and combine this with temporal information to generate structured task boundary data.

[0078] Finally, the state machine transition process refers to the sequence of changes in the internal states of a twin entity over time during task execution. Internal states include idle, moving, grasping, releasing, and waiting. Runtime sequence data is a collection of data recording the time points, durations, and timestamps of key events during each state transition of the twin entity during task execution. Its role is to provide temporal dimension information about the agent's behavior, which, combined with spatial information, forms a complete spatiotemporal behavioral description. Runtime sequence data can be generated by defining a state machine model for each twin entity in the simulation engine and recording the current state, transition events, and timestamps of the state machine in real time during simulation. Another approach is to parse the execution logs of twin-driven instructions, extracting the start time, end time, and duration of key operations to construct runtime sequence data.

[0079] Through the above technical solution, this application can accurately acquire dynamic spatiotemporal behavior data of industrial multi-agents in a digital twin environment. Specifically, by converting the initial execution sequence into twin-driven instructions and driving the twin entity to perform kinematic simulation in the digital twin environment, the actual operation of the agent can be simulated in advance. Furthermore, by constructing a multi-dimensional spatial envelope based on real-time motion parameters and preset safety margins, and by collecting its swept volume topology and resource occupancy range in real time, task boundary data that accurately characterizes the maximum interference boundary of the twin entity can be generated. At the same time, synchronously tracking the state machine transition process of the twin entity and generating runtime sequence data ensures the integrity of time information. Based on this, the acquired task boundary data and runtime sequence data can provide a basis for subsequent judgment on whether there is spatial interference in the initial execution sequence, improve the accuracy and reliability of spatial interference detection, and thus effectively avoid production interruptions, equipment damage, or safety accidents caused by agent collisions in actual industrial manufacturing processes, ensuring the smooth and efficient operation of industrial production.

[0080] In one embodiment of this application, the specific process of "determining whether there is spatial interference in the initial execution sequence based on the task boundary data and the runtime sequence data" in step S140 can be further explained in conjunction with the following description.

[0081] As described in the following steps The task boundary data and runtime sequence data of each dedicated intelligent agent are timestamped to generate a dynamic spatiotemporal sequence of each dedicated intelligent agent. Boolean intersection operation is performed on the dynamic spatiotemporal sequence based on a sliding time window to calculate the spatial overlap volume of each of the dedicated intelligent agents at the same timestamp; A spatiotemporal interference matrix is ​​constructed based on the spatial overlapping volume, and it is determined whether there are non-zero matrix elements in the spatiotemporal interference matrix that represent spatial interference.

[0082] It should be noted that task boundary data refers to the data obtained through kinematic simulation in a digital twin environment, used to characterize the maximum interference boundary of the twin entity. The concept involves mapping the swept volume topology and resource occupancy range of the specialized intelligent agent in a three-dimensional virtual space, thereby forming a geometric model or dataset that represents the maximum spatial range that the agent may occupy during task execution. Task boundary data can be a polygonal mesh, a voxel set, or an envelope description based on a parametric model. Runtime sequence data refers to the data generated in a digital twin environment by tracking the state machine transitions of each twin entity when executing twin-driven instructions. The concept involves recording the behavioral trajectory, state changes, and key event occurrence times of the specialized intelligent agent in the time dimension. Runtime sequence data can be time series, event logs, or state transition diagrams. Timestamp alignment refers to synchronizing data from different sources or of different types, such as spatial and temporal data, according to their common timestamps to ensure accurate association of different events or states occurring at the same time during analysis. For example, data synchronization can be achieved through a unified system clock, event trigger time, or a preset time interval. Dynamic spatiotemporal sequences refer to the combination of spatial occupancy information and temporal evolution information of a dedicated intelligent agent to form a continuously changing spatial state description in the time dimension. Specifically, a dynamic spatiotemporal sequence can be a list of spatial geometries with timestamps, or a data structure that allows querying the spatial state at any given time using time parameters.

[0083] Furthermore, a sliding time window is a data processing technique that defines a fixed or variable-length time interval and gradually moves this time interval along the time axis to analyze the data within the window. For example, a 100-millisecond window can be set, sliding forward every 50 milliseconds to process the data within the window. Boolean intersection is a fundamental operation in set theory used to find elements common to two or more sets. In spatial geometry, Boolean intersection typically refers to calculating the overlapping portion of two or more geometric objects. For example, for two three-dimensional envelopes A and B, the Boolean intersection result is the spatial region jointly occupied by A and B. Spatial overlap volume refers to the volume of the overlapping portion of the spatial regions represented by the task boundary data of two or more specialized agents at a specific point in time or time period. This spatial overlap volume can be calculated using geometric algorithms. For example, for voxelized spatial data, it can be calculated using a logical AND operation on a voxel-by-voxel basis; for parametric geometry, the volume of the intersection region can be solved using analytical geometry or numerical methods.

[0084] Furthermore, the spatiotemporal interferometry matrix is ​​a two-dimensional data structure. Its rows and columns typically correspond to different specialized agents, and the elements in the matrix record whether spatial interference exists between these agents within a specific time period and the degree of interference. For example, a matrix element can be a Boolean value, where 0 indicates no interference and 1 indicates interference; or it can be a specific spatial overlap volume value or interference duration. A non-zero matrix element refers to an element in the spatiotemporal interferometry matrix whose value is not zero. In the context of interferometry detection, a non-zero matrix element explicitly indicates that two or more specialized agents have experienced spatial overlap at a certain point in time or within a certain time period, i.e., spatial interference exists.

[0085] Through the above technical solution, this application can accurately align the task boundary data and runtime sequence data acquired by dedicated intelligent agents in a digital twin environment with timestamps, thereby constructing a dynamic spatiotemporal sequence that comprehensively reflects the spatiotemporal behavior of the agents. Based on this, by utilizing a sliding time window and Boolean intersection operations, the spatial overlap volume of each dedicated intelligent agent at any given time can be calculated efficiently and accurately, avoiding potential omissions or misjudgments in traditional methods. Furthermore, by constructing a spatiotemporal interference matrix and determining whether non-zero elements exist, spatial interference in the initial execution sequence can be intuitively identified, providing a clear basis for subsequent conflict resolution and scheduling optimization. Based on this, the solution in this embodiment can improve the accuracy and efficiency of spatial interference detection in industrial multi-agent collaborative scheduling, effectively reducing the risk of collisions in actual production, thereby ensuring the safety and continuity of the industrial manufacturing process.

[0086] In one embodiment of this application, the specific process of "adjusting the execution order and / or operating range of the dedicated intelligent agent when the judgment result is that spatial interference exists, until the spatial interference is eliminated" can be further described in conjunction with the following description.

[0087] As described in the following steps If the non-zero matrix element exists, then the spatial interference is determined to exist; Based on the row and column index mapping relationship of the non-zero matrix elements in the spatiotemporal interference matrix, the unique physical identifier of the dedicated intelligent agent that generates the spatial interference is obtained; Based on the unique physical identifier, the execution flexibility margin of the dedicated intelligent agent corresponding to the subtask is extracted; the execution flexibility margin includes temporal relaxation and spatial avoidance tolerance; Calculate the conflict concession cost of each of the dedicated intelligent agents based on the aforementioned execution flexibility margin; Based on the conflict concession value, the dominant executor and the cooperating executor are determined in the dedicated intelligent agent that generates the spatial interference; Spatiotemporal perturbation variables are applied to the cooperative executor until the spatial interference is eliminated; wherein, the spatiotemporal perturbation variables include a temporal translation vector generated based on the temporal relaxation, and / or a spatial deformation vector generated based on the spatial avoidance tolerance; The execution order and / or job scope of the collaborative executor are updated by the spatiotemporal perturbation variables.

[0088] It should be noted that when a non-zero matrix element is detected in the spatiotemporal interference matrix, spatial interference is definitively identified. This step aims to clarify the criteria for spatial interference and ensure the accuracy of subsequent processing. Spatial interference prevents specialized agents from executing their tasks simultaneously or sequentially according to the initial plan, thus affecting the successful completion of the overall task. Subsequently, based on the row and column index mapping relationship of the non-zero matrix elements in the spatiotemporal interference matrix, the unique physical identifier of the specialized agent causing the spatial interference is obtained. The row and column indices of the spatiotemporal interference matrix typically correspond one-to-one with the specialized agents participating in the simulation. Through a pre-established mapping relationship, the indices can be converted into unique physical identifiers for the specialized agents, facilitating subsequent targeted adjustments.

[0089] Furthermore, based on the unique physical identifier, the system extracts the execution flexibility margin of the corresponding subtask for the dedicated agent. This execution flexibility margin includes temporal slack and spatial avoidance tolerance. The execution flexibility margin is a key parameter measuring the adjustability of the dedicated agent in time and space. For example, temporal slack can represent the difference between the earliest start time and the latest finish time of a task, while spatial avoidance tolerance can represent the maximum offset of its end effector in a specific direction. Additionally, the execution flexibility margin data can typically be preset and stored during the dedicated agent registration or task planning phase.

[0090] Furthermore, the conflict concession value of each specialized agent is calculated based on the execution flexibility margin. The conflict concession value is a comprehensive indicator used to evaluate the negative impact when a specialized agent adjusts its execution plan to eliminate spatial interference. For example, it can be calculated by weighting factors such as time relaxation and spatial avoidance tolerance based on a preset weight function.

[0091] Furthermore, based on the conflict concession cost, the dominant and cooperating agents are identified among the specialized agents that generate spatial interference. The dominant agent is the specialized agent with higher priority or lower concession cost in conflict resolution, while the cooperating agent is the specialized agent that needs to make significant adjustments to avoid the dominant agent. This identification of the dominant and cooperating agents can be achieved by comparing the conflict concession costs of each specialized agent.

[0092] Furthermore, spatiotemporal perturbation variables are applied to the cooperating executors until spatial interference is eliminated. These spatiotemporal perturbation variables are specific parameters used to adjust the execution plan of the cooperating executors, including temporal translation vectors generated based on temporal relaxation and / or spatial deformation vectors generated based on spatial avoidance tolerance. Specifically, the system iteratively applies spatiotemporal perturbation variables to the cooperating executors and re-detects spatial interference after each adjustment until all spatial interference is completely eliminated.

[0093] Finally, the execution order and / or scope of the cooperating agents are updated through spatiotemporal perturbation variables, thereby solidifying the results of interference elimination to ensure that the dedicated intelligent agent can operate according to the interference-free plan.

[0094] Through the above technical solution, this application provides an efficient interference elimination mechanism after detecting spatial interference between dedicated intelligent agents. Specifically, by quantifying the execution elasticity margin of dedicated intelligent agents and calculating the conflict concession cost, the responsibility for interference and the priority of adjustment can be determined, avoiding blind or random adjustments. By distinguishing between the dominant executor and the cooperating executor and applying spatiotemporal perturbation variables to the cooperating executor in a targeted manner, spatial interference can be eliminated at the lowest cost and in the most reasonable way, thereby effectively preventing collisions between dedicated intelligent agents in actual operation and ensuring the safety and continuity of industrial manufacturing processes. Furthermore, through this iterative adjustment and verification mechanism, it can be ensured that the final generated cooperative execution sequence is interference-free and feasible, thus improving the efficiency of multi-agent cooperative scheduling.

[0095] In one embodiment of this application, the specific process of "calculating the conflict concession cost of each of the dedicated intelligent agents based on the execution flexibility margin" can be further explained in conjunction with the following description.

[0096] As described in the following steps Based on the time relaxation and the spatial avoidance tolerance, calculate the local concession cost of each of the dedicated intelligent agents; Based on the temporal constraints of each of the sub-tasks, extract the associated sub-task chains corresponding to each of the dedicated intelligent agents that generate the spatial interference; The local concession cost is converted into a temporal perturbation factor, and the temporal perturbation factor is injected into the associated subtask chain for temporal deduction to generate a cascading propagation penalty term; The conflict concession cost is obtained by weighting and fusing the local concession cost with the cascading propagation penalty term to obtain the conflict concession cost of each of the dedicated intelligent agents.

[0097] It should be noted that calculating the local concession cost of each specialized intelligent agent refers to quantifying the direct costs that the agent is willing or able to incur in resolving conflicts, based on its flexibility in the time and space dimensions. For example, the local concession cost can be obtained by multiplying the time slack by a preset unit time concession cost, or by multiplying the spatial avoidance tolerance by a preset unit space concession cost. In some implementations, the local concession cost can also be a weighted sum of the time concession cost and the space concession cost, or it can be comprehensively evaluated through methods such as table lookup or fuzzy logic.

[0098] Furthermore, extracting the associated subtask chains corresponding to each specialized agent that causes spatial interference refers to identifying a series of subtasks directly or indirectly related to the subtasks performed by the specialized agent that caused the spatial interference, and these subtasks have temporal dependencies. For example, by backtracking or forward tracing the task dependency graph, starting from the subtask that caused the interference, all predecessor and successor subtasks can be identified along the temporally connected edges until a task without dependencies is encountered or a preset depth is reached. In another implementation, a task dependency matrix can be maintained. When interference occurs, this task dependency matrix can be queried to quickly identify all directly and indirectly dependent tasks associated with the interfering subtask, thereby constructing the associated subtask chains.

[0099] Furthermore, the cost of local concessions is transformed into a timing perturbation factor, which is then injected into the associated sub-task chain for timing extrapolation, generating a cascading propagation penalty term. This involves simulating the impact of local concessions on the execution time of other tasks in the task chain and quantifying the additional penalty resulting from this impact. For example, the cost of local concessions can be proportionally converted into time delays based on its main components, serving as a timing perturbation factor. This timing perturbation factor is added to the start or end times of the affected sub-tasks in the associated sub-task chain, and then the expected start / end times of all subsequent tasks in the chain are recalculated according to the timing constraints of the tasks. The cascading propagation penalty term can be the extra total delay time extrapolated, a task overdue penalty due to delay, or an additional cost incurred due to extended resource occupation time.

[0100] Finally, the local concession cost and the cascading propagation penalty term are weighted and fused to obtain the conflict concession cost of each specialized agent. This means that through weighted fusion, the direct cost of the local concession is combined with the indirect and cascading costs it causes throughout the entire task chain, thus obtaining a more accurate conflict concession cost. For example, the conflict concession cost can be equal to the product of w1 and the local concession cost plus the product of w2 and the cascading propagation penalty term, where w1 and w2 are preset weights that can be adjusted according to the importance of local and global impacts in the actual application scenario. In some implementations, a nonlinear fusion function can also be used. For example, when the cascading propagation penalty term exceeds a certain threshold, its weight will increase significantly to avoid causing excessive impact on the entire task chain.

[0101] In the flexible adjustment process to eliminate spatial interference, the game outcome of each agent is determined by the conflict concession cost. This conflict concession cost calculation combines local costs with global cascading propagation penalties. For example, if a core robotic arm on a critical process path is forced to make an avoidance maneuver, it will cause a chain reaction of stagnation in multiple downstream processes. In this case, the cascading propagation penalty calculated by the system increases sharply, making it the dominant executor and maintaining the original instruction unchanged. Conversely, for AGVs on non-critical paths with large time slack, their concession cost is minimal, and they will be judged as cooperating executors and receive spatiotemporal perturbation variables. On the physical control side, spatiotemporal perturbation variables can be specifically converted into low-level driving instructions: a temporal translation vector corresponds to ordering the AGV to wait in place at the current coordinates within microseconds, and a spatial deformation vector corresponds to ordering the AGV to deviate from the original planned trajectory by 0.5 meters to generate a local avoidance arc. Through this mechanism, the shortest path algorithm of the topology graph can be strongly bound to the physical causality of industrial control, exchanging minimal local temporal fluctuation costs for the resolution of global deadlock.

[0102] Through the above technical solution, this application, when a dedicated intelligent agent experiences spatial interference, not only considers the direct cost of a single agent's local concession, but also evaluates the cascading impact of this local concession on the entire associated sub-task chain through temporal deduction, quantifying it as a cascading propagation penalty term. Based on this, the solution in this embodiment, through this comprehensive conflict concession cost calculation method, can more accurately reflect the true cost of the adjustment strategy, thereby avoiding the problem of global suboptimal results from local optimization. This enables more reasonable and efficient decisions when determining the leading and collaborating executors, ensuring that the adjusted execution order and scope of operations minimize the impact on the overall task progress and improve the efficiency of multi-agent collaborative scheduling.

[0103] As the system implementation is basically similar to the method implementation, it is described in a relatively simple way. For relevant details, please refer to the description of the method implementation.

[0104] Reference Figure 2 This application illustrates a hierarchical industrial multi-agent collaborative scheduling system according to an embodiment of the present application, which specifically includes the following modules: The task splitting module is used to split a requirement task into multiple sub-tasks with temporal constraints in response to the requirement task; the determination of the temporal constraints specifically includes: Extract the process dependency logic, state transition delay and resource mutual exclusion conditions of each subtask, instantiate each subtask as an action node, and instantiate the state transition delay as an implicit transition node without an execution subject. Based on the process dependency logic, a temporal connectivity edge is constructed between the action node and the implicit transition node, and based on the resource mutual exclusion condition, a mutual exclusion lock edge is constructed between the action nodes where parallel scheduling is possible. A directed topology graph is generated based on the action nodes, implicit transition nodes, temporally connected edges, and mutually exclusive locked edges. The temporal constraints between the subtasks are determined based on the directed topology graph; the temporal constraints include sequential or parallel relationships. A sequence generation module is used to match the dedicated agent executing the subtask in the agent candidate pool, and generate the initial execution sequence of the dedicated agent according to the temporal constraint relationship; The sequence inference module is used to import the initial execution sequence into a pre-built digital twin environment for kinematic simulation inference, and to obtain the task boundary data and runtime sequence data of each dedicated intelligent agent during the inference process; The data comparison module is used to determine whether there is spatial interference in the initial execution sequence based on the task boundary data and the runtime sequence data; The conflict resolution module is used to adjust the execution order and / or operating range of the dedicated intelligent agent when the judgment result is that spatial interference exists, until the spatial interference is eliminated; The sequence distribution module is used to generate a collaborative execution sequence based on the adjusted execution order and / or job scope, and distribute the collaborative execution sequence to the corresponding dedicated intelligent agent for execution.

[0105] Reference Figure 3 The illustration shows a computer electronic device for implementing a hierarchical architecture industrial multi-agent cooperative scheduling method according to this application, which may specifically include the following: The aforementioned computer electronic device 1 is manifested in the form of a general-purpose computing device. The components of the computer electronic device 1 may include, but are not limited to: one or more processors or processing units 3, memory 8, and a bus 4 connecting different system components (including memory 8 and processing unit 3).

[0106] Bus 4 represents one or more of several bus architectures, including memory buses or memory controllers, peripheral buses, graphics acceleration ports, processors, or local buses using any of the various bus architectures. For example, these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Audio / Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0107] Computer electronic device 1 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by computer electronic device 1, including volatile and non-volatile media, removable and non-removable media.

[0108] Memory 8 may include computer system readable media in the form of volatile memory, such as random access memory 9 and / or cache memory 10. Computer electronic device 1 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 11 may be used to read and write non-removable, non-volatile magnetic media (commonly referred to as a "hard disk drive"). Although Figure 3 As not shown, a disk drive for reading and writing to a removable non-volatile disk (such as a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk (such as a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 4 via one or more data media interfaces. The memory may include at least one program product having a set (e.g., at least one) of program modules 13 configured to perform the functions of the embodiments of this application.

[0109] A program / utility 12 having a set (at least one) of program modules 13 may be stored, for example, in memory. Such program modules 13 include—but are not limited to—an operating system, one or more application programs, other program modules 13, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 13 typically perform the functions and / or methods described in the embodiments of this application.

[0110] The computer electronic device 1 can also communicate with one or more external devices 2 (e.g., keyboard, pointing device, display 7, camera, etc.), and with one or more devices that enable an operator to interact with the computer electronic device 1, and / or with any device that enables the computer electronic device 1 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed through the I / O interface 6. Furthermore, the computer electronic device 1 can also communicate with one or more networks (e.g., local area network (LAN)), wide area network (WAN), and / or public networks (e.g., the Internet) through the network adapter 5. Figure 3 As shown, network adapter 5 communicates with other modules of computer electronic device 1 via bus 4. It should be understood that, although... Figure 3 Not shown, it may be combined with other hardware and / or software modules, including but not limited to: microcode, device drivers, redundant processing unit 3, external disk drive array, RAID system, tape drive and data backup storage system 11, etc.

[0111] The processing unit 3 executes various functional applications and data processing by running programs stored in memory 8, such as implementing a hierarchical industrial multi-agent collaborative scheduling method provided in the embodiments of this application.

[0112] That is, when the processing unit 3 executes the above program, it performs the following: in response to the demand task, it breaks down the demand task into multiple sub-tasks with temporal constraints; the temporal constraints include sequential or parallel relationships; it matches the dedicated intelligent agents that execute the sub-tasks in the intelligent agent candidate pool, and generates an initial execution sequence for the dedicated intelligent agents according to the temporal constraints; it imports the initial execution sequence into a pre-built digital twin environment for kinematic simulation and deduction, and obtains the task boundary data and runtime sequence data of each dedicated intelligent agent during the deduction process; based on the task boundary data and the runtime sequence data, it determines whether there is spatial interference in the initial execution sequence; when the determination result is that there is spatial interference, it adjusts the execution order and / or work range of the dedicated intelligent agents until the spatial interference is eliminated; it generates a cooperative execution sequence according to the adjusted execution order and / or work range, and sends the cooperative execution sequence to the corresponding dedicated intelligent agents for execution.

[0113] In this application embodiment, the application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a hierarchical industrial multi-agent cooperative scheduling method as provided in all embodiments of the application.

[0114] That is, when the program is executed by the processor, it performs the following: In response to a demand task, it breaks down the demand task into multiple sub-tasks with temporal constraints; the temporal constraints include sequential or parallel relationships; it matches the dedicated agents that execute the sub-tasks in the agent candidate pool, and generates an initial execution sequence for the dedicated agents based on the temporal constraints; it imports the initial execution sequence into a pre-built digital twin environment for kinematic simulation and deduction, and obtains the task boundary data and runtime sequence data of each dedicated agent during the deduction process; based on the task boundary data and the runtime sequence data, it determines whether there is spatial interference in the initial execution sequence; when the determination result is that there is spatial interference, it adjusts the execution order and / or work range of the dedicated agents until the spatial interference is eliminated; it generates a cooperative execution sequence based on the adjusted execution order and / or work range, and sends the cooperative execution sequence to the corresponding dedicated agents for execution.

[0115] Any combination of one or more computer-readable media may be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that includes or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.

[0116] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including—but not limited to—electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0117] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the operator's computer, partially on the operator's computer, as a standalone software package, partially on the operator's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the operator's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider). The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably.

[0118] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0119] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0120] The above provides a detailed description of a hierarchical industrial multi-agent collaborative scheduling method and system provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A hierarchical architecture industrial multi-agent cooperative scheduling method, characterized in that, The method involves an agent candidate pool, which includes several dedicated agents for performing specific process operations in industrial manufacturing. The method includes the following steps: In response to the requirement task, the requirement task is broken down into multiple sub-tasks with temporal constraints; The determination of the timing constraint relationship specifically includes: Extract the process dependency logic, state transition delay and resource mutual exclusion conditions of each subtask, instantiate each subtask as an action node, and instantiate the state transition delay as an implicit transition node without an execution subject. Based on the process dependency logic, a temporal connectivity edge is constructed between the action node and the implicit transition node, and based on the resource mutual exclusion condition, a mutual exclusion lock edge is constructed between the action nodes where parallel scheduling is possible. A directed topology graph is generated based on the action nodes, implicit transition nodes, temporally connected edges, and mutually exclusive locked edges. The temporal constraints between the subtasks are determined based on the directed topology graph; the temporal constraints include sequential or parallel relationships. The dedicated agent that executes the subtask is matched in the agent candidate pool, and an initial execution sequence of the dedicated agent is generated according to the temporal constraint relationship; The initial execution sequence is imported into a pre-built digital twin environment for kinematic simulation and deduction, and the task boundary data and runtime sequence data of each dedicated intelligent agent are obtained during the deduction process. Based on the task boundary data and the runtime sequence data, determine whether there is spatial interference in the initial execution sequence; When the determination result indicates the existence of spatial interference, the execution order and / or operating range of the dedicated intelligent agent are adjusted until the spatial interference is eliminated; A collaborative execution sequence is generated based on the adjusted execution order and / or task scope, and the collaborative execution sequence is sent to the corresponding dedicated intelligent agent for execution.

2. The hierarchical architecture industrial multi-agent cooperative scheduling method according to claim 1, characterized in that, The step of breaking down the requirement task into multiple sub-tasks with temporal constraints in response to the requirement task specifically includes: In response to the requested task, extract the global task intent of the requested task; The global task intent is decomposed into multiple sub-tasks.

3. The hierarchical architecture industrial multi-agent cooperative scheduling method according to claim 1, characterized in that, The step of determining the temporal constraint relationships between the subtasks based on the directed topology graph specifically includes: Based on the earliest trigger time of the action node in the directed topology graph, the decay time of the implicit transition node, and the blocking condition of the mutual exclusion edge locking, calculate the absolute time window of each subtask. The absolute time window is used to determine the sequential or parallel relationships between the subtasks.

4. The hierarchical architecture industrial multi-agent cooperative scheduling method according to claim 1, characterized in that, The step of matching the dedicated agent to execute the subtask in the agent candidate pool and generating the initial execution sequence of the dedicated agent according to the temporal constraint relationship specifically includes: Obtain the operating parameters and static capability attributes of each dedicated intelligent agent in the candidate pool of intelligent agents, and generate the capability feature vector of each dedicated intelligent agent based on the operating parameters and static capability attributes; Extract the capability requirements and accuracy indicators corresponding to each subtask, and perform feature space mapping on the capability requirements and accuracy indicators to generate execution feature labels; Calculate the spatial distance between each execution feature label and each capability feature vector, and generate a bipartite graph representing the resource allocation relationship; In the bipartite graph, an optimization objective is set and a global optimization allocation is performed to determine the target dedicated intelligent agent that performs each of the sub-tasks; Based on the aforementioned temporal constraints, the execution process of each target-specific intelligent agent is time-mapped to generate the initial execution sequence.

5. The hierarchical architecture industrial multi-agent cooperative scheduling method according to claim 4, characterized in that, The step of setting an optimization objective and performing global optimization allocation in the bipartite graph to determine the target dedicated intelligent agent for executing each sub-task specifically includes: The optimization objective is to set global resource utilization balancing. Based on the bipartite graph, the feature overflow component of the capability feature vector of the dedicated agent relative to the execution feature label of the associated subtask is extracted; When the feature overflow component is greater than a preset performance redundancy threshold, the resource dimensionality reduction impedance is calculated based on the feature overflow component; the resource dimensionality reduction impedance is used to simulate the resistance caused by resource over-configuration. The resource dimensionality reduction impedance is superimposed as a penalty term onto the bipartite graph, and the global maximum weighted matching combination that meets the optimization objective is calculated by a weighted bipartite graph matching algorithm. Based on the global maximum weight matching combination, each of the subtasks is assigned to the corresponding dedicated agent to determine the target dedicated agent.

6. The hierarchical architecture industrial multi-agent cooperative scheduling method according to claim 1, characterized in that, The step of importing the initial execution sequence into a pre-constructed digital twin environment for kinematic simulation and obtaining task boundary data and runtime sequence data of each dedicated intelligent agent during the simulation process specifically includes: Generate twin-drive instructions based on the initial execution sequence; Based on the twin-driven instructions, the twin entities corresponding to each of the dedicated intelligent agents are driven to perform kinematic simulation in the digital twin environment; During the kinematic simulation, a multidimensional spatial envelope is constructed based on the real-time motion parameters of each twin entity and a preset safety margin. The swept volume topology and resource occupation range formed by the multidimensional spatial envelope in the three-dimensional virtual space are collected and mapped to generate the task boundary data used to characterize the maximum interference boundary of the twin entity; The state machine transition process of each twin entity is tracked when executing the twin-driven instructions to generate the runtime sequence data.

7. The hierarchical architecture industrial multi-agent cooperative scheduling method according to claim 1, characterized in that, The step of determining whether there is spatial interference in the initial execution sequence based on the task boundary data and the runtime sequence data specifically includes: The task boundary data and runtime sequence data of each dedicated intelligent agent are timestamped to generate a dynamic spatiotemporal sequence of each dedicated intelligent agent. Boolean intersection operation is performed on the dynamic spatiotemporal sequence based on a sliding time window to calculate the spatial overlap volume of each of the dedicated intelligent agents at the same timestamp; A spatiotemporal interference matrix is ​​constructed based on the spatial overlapping volume, and it is determined whether there are non-zero matrix elements in the spatiotemporal interference matrix that represent spatial interference.

8. The hierarchical architecture industrial multi-agent cooperative scheduling method according to claim 7, characterized in that, The step of adjusting the execution order and / or operating range of the dedicated intelligent agent when the judgment result indicates the existence of spatial interference, until the spatial interference is eliminated, specifically includes: If the non-zero matrix element exists, then the spatial interference is determined to exist; Based on the row and column index mapping relationship of the non-zero matrix elements in the spatiotemporal interference matrix, the unique physical identifier of the dedicated intelligent agent that generates the spatial interference is obtained; Based on the unique physical identifier, the execution flexibility margin of the dedicated intelligent agent corresponding to the subtask is extracted; the execution flexibility margin includes temporal relaxation and spatial avoidance tolerance; Calculate the conflict concession cost of each of the dedicated intelligent agents based on the aforementioned execution flexibility margin; Based on the conflict concession value, the dominant executor and the cooperating executor are determined in the dedicated intelligent agent that generates the spatial interference; Spatiotemporal perturbation variables are applied to the cooperative executor until the spatial interference is eliminated; wherein, the spatiotemporal perturbation variables include a temporal translation vector generated based on the temporal relaxation, and / or a spatial deformation vector generated based on the spatial avoidance tolerance; The execution order and / or job scope of the collaborative executor are updated by the spatiotemporal perturbation variables.

9. The hierarchical architecture industrial multi-agent cooperative scheduling method according to claim 8, characterized in that, The step of calculating the conflict concession value of each dedicated agent based on the execution elasticity margin specifically includes: Based on the time relaxation and the spatial avoidance tolerance, calculate the local concession cost of each of the dedicated intelligent agents; Based on the temporal constraints of each of the sub-tasks, extract the associated sub-task chains corresponding to each of the dedicated intelligent agents that generate the spatial interference; The local concession cost is converted into a temporal perturbation factor, and the temporal perturbation factor is injected into the associated subtask chain for temporal deduction to generate a cascading propagation penalty term; The conflict concession cost is obtained by weighting and fusing the local concession cost with the cascading propagation penalty term to obtain the conflict concession cost of each of the dedicated intelligent agents.

10. A hierarchical architecture industrial multi-agent collaborative scheduling system, characterized in that, The steps for implementing the hierarchical architecture industrial multi-agent cooperative scheduling method as described in any one of claims 1-9 include: The task splitting module is used to split a requirement task into multiple sub-tasks with temporal constraints in response to the requirement task; the determination of the temporal constraints specifically includes: Extract the process dependency logic, state transition delay and resource mutual exclusion conditions of each subtask, instantiate each subtask as an action node, and instantiate the state transition delay as an implicit transition node without an execution subject. Based on the process dependency logic, a temporal connectivity edge is constructed between the action node and the implicit transition node, and based on the resource mutual exclusion condition, a mutual exclusion lock edge is constructed between the action nodes where parallel scheduling is possible. A directed topology graph is generated based on the action nodes, implicit transition nodes, temporally connected edges, and mutually exclusive locked edges. The temporal constraints between the subtasks are determined based on the directed topology graph; the temporal constraints include sequential or parallel relationships. A sequence generation module is used to match the dedicated agent executing the subtask in the agent candidate pool, and generate the initial execution sequence of the dedicated agent according to the temporal constraint relationship; The sequence inference module is used to import the initial execution sequence into a pre-built digital twin environment for kinematic simulation inference, and to obtain the task boundary data and runtime sequence data of each dedicated intelligent agent during the inference process; The data comparison module is used to determine whether there is spatial interference in the initial execution sequence based on the task boundary data and the runtime sequence data; The conflict resolution module is used to adjust the execution order and / or operating range of the dedicated intelligent agent when the judgment result is that spatial interference exists, until the spatial interference is eliminated; The sequence distribution module is used to generate a collaborative execution sequence based on the adjusted execution order and / or job scope, and distribute the collaborative execution sequence to the corresponding dedicated intelligent agent for execution.

Images