A task scheduling method and system based on multi-agent collaborative optimization computer model, medium

Abstract

The application provides a task scheduling method and system based on a multi-agent collaborative optimization computer model, constructs an agent collaboration network to efficiently process complex tasks based on multi-point collaboration logic of a transaction processing network. The system simulates a transaction processing collaboration mode, provides a unified agent for task receiving, understanding, decomposition and scheduling as a processing core node, and based on a unique load balancing mode different from the prior art, uses a business processing special agent as a transaction unloading and sub-task processing node, performs three-party task placement through the unified agent, a collaborative processing agent module and a sub-task placement agent, performs two-level decomposition of the sub-task through multi-module data fusion, combines traceability query and associated task information, realizes secondary balancing of task allocation, further improves the global performance of the computer collaborative model when processing input tasks, improves the task decomposition refinement degree, and realizes high system task load processing efficiency.

Description

Technical Field

[0001] This invention belongs to the field of next-generation information technology, and specifically relates to a computer task scheduling control and regulation system (G05B) and an electrical digital data processing system (G06F) for computer task processing information systems. In particular, it relates to a task scheduling method, system, and medium based on a multi-agent collaborative optimization computer model. Background Technology

[0002] With the continuous upgrading of artificial intelligence information processing technology, information processing is gradually becoming more intelligent. More and more information and big data processing work, as well as information system management and control processes, are being delegated to intelligent agents to perform their duties, thereby improving the accuracy of data processing and the efficiency of information transmission and iteration.

[0003] An intelligent agent is a computational entity with autonomous perception and decision-making capabilities. Its core characteristic lies in its ability to operate independently in a specific environment, achieving predetermined goals without continuous human intervention. Compared to traditional software programs, intelligent agents exhibit significant initiative—they not only passively respond to external instructions but also actively monitor environmental changes, assess the current state, and adjust their behavioral strategies accordingly. This autonomy stems from their internal architecture, which typically comprises three basic components: a perception module, an inference engine, and an execution mechanism. The perception module acquires information from the external environment, the inference engine performs logical deductions based on a built-in knowledge base or learning algorithms, and the execution mechanism translates decisions into concrete actions. At the implementation level, intelligent agents can be purely software-based, such as algorithms running on servers, or they can be a combination of hardware and software, such as mobile robots equipped with sensors. Their intelligence levels span a wide range, from expert systems following fixed rules to modern applications using deep learning for complex pattern recognition—all falling within the scope of intelligent agent technology. It is important to note that the "intelligence" of an intelligent agent is not equivalent to general intelligence in the human sense, but rather a specialized capability specific to a particular task domain. This limitation makes intelligent agents more controllable and reliable in engineering practice.

[0004] From a systems theory perspective, intelligent agents constitute a key node at the intersection of distributed computing and artificial intelligence. In multi-agent systems, individual agents are no longer isolated units, but rather form an organic network with other agents through specific communication protocols and coordination mechanisms. This architecture brings two important advantages: first, parallel decomposition of tasks, complex problems can be broken down into several sub-tasks and processed synchronously by different agents, significantly improving overall efficiency; second, enhanced system robustness, the failure of a single node will not lead to global collapse, and the remaining agents can maintain basic functions through dynamic reorganization. Currently, the theoretical framework of intelligent agent technology is mainly based on the Belief-Desire-Intention (BDI) model, which uses three abstract dimensions to characterize the cognitive state of an agent: beliefs represent its understanding of the world, desires reflect its pursued goals, and intentions refer to the action plans it has committed to executing. This formal description provides a computable analytical basis for the behavioral logic of agents and also provides a theoretical tool for verifying the correctness of the system. With the popularization of edge computing and IoT technologies, intelligent agents are migrating from centralized cloud to distributed terminals. This ubiquitous trend extends their application scenarios from virtual information processing to real-time control in the physical world.

[0005] In the field of computer information processing, agent technology has reshaped the fundamental paradigm of data flow and knowledge extraction. Faced with the processing demands of massive amounts of heterogeneous data, traditional batch processing often suffers from slow response times and insufficient scalability, while agent-driven streaming architectures offer a more flexible solution. Specifically, information processing agents are deployed as autonomous nodes in the data pipeline. Each node undertakes specific preprocessing, feature extraction, or fusion tasks, and nodes collaborate loosely through asynchronous message passing. A key advantage of this architecture is adaptive load balancing: when data traffic surges, the system can dynamically instantiate new processing agents to share the load; when traffic decreases, idle agents automatically release resources. At the semantic understanding level, agents based on large language models have demonstrated deep parsing capabilities for unstructured text. They can identify entity relationships in documents, extract key event contexts, and generate structured knowledge representations. Furthermore, multi-agent collaborative information extraction systems can achieve cross-source verification. By comparing the parsing results of different agents on the same information, content with low confidence is automatically marked for manual review, thereby controlling error propagation while ensuring processing efficiency. This human-machine collaborative processing model is being widely used in professional fields such as financial risk control, medical literature analysis, and legal contract review, significantly reducing the cognitive load on knowledge workers in the information screening process.

[0006] Task scheduling, as a core component of computing resource management, is undergoing a paradigm shift from static planning to dynamic game theory, empowered by agent technology. Traditional scheduling algorithms typically rely on centralized decision-making based on pre-defined priority rules or optimization objective functions, making it difficult to cope with the suddenness of task demands and the volatility of resource status in cloud-native environments. Agent-based scheduling systems, on the other hand, employ a decentralized negotiation mechanism, modeling task submitters, resource providers, and execution monitors as agents with different utility functions. During the task allocation phase, each resource agent bids based on its own load, energy consumption level, and task characteristics, while the task agent selects the optimal match based on a multi-objective optimization strategy. This process essentially constitutes a two-way auction model in computational economics. For complex workflows with dependencies, agents decompose tasks into multiple levels through contract network protocols or variations of auction mechanisms. Parent task agents outsource sub-tasks to sub-agents with corresponding capabilities, and milestone checks and penalty clauses ensure execution quality. In terms of fault tolerance, the monitoring agent continuously tracks the task execution status. Once the failure or performance degradation of the executing agent is detected, the task migration process is immediately triggered, transferring the unfinished computation state to a backup node. This self-organizing scheduling ecosystem not only improves resource utilization but also enhances the system's resilience to anomalies such as network partitions and node failures, providing reliable operational assurance for large-scale distributed computing.

[0007] In the field of task allocation and scheduling, agent technology has formed a unique methodological system distinct from traditional operations research methods. Its core lies in reconstructing the scheduling problem as a negotiation and game process among multiple agents. Traditional scheduling algorithms typically rely on global state information and centralized optimization solutions. As the system scales up or environmental uncertainty increases, computational complexity and information acquisition costs rise sharply. Agent-based distributed scheduling architectures, however, resolve this dilemma through local decision-making. Specifically, each task is encapsulated as a task agent with specific constraints, and each computing node or execution unit is modeled as a resource agent, possessing awareness of its own state. The scheduling process is not dominated by a single central controller but is achieved through a two-way selection mechanism among agents.

[0008] Furthermore, agent technology demonstrates a refined ability to handle task dependencies in the scheduling optimization of complex workflows. In cross-domain scheduling scenarios, such as multi-cloud environments or edge-center collaborative computing, agents accumulate knowledge of the performance characteristics of different platforms through federated learning mechanisms, forming empirical task-resource matching models. This allows them to quickly generate near-optimal initial allocation schemes when new tasks arrive, and then iteratively improve them through local search. It is worth noting that the performance of agent scheduling systems is highly dependent on the design of the negotiation protocol—excessive competition can lead to resource fragmentation and increased negotiation overhead, while excessive coordination may negate the flexibility advantages of distributed architectures. Therefore, current research tends to introduce mechanism design theory, guiding the self-interested behavior of agents towards consistency with global social welfare through the design of reasonable payment rules and reputation systems. In industrial practice, agent scheduling technology has been applied to custom schedulers in Kubernetes clusters, scientific research workflow management systems, and production scheduling systems in intelligent manufacturing. Its core value lies in decoupling scheduling logic from hard-coded business rules, endowing the system with the ability to continuously self-optimize based on operational data. However, the applicant found that intelligent agents often neglect the issue of multi-agent collaboration during task processing, or merely remain at the level of collaborative data calculation or information transmission, and only adopt the basic existing load balancing mode composed of simple distribution and multiple processing nodes, without reflecting the peer-to-peer multi-mode collaboration of intelligent agents in the task allocation process, or further refined processing of sub-task allocation in the task allocation process.

[0009] This invention proposes a task scheduling method and system based on a multi-agent collaborative optimization computer model. By leveraging the multi-point collaborative logic of a transaction processing network, an agent collaborative network is constructed to efficiently handle complex tasks. The system simulates a transaction processing collaboration mode, providing a coordinating agent as the core processing node. This agent comprises a first external service interface, a first business processing output interface, and a second business processing output interface, used for task reception, understanding, decomposition, and scheduling. Based on a unique load balancing mode distinct from existing technologies, specialized business processing agents serve as transaction offloading and sub-task processing nodes. Through collaborative processing and sub-task placement, multiple specialized business processing agents perform their respective duties and work collaboratively. Specifically, the coordinating agent, the collaborative processing agent module, and the sub-task placement agent execute task placement. Through multi-module data fusion, a secondary decomposition of sub-tasks is performed. Combined with traceability queries and accompanying task information, a secondary balance in task allocation is achieved, further improving the global performance of the computer collaborative model when processing input tasks, enhancing the granularity of task decomposition, and achieving high system task load processing efficiency. Summary of the Invention

[0010] The present invention aims to provide a task scheduling system based on a multi-agent cooperative optimization computer model that is superior to existing technologies.

[0011] To achieve the above objectives, the technical solution of the present invention is as follows: A task scheduling system based on a multi-agent cooperative optimization computer model, the system comprising at least a coordinating agent, a business processing agent, a task scheduling platform, a cooperative processing agent module, and a subtask placement agent, wherein: The overall planning intelligent agent includes a first external service interface, a first business processing output interface, and a second business processing output interface, used for task reception, understanding, decomposition, and scheduling. The overall planning intelligent agent analyzes the input task received by the user from the first external service interface, sends input task traceability information and initial task placement information to the sub-task placement agent through the first business processing output interface, obtains the intelligent agent task decomposition structure sent by the collaborative processing agent module, and obtains the sub-task decomposition and placement structure provided by the sub-task placement agent. Based on the intelligent agent task decomposition structure and the sub-task decomposition and placement structure, it performs sub-task placement on the user input task, generates a sub-task placement entropy set, and sends it to the task scheduling platform through the second business processing output interface. The collaborative processing agent module includes a second external service interface, which receives the accompanying task information of the input task from the second external service interface, generates an agent task decomposition structure based on the accompanying task information, and sends it to the overall planning agent; the agent task decomposition structure is used to provide the overall planning agent with a first data encapsulation standard for task decomposition. The subtask allocation agent includes an internal service interface, which connects to the overall planning agent and receives input task tracing information and initial task allocation information sent by the overall planning agent. The input task tracing information includes at least a first tracing parameter of the input task, which is used to characterize the relevant information of the corresponding input task requester. The initial task allocation information is the initial load balancing result of the overall planning agent. The subtask allocation agent traces the input task tracing information, obtains the subtask processing standard decomposition indicator from the tracing party, generates a subtask decomposition and allocation structure based on the initial task allocation information and the subtask processing standard decomposition indicator, and sends it to the overall planning agent. The task scheduling platform is used to receive the subtask placement entropy set, and based on the subtask processing target intelligent agent ID of each subtask placement entropy information in the subtask placement entropy set, distribute the subtasks to the corresponding business processing special intelligent agents for execution according to the corresponding weights, and collect the processing feedback of each business processing special intelligent agent to confirm whether all subtask branches of the input task have been successfully processed. The dedicated intelligent agent for business processing is used to receive sub-tasks distributed by the task scheduling platform, and after completing the processing of the sub-tasks, send the task processing feedback information to the task scheduling platform.

[0012] Preferably, the step of receiving the accompanying task information of the input task from the second external service interface, and generating an agent task decomposition structure based on the accompanying task information, specifically involves: When the task requester inputs a task, they simultaneously submit accompanying task information to the second external service interface. The accompanying task information includes the following task metadata: Task requester ID; The system exempts the dedicated intelligent agent for business processing. The dedicated intelligent agent for business processing is the business processing agent ID that does not need to be configured when the system is processing the current input task because its corresponding sub-tasks can be processed out of band. System information security protection permission request, used to request data security access permission from the system; Task decomposition data processing instruction information is used to indicate the minimum data processing length after task decomposition; Based on the intelligent agent task decomposition structure and subtask decomposition and placement structure, the user input task is sub-tasked and a subtask placement entropy set is generated, which is at least: Determine the relevant information within the intelligent agent task decomposition structure, remove the subtask processing target intelligent agent ID that belongs to the system exempt business processing special intelligent agent in the corresponding subtask decomposition and placement structure, obtain the remaining subtask processing target intelligent agent ID and corresponding weight after the removal operation, add the minimum data processing length indicated by the task decomposition data processing indication information to the subtask meta-information, and use it as a subsequent data processing limit. Confirm whether the task requester ID is in the system whitelist. If so, request permission through the system information security protection. Based on the target agent ID and corresponding weight of each remaining subtask processing target agent, and the pre-configured subtask metadata of the system for the demand agent ID, basic subtask placement entropy information is generated. The basic subtask placement entropy information includes at least the target agent ID and corresponding weight of each remaining subtask processing target agent, and the pre-configured subtask metadata of the system for the demand agent ID. A subtask ID is generated for each basic subtask placement entropy information, and subtask placement entropy information is generated by adding the basic subtask placement entropy information. The subtask placement entropy information corresponding to all remaining subtask processing target agents is combined into a subtask placement entropy set.

[0013] Preferably, the subtask placement entropy set includes multiple subtask placement entropy information, each of which includes a subtask ID for identifying the subtask; a subtask processing target agent ID and corresponding weight for indicating the dedicated agent for processing the subtask and the corresponding resource limit; subtask metadata and task requester ID for providing subtask processing parameters and task requester information, respectively. The subtask processing parameters are used to manage data access permissions and task information during specific task processing. The system is built on the ThinkPHP backend framework and the Vue3 frontend framework, using MySQL to store task and agent relationship data, and Redis to provide caching support.

[0014] Preferably, the input task traceability information includes at least a first traceability parameter of the input task, the first traceability parameter being used to characterize the relevant information of the corresponding input task requester; the initial task allocation information is the initial load balancing result of the coordinating agent, specifically: An input task input according to a system preset input structure includes at least input task traceability information. The input task traceability information contains a first traceability parameter of the input task, which includes: The system inputs a task requester ID and a pre-defined identifier representing the user requesting the task. It also inputs a task requester traceability data chain pointer, indicating the destination storage address for the corresponding input task traceability information subtask processing standard decomposition indicator. This destination storage address is the memory partition address in the system database that stores the corresponding input task requester's subtask processing standard decomposition indicator. The subtask processing standard decomposition indicator is configured by the system based on a one-to-one correspondence with the input task requester ID and includes at least: When the user who inputs the task request is a specific user, the system recommends the target agent IDs for each subtask based on the user's pre-set template. as well as, The initial task allocation information is the initial subtask allocation result information performed by the overall planning agent when it has not received the agent task decomposition structure sent by the collaborative processing agent module, or obtained the subtask decomposition and allocation structure provided by the subtask allocation agent, based on the specialized processing information of multiple business processing specialized agents in the system. The initialization task allocation information includes at least: When participating in the processing of corresponding input tasks, the target agent ID of each subtask and the processing resource allocation weight of each target agent in the subtask are determined.

[0015] Preferably, the initial task allocation information includes at least: the ID of the target intelligent agent for each subtask and the processing resource allocation weight of each target intelligent agent for each subtask when participating in the processing of the corresponding input task, specifically: The overall planning agent, based on the estimated processing steps and processing costs of the input task, allocates multiple corresponding business processing specialized agents to perform processing through intelligent computing load balancing, and allocates processing resources to these agents. The size of the processing resources is represented by weights, which are multi-level processing resource allocation limits preset by the system. The business processing specialized agents are the sub-task processing target agents of the input task, and the IDs of each sub-task processing target agent are the IDs of the business processing specialized agents.

[0016] Preferably, the processing resource size is represented by a weight, and the weight is a multi-level processing resource allocation limit preset by the system, including at least: The system's preset weights are Lv0x01, Lv0x02, and Lv0x03; The processing resource overhead unit limit for Lv0x01 can be a fixed value preset by the system, which is one-quarter of that of Lv0x02 and one-sixteenth of that of Lv0x03.

[0017] Preferably, the step of generating a subtask decomposition and placement structure based on the initial task placement information and the subtask processing standard decomposition indicator, and sending it to the coordinating agent, includes at least: Match the IDs of the target agents for each subtask within the initial task placement information with the IDs of the target agents for each subtask in the subtask processing standard decomposition indicator. For a successfully matched target agent ID, increase its processing weight if the processing weight has not reached the upper limit; otherwise, maintain the upper limit of the weight. For a target agent ID that does not match successfully, the ID weight remains unchanged. Record the target agent ID and weight of each subtask processing target agent ID and weight for each of the above two categories of successfully matched and unmatched subtasks, and generate a subtask decomposition and placement structure containing the target agent ID and weight of each of the above two categories of subtask processing target agents.

[0018] Preferably, the system supports the integration of multiple large language models and external tool APIs. Through standardized communication protocols, agents can efficiently exchange information, share resources, and coordinate progress. The system also has an error handling and recovery module to perform error handling and recovery, ensuring that the execution of the overall task is not affected when a problem occurs in a single agent.

[0019] Simultaneously, this invention also proposes a task scheduling method for a task scheduling system based on a multi-agent cooperative optimization computer model as described in any of the above claims, characterized in that: Step 1: Use the overall planning intelligent agent, which includes a first external service interface, a first business processing output interface, and a second business processing output interface, to perform task reception, understanding, decomposition, and scheduling. The overall planning intelligent agent analyzes the input task received by the user from the first external service interface, sends the input task traceability information and initial task placement information to the sub-task placement agent through the first business processing output interface, obtains the intelligent agent task decomposition structure sent by the collaborative processing agent module, and obtains the sub-task decomposition and placement structure provided by the sub-task placement agent. Based on the intelligent agent task decomposition structure and the sub-task decomposition and placement structure, the user input task is placed into sub-tasks, a sub-task placement entropy set is generated, and the set is sent to the task scheduling platform through the second business processing output interface. Step 2: The collaborative processing agent module, which includes a second external service interface, receives the accompanying task information of the input task from the second external service interface, generates an agent task decomposition structure based on the accompanying task information, and sends it to the overall planning agent; the agent task decomposition structure is used to provide the overall planning agent with a first data encapsulation standard for task decomposition. Step 3: Use the subtask allocation agent containing the internal service interface to connect to the overall planning agent based on the internal service interface and receive the input task tracing information and initial task allocation information sent by the overall planning agent. The input task tracing information includes at least the first tracing parameter of the input task, which is used to characterize the relevant information of the corresponding input task requester. The initial task allocation information is the initial load balancing result of the overall planning agent. Step 4: Further use the subtask placement agent to trace the input task tracing information based on the input task tracing information, obtain the subtask processing standard decomposition indicator of the tracing party, generate the subtask decomposition and placement structure based on the initial task placement information and the subtask processing standard decomposition indicator, and send it to the overall planning agent. Step 5: Receive the subtask placement entropy set using the task scheduling platform. Based on the subtask processing target agent ID of each subtask placement entropy information in the set, distribute the subtasks according to their respective weights to the corresponding business processing specialized agents for execution. Collect processing feedback from each business processing specialized agent to confirm whether all subtask branches of the input task have been successfully processed. The business processing specialized agent receives the subtasks distributed by the task scheduling platform, completes the subtask processing, and sends task processing feedback information to the task scheduling platform afterward.

[0020] Simultaneously, the present invention also proposes a computer-readable storage medium storing a program for electronic data processing, wherein the program causes a terminal to perform the corresponding functions of a task scheduling system based on a multi-agent cooperative optimization computer model as described in any of the preceding claims.

[0021] At the same time, the present invention also proposes a computer program product, which includes computer instructions that, when executed by a processor, perform the corresponding functions of the task scheduling system based on the multi-agent cooperative optimization computer model described above.

[0022] This invention proposes a task scheduling method and system based on a multi-agent collaborative optimization computer model. By leveraging the multi-point collaborative logic of a transaction processing network, an agent collaborative network is constructed to efficiently handle complex tasks. The system simulates a transaction processing collaboration mode, providing a coordinating agent as the core processing node. This agent comprises a first external service interface, a first business processing output interface, and a second business processing output interface, used for task reception, understanding, decomposition, and scheduling. Based on a unique load balancing mode distinct from existing technologies, specialized business processing agents serve as transaction offloading and sub-task processing nodes. Through collaborative processing and sub-task allocation, multiple specialized business processing agents perform their respective duties and work collaboratively. This system has been successfully applied in various scenarios such as intelligent content creation, software development, and business decision support, significantly improving task execution efficiency and quality. Compared to traditional single-agent processing methods, the multi-agent collaborative mechanism can better utilize distributed professional capabilities to achieve parallel processing and professional division of labor, providing new ideas and methods for solving complex problems. Future systems will further enhance the self-learning and cross-platform collaboration capabilities of agents, building a more open and intelligent agent ecosystem.

[0023] Among them, the task allocation of the intelligent agent, the collaborative processing agent module, and the subtask allocation agent is coordinated. Through the data fusion of multiple modules, the secondary decomposition of the subtasks is executed. Combined with traceability query and accompanying task information, the secondary balance of task allocation is achieved, which further improves the global performance of the computer collaborative model when processing input tasks, improves the fineness of task decomposition, and achieves a high system task load processing efficiency. Attached Figure Description

[0024] Figure 1 This is a basic example diagram of a task scheduling system based on a multi-agent cooperative optimization computer model as shown in this invention; Figure 2 This is a basic example diagram illustrating the interconnection relationship between agents and related modules in a task scheduling system based on a multi-agent cooperative optimization computer model, as shown in this invention. Figure 3 This is an example diagram of the downward transmission of tasks in a task scheduling system based on a multi-agent cooperative optimization computer model, which is the subject of this invention. Figure 4 This is one of the embodiments of the task scheduling method of the task scheduling system based on the multi-agent cooperative optimization computer model claimed in this invention; Figure 5 This is one of the specific embodiments of the relevant steps executed by the scheduling platform in the task scheduling method of the task scheduling system based on the multi-agent cooperative optimization computer model claimed in this invention. Detailed Implementation

[0025] The following describes in detail several embodiments and beneficial effects of the task scheduling system and method based on a multi-agent cooperative optimization computer model claimed in this invention, in order to facilitate a more detailed examination and breakdown of this invention.

[0026] To better understand the technical solution of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0027] It should be understood that the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0028] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0029] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0030] It should be understood that although the terms "first," "second," etc., may be used to describe the methods and corresponding apparatus in the embodiments of the present invention, these keywords should not be limited to these terms. These terms are only used to distinguish keywords from each other. For example, without departing from the scope of the embodiments of the present invention, "first service processing output interface," "first data encapsulation standard," etc., may also be referred to as "second service processing output interface," "second data encapsulation standard," etc., and "second service processing output interface," "second data encapsulation standard," etc., may also be referred to as "first service processing output interface," "first data encapsulation standard."

[0031] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0032] As per the instruction manual Figure 1 - Appendix Figure 3 The diagram shown is a basic example of a task scheduling system based on a multi-agent cooperative optimization computer model according to the present invention. As a preferred embodiment that can be superimposed, each node or module can preferably interconnect with other nodes or modules for data and instruction transmission. Of course, as another preferred embodiment that can be superimposed, some nodes may not have interconnection with some other nodes, or may be allowed to disable or enable interconnection with other nodes.

[0033] The task scheduling system based on a multi-agent cooperative optimization computer model claimed in this invention includes at least a coordinating agent, a business processing specialized agent, a task scheduling platform, a cooperative processing agent module, and a subtask placement agent, wherein: As per the instruction manual Figure 2 The diagram shown illustrates a basic example of the interconnection between the coordinating agent and related modules in a task scheduling system based on a multi-agent collaborative optimization computer model, as described in this invention. The coordinating agent includes a first external service interface, a first business processing output interface, and a second business processing output interface, used for task reception, understanding, decomposition, and scheduling. The coordinating agent analyzes the input task received by the user from the first external service interface, sends input task tracing information and initial task placement information to the subtask placement agent through the first business processing output interface, obtains the agent task decomposition structure sent by the collaborative processing agent module, and obtains the subtask decomposition and placement structure provided by the subtask placement agent. Based on the agent task decomposition structure and the subtask decomposition and placement structure, it performs subtask placement on the user input task, generates a subtask placement entropy set, and sends it to the task scheduling platform through the second business processing output interface. The collaborative processing agent module includes a second external service interface, which receives the accompanying task information of the input task from the second external service interface, generates an agent task decomposition structure based on the accompanying task information, and sends it to the overall planning agent; the agent task decomposition structure is used to provide the overall planning agent with a first data encapsulation standard for task decomposition. The subtask allocation agent includes an internal service interface, which connects to the overall planning agent and receives input task tracing information and initial task allocation information sent by the overall planning agent. The input task tracing information includes at least a first tracing parameter of the input task, which is used to characterize the relevant information of the corresponding input task requester. The initial task allocation information is the initial load balancing result of the overall planning agent. The subtask allocation agent traces the input task tracing information, obtains the subtask processing standard decomposition indicator from the tracing party, generates a subtask decomposition and allocation structure based on the initial task allocation information and the subtask processing standard decomposition indicator, and sends it to the overall planning agent. As per the instruction manual Figure 3 The diagram shown illustrates a basic example of task scheduling in a multi-agent collaborative optimization computer model-based task scheduling system, where the task scheduling platform transmits data downwards. The task scheduling platform receives a set of subtask placement entropies, and based on the subtask processing target agent IDs of each subtask placement entropy in the set, distributes the subtasks to corresponding business processing specialized agents according to their respective weights for execution. It also collects processing feedback from each business processing specialized agent to confirm whether all subtask branches of the input task have been successfully processed. The dedicated intelligent agent for business processing is used to receive sub-tasks distributed by the task scheduling platform, and after completing the processing of the sub-tasks, send the task processing feedback information to the task scheduling platform.

[0034] To further differentiate it from existing technologies, as a preferred embodiment that can be overlaid, the dedicated intelligent agent for business processing is used to receive sub-tasks distributed by the task scheduling platform, and after completing the sub-task processing, sends task processing feedback information to the task scheduling platform. This includes at least: the dedicated intelligent agent for business processing is used to receive sub-tasks distributed by the task scheduling platform, and after completing the sub-task processing, sends information on whether the task processing was successfully completed, as well as task ID and / or sub-task ID and requester ID feedback information to the task scheduling platform, so that the task scheduling platform can confirm the task completion progress or whether the task needs to be re-initiated.

[0035] As a preferred embodiment that can be overlaid, the step of receiving the accompanying task information of the input task from the second external service interface, and generating an agent task decomposition structure based on the accompanying task information, specifically involves: When the task requester inputs a task, they simultaneously submit accompanying task information to the second external service interface. The accompanying task information includes the following task metadata: Task requester ID; The system exempts the dedicated intelligent agent for business processing. The dedicated intelligent agent for business processing is the business processing agent ID that does not need to be configured when the system is processing the current input task because its corresponding sub-tasks can be processed out of band. System information security protection permission request, used to request data security access permission from the system; Task decomposition data processing instruction information is used to indicate the minimum data processing length after task decomposition; Based on the intelligent agent task decomposition structure and subtask decomposition and placement structure, the user input task is sub-tasked and a subtask placement entropy set is generated, which is at least: Determine the relevant information within the intelligent agent task decomposition structure, remove the subtask processing target intelligent agent ID that belongs to the system exempt business processing special intelligent agent in the corresponding subtask decomposition and placement structure, obtain the remaining subtask processing target intelligent agent ID and corresponding weight after the removal operation, add the minimum data processing length indicated by the task decomposition data processing indication information to the subtask meta-information, and use it as a subsequent data processing limit. Confirm whether the task requester ID is in the system whitelist. If so, request permission through the system information security protection. Based on the target agent ID and corresponding weight of each remaining subtask processing target agent, and the pre-configured subtask metadata of the system for the demand agent ID, basic subtask placement entropy information is generated. The basic subtask placement entropy information includes at least the target agent ID and corresponding weight of each remaining subtask processing target agent, and the pre-configured subtask metadata of the system for the demand agent ID. A subtask ID is generated for each basic subtask placement entropy information, and subtask placement entropy information is generated by adding the basic subtask placement entropy information. The subtask placement entropy information corresponding to all remaining subtask processing target agents is combined into a subtask placement entropy set.

[0036] To further differentiate it from existing technologies, as a preferred embodiment that can be superimposed, the subtask meta-information may include various task processing parameters that are generally understood in the technical field during routine task processing and scheduling, such as resource operation parameters, task allocation index parameters or path parameters, control and operation parameters of neural networks, instruction parameters during data processing and storage, etc.

[0037] As another preferred embodiment that can be overlaid, the subtask placement entropy set includes multiple subtask placement entropy information. Each subtask placement entropy information includes a subtask ID for identifying the subtask; a subtask processing target agent ID and corresponding weight for indicating the dedicated agent for processing the subtask and the corresponding resource limit; subtask metadata and task requester ID for providing subtask processing parameters and task requester information, respectively. The subtask processing parameters are used to manage data access permissions and task information when processing specific tasks. The system is built on the ThinkPHP backend framework and the Vue3 frontend framework, uses MySQL to store task and Agent relationship data, and Redis provides caching support.

[0038] As another preferred, superimposed implementation, the task scheduling algorithm is the core technology of the system, optimizing execution efficiency through intelligent task decomposition and agent matching strategies. During task decomposition, the system comprehensively considers task objectives, constraints, and dependencies between subtasks to generate the optimal execution plan. Agent matching is based on multi-dimensional evaluation, including professional matching degree, historical performance, current load, and collaboration tendency, achieving optimal allocation of agent capabilities to task requirements through optimization algorithms. During execution, the system supports parallel processing of multiple independent subtasks and uses dynamic priority scheduling to handle unexpected situations, ensuring maximum overall execution efficiency.

[0039] As another preferred embodiment that can be overlaid, the input task traceability information includes at least a first traceability parameter of the input task, which is used to characterize the relevant information of the corresponding input task requester; the initial task placement information is the initial load balancing result of the coordinating agent, specifically: An input task input according to a system preset input structure includes at least input task traceability information. The input task traceability information contains a first traceability parameter of the input task, which includes: The system inputs a task requester ID and a pre-defined identifier representing the user requesting the task. It also inputs a task requester traceability data chain pointer, indicating the destination storage address for the corresponding input task traceability information subtask processing standard decomposition indicator. This destination storage address is the memory partition address in the system database that stores the corresponding input task requester's subtask processing standard decomposition indicator. The subtask processing standard decomposition indicator is configured by the system based on a one-to-one correspondence with the input task requester ID and includes at least: When the user who requests the task is a specific user, the system recommends the target agent IDs for each subtask based on the user's pre-set template. as well as, The initial task allocation information is the initial subtask allocation result information performed by the overall planning agent when it has not received the agent task decomposition structure sent by the collaborative processing agent module, or obtained the subtask decomposition and allocation structure provided by the subtask allocation agent, based on the specialized processing information of multiple business processing specialized agents in the system. To further differentiate itself from existing technologies, as a preferred embodiment that can be superimposed, the initial task allocation information is the initial sub-task allocation result information performed by the coordinating agent when it has not received the agent task decomposition structure sent by the collaborative processing agent module, nor obtained the sub-task decomposition and allocation structure provided by the sub-task allocation agent. This is based on the specialized processing information of multiple business processing specialized agents in the system. Specifically, when the coordinating agent receives an input task, if the collaborative processing agent module has not yet obtained the agent task decomposition structure, and the sub-task allocation agent has not provided the coordinating agent with its sub-task decomposition and allocation structure, the coordinating agent performs task decomposition on the input task information according to any intelligent neural network processing flow and rules in the existing technology. Based on the task structured data or text information after decomposition, it matches each sub-task to a business processing specialized agent associated with the corresponding theme. As a preferred embodiment that can be superimposed, the multiple business processing specialized agents with corresponding themes can be, for example, specialized agents for data analysis, document text editing, specific functions in code development, or design information processing. These specialized agents are organized through a tree-like binding network, allowing for flexible combinations to handle various complex scenarios. After the task is initially decomposed and matched to a business-specific intelligent agent associated with the corresponding topic, clustering is performed based on the intelligent matching results and the processing complexity of each subtask after task decomposition. The processing resource overhead weights for each subtask are then set. Furthermore, using each specialized business-specific intelligent agent as a distinguishing element, the ID of each specialized business-specific intelligent agent (i.e., the target intelligent agent for each subtask) and the processing resource allocation weights of each target intelligent agent are stored as initial task placement information.

[0040] The initialization task allocation information includes at least: When participating in the processing of corresponding input tasks, the target agent ID of each subtask and the processing resource allocation weight of each target agent in the subtask are determined.

[0041] As another preferred embodiment that can be superimposed, the initialization task allocation information includes at least: the ID of the target intelligent agent for each subtask and the processing resource allocation weight of each target intelligent agent for each subtask when participating in the processing of the corresponding input task, specifically: The overall planning agent, based on the estimated processing steps and processing costs of the input task, allocates multiple corresponding business processing specialized agents to perform processing through intelligent computing load balancing, and allocates processing resources to these agents. The size of the processing resources is represented by weights, which are multi-level processing resource allocation limits preset by the system. The business processing specialized agents are the sub-task processing target agents of the input task, and the IDs of each sub-task processing target agent are the IDs of the business processing specialized agents.

[0042] As another preferred embodiment that can be superimposed, the processing resource size is characterized by weights, where the weights are system-preset multi-level processing resource allocation limits, including at least: The system's preset weights are Lv0x01, Lv0x02, and Lv0x03; Furthermore, the processing resource overhead unit limit of Lv0x01 is one-quarter of that of Lv0x02 and one-sixteenth of that of Lv0x03.

[0043] As another preferred embodiment that can be overlaid, the step of generating a subtask decomposition and placement structure based on the initial task placement information and the subtask processing standard decomposition indicator, and sending it to the overall planning agent, includes at least: Match the IDs of the target agents for each subtask within the initial task placement information with the IDs of the target agents for each subtask in the subtask processing standard decomposition indicator. For a successfully matched target agent ID, increase its processing weight if the processing weight has not reached the upper limit; otherwise, maintain the upper limit of the weight. For a target agent ID that does not match successfully, the ID weight remains unchanged. Record the target agent ID and weight of each subtask processing target agent ID and weight for each of the above two categories of successfully matched and unmatched subtasks, and generate a subtask decomposition and placement structure containing the target agent ID and weight of each of the above two categories of subtask processing target agents.

[0044] As another preferred embodiment that can be overlaid, the system supports the integration of multiple large language models and external tool APIs. Through standardized communication protocols, agents can efficiently exchange information, share resources, and coordinate progress. The system also has an error handling and recovery module to perform error handling and recovery, ensuring that the execution of the overall task is not affected when a problem occurs in a single agent.

[0045] Simultaneously, this invention also proposes a task scheduling method for a task scheduling system based on a multi-agent cooperative optimization computer model as described in any of the above claims, as shown in the appendix to the specification. Figure 4 The diagram shown is a basic example of a task scheduling method for a task scheduling system based on a multi-agent cooperative optimization computer model, as illustrated in this invention. Its features include: Step S102: Use the overall planning agent, which includes a first external service interface, a first business processing output interface, and a second business processing output interface, to perform task reception, understanding, decomposition, and scheduling; the overall planning agent analyzes the input task received by the user from the first external service interface, sends the input task traceability information and initial task placement information to the subtask placement agent through the first business processing output interface, obtains the agent task decomposition structure sent by the collaborative processing agent module, and obtains the subtask decomposition and placement structure provided by the subtask placement agent. Based on the agent task decomposition structure and the subtask decomposition and placement structure, the user input task is placed into subtasks, a subtask placement entropy set is generated, and sent to the task scheduling platform through the second business processing output interface; Step S104: The collaborative processing agent module, which includes a second external service interface, receives the accompanying task information of the input task from the second external service interface, generates an agent task decomposition structure based on the accompanying task information, and sends it to the overall planning agent; the agent task decomposition structure is used to provide the overall planning agent with a first data encapsulation standard for task decomposition. Step S106: Use a subtask allocation agent containing an internal service interface to connect to the overall planning agent based on the internal service interface and receive input task tracing information and initial task allocation information sent by the overall planning agent. The input task tracing information includes at least a first tracing parameter of the input task, which is used to characterize the relevant information of the corresponding input task requester. The initial task allocation information is the initial load balancing result of the overall planning agent. Step S108: Further use the subtask placement agent to trace the input task requester based on the input task tracing information, obtain the subtask processing standard decomposition indicator of the requester, generate the subtask decomposition and placement structure based on the initial task placement information and the subtask processing standard decomposition indicator, and send it to the overall planning agent. As per the instruction manual Figure 5 The diagram shown is a basic example of the relevant steps executed by the scheduling platform in the task scheduling method of the task scheduling system based on the multi-agent cooperative optimization computer model of the present invention.

[0046] Step S110: The task scheduling platform receives the subtask placement entropy set. Based on the subtask processing target agent ID of each subtask placement entropy information in the subtask placement entropy set, the subtasks are distributed to the corresponding business processing special agents according to the corresponding weights for execution. The processing feedback of each business processing special agent is collected to confirm whether all subtask branches of the input task have been successfully processed. The business processing special agent receives the subtasks distributed by the task scheduling platform, completes the subtask processing, and sends the task processing feedback information to the task scheduling platform.

[0047] Simultaneously, the present invention also proposes a computer-readable storage medium storing a program for electronic data processing, wherein the program causes a terminal to perform the corresponding functions of a task scheduling system based on a multi-agent cooperative optimization computer model as described in any of the preceding claims.

[0048] At the same time, the present invention also proposes a computer program product, which includes computer instructions that, when executed by a processor, perform the corresponding functions of the task scheduling system based on the multi-agent cooperative optimization computer model described above.

[0049] This invention proposes a task scheduling method and system based on a multi-agent collaborative optimization computer model. By leveraging the multi-point collaborative logic of a transaction processing network, an agent collaborative network is constructed to efficiently handle complex tasks. The system simulates a transaction processing collaboration mode, providing a coordinating agent as the core processing node. This agent comprises a first external service interface, a first business processing output interface, and a second business processing output interface, used for task reception, understanding, decomposition, and scheduling. Based on a unique load balancing mode distinct from existing technologies, specialized business processing agents serve as transaction offloading and sub-task processing nodes. Through collaborative processing and sub-task placement, multiple specialized business processing agents perform their respective duties and work collaboratively. Specifically, the coordinating agent, the collaborative processing agent module, and the sub-task placement agent execute task placement. Through multi-module data fusion, a secondary decomposition of sub-tasks is performed. Combined with traceability queries and accompanying task information, a secondary balance in task allocation is achieved, further improving the global performance of the computer collaborative model when processing input tasks, enhancing the granularity of task decomposition, and achieving high system task load processing efficiency.

[0050] In all the above embodiments, in order to achieve certain special data transmission and read / write function requirements, the above methods and corresponding devices can be expanded by adding devices, modules, components, hardware, pin connections or memory, processor differences during operation.

[0051] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the methods, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0052] In the embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of method steps is only a logical or functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections through some interfaces, apparatuses, or units, and may be electrical, mechanical, or other forms.

[0053] The units described as separate components of the method and apparatus may or may not be logically or physically separate, and may not be physical units. That is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0054] Furthermore, the method steps and their implementations, as well as the functional units, in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit described above can be implemented in hardware or in the form of hardware plus software functional units.

[0055] The aforementioned methods and apparatus can be implemented as integrated units in the form of software functional units, which can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), NVRAM, magnetic disks, or optical disks.

[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0057] It should be noted that the above embodiments are only used to more clearly explain and illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A task scheduling system based on a multi-agent cooperative optimization computer model, the system comprising at least a coordinating agent, a business processing specialized agent, a task scheduling platform, a cooperative processing agent module, and a subtask placement agent, wherein: The overall planning intelligent agent includes a first external service interface, a first business processing output interface, and a second business processing output interface, which are used for task reception, understanding, decomposition, and scheduling; The collaborative processing agent module includes a second external service interface, which receives the accompanying task information of the input task from the second external service interface, generates an agent task decomposition structure based on the accompanying task information, and sends it to the overall planning agent; the agent task decomposition structure is used to provide the overall planning agent with a first data encapsulation standard for task decomposition. The system receives companion task information of the input task from the second external service interface, and generates an intelligent agent task decomposition structure based on the companion task information, specifically as follows: When the task requester inputs a task, they simultaneously submit accompanying task information to the second external service interface. The accompanying task information includes the following task metadata: Task requester ID; The system exempts the dedicated intelligent agent for business processing. The dedicated intelligent agent for business processing is the business processing agent ID that does not need to be configured when the system is processing the current input task because its corresponding sub-tasks can be processed out of band. System information security protection permission request, used to request data security access permission from the system; Task decomposition data processing instruction information is used to indicate the minimum data processing length after task decomposition; Based on the intelligent agent task decomposition structure and subtask decomposition and placement structure, the user input task is sub-tasked and a subtask placement entropy set is generated, which is at least: Determine the relevant information within the intelligent agent task decomposition structure, remove the subtask processing target intelligent agent ID that belongs to the system exempt business processing special intelligent agent in the corresponding subtask decomposition and placement structure, obtain the remaining subtask processing target intelligent agent ID and corresponding weight after the removal operation, add the minimum data processing length indicated by the task decomposition data processing indication information to the subtask meta-information, and use it as a subsequent data processing limit. Confirm whether the task requester ID is in the system whitelist. If so, request permission through the system information security protection. Based on the target agent ID and corresponding weight of each remaining subtask processing target agent, and the pre-configured subtask meta-information of the system for the demand party ID, the basic information of subtask placement entropy is generated. The basic information of subtask placement entropy includes at least the target agent ID and corresponding weight of each remaining subtask processing target agent, and the pre-configured subtask meta-information of the system for the demand party ID. For each subtask, generate a subtask ID based on the basic information of subtask placement entropy, add the basic information of subtask placement entropy to generate subtask placement entropy information, and combine the subtask placement entropy information corresponding to the target intelligent agents of all remaining subtasks into a subtask placement entropy set. The subtask allocation agent includes an internal service interface, which connects to the overall planning agent and receives input task tracing information and initial task allocation information sent by the overall planning agent. The task scheduling platform is used to receive the subtask placement entropy set, and based on the subtask processing target intelligent agent ID of each subtask placement entropy information in the subtask placement entropy set, distribute the subtasks to the corresponding business processing special intelligent agents for execution according to the corresponding weights, and collect the processing feedback of each business processing special intelligent agent to confirm whether all subtask branches of the input task have been successfully processed. A dedicated intelligent agent for business processing is used to receive sub-tasks distributed by the task scheduling platform, and after completing the processing of the sub-tasks, send the task processing feedback information back to the task scheduling platform.

2. The task scheduling system based on a multi-agent cooperative optimization computer model as described in claim 1, characterized in that: The overall planning intelligent agent includes a first external service interface, a first business processing output interface, and a second business processing output interface, used for task reception, understanding, decomposition, and scheduling, specifically including: The overall planning agent analyzes the input tasks received by the user from the first external service interface, sends the input task traceability information and initial task placement information to the subtask placement agent through the first business processing output interface, obtains the agent task decomposition structure sent by the collaborative processing agent module, and obtains the subtask decomposition and placement structure provided by the subtask placement agent. Based on the agent task decomposition structure and the subtask decomposition and placement structure, the agent performs subtask placement on the user input task, generates a subtask placement entropy set, and sends it to the task scheduling platform through the second business processing output interface. The input task traceability information includes at least a first traceability parameter of the input task, which is used to characterize the relevant information of the corresponding input task requester; the initial task placement information is the initial load balancing result of the coordinating agent.

3. The task scheduling system based on a multi-agent cooperative optimization computer model as described in claim 1, characterized in that: The subtask allocation entropy set includes multiple subtask allocation entropy information. Each subtask allocation entropy information includes a subtask ID, used to identify the subtask; a subtask processing target agent ID and corresponding weight, used to indicate the dedicated business processing agent and corresponding resource limit for processing the subtask; subtask metadata and task requester ID, used to provide subtask processing parameters and task requester information respectively. The subtask processing parameters are used to manage data access permissions and task information when specifically processing tasks. The system is built based on the ThinkPHP backend framework and the Vue3 frontend framework, using MySQL to store task and Agent relationship data, and Redis to provide caching support. The subtask allocation agent also traces the input task requester based on the input task tracing information, obtains the subtask processing standard decomposition indicator of the requester, generates a subtask decomposition and allocation structure based on the initial task allocation information and the subtask processing standard decomposition indicator, and sends it to the overall planning agent.

4. The task scheduling system based on a multi-agent cooperative optimization computer model as described in claim 3, characterized in that: The input task traceability information includes at least a first traceability parameter for the input task, which is used to characterize the relevant information of the corresponding input task requester; the initial task allocation information is the initial load balancing result of the coordinating agent, specifically: An input task input according to a system preset input structure includes at least input task traceability information. The input task traceability information contains a first traceability parameter of the input task, which includes: The system inputs a task requester ID and a pre-defined identifier representing the user requesting the task. It also inputs a task requester traceability data chain pointer, indicating the destination storage address for the corresponding input task traceability information subtask processing standard decomposition indicator. This destination storage address is the memory partition address in the system database that stores the corresponding input task requester's subtask processing standard decomposition indicator. The subtask processing standard decomposition indicator is configured by the system based on a one-to-one correspondence with the input task requester ID and includes at least: When the user who inputs the task request is a specific user, the system recommends the target agent IDs for each subtask based on the user's pre-set template. as well as, The initial task allocation information is the initial subtask allocation result information performed by the overall planning agent when it has not received the agent task decomposition structure sent by the collaborative processing agent module, or obtained the subtask decomposition and allocation structure provided by the subtask allocation agent, based on the specialized processing information of multiple business processing specialized agents in the system. The initialization task allocation information includes at least: When participating in the processing of corresponding input tasks, the target agent ID of each subtask and the processing resource allocation weight of each target agent in the subtask are determined.

5. The task scheduling system based on a multi-agent cooperative optimization computer model as described in claim 3 or 4, characterized in that: The initialization task allocation information includes at least: the ID of the target intelligent agent for each subtask and the processing resource allocation weight of each target intelligent agent for each subtask when participating in the processing of the corresponding input task, specifically: The overall planning agent, based on the estimated processing steps and overhead of each step required to process the input task, allocates multiple corresponding business processing specialized agents to perform processing through intelligent computing load balancing, and allocates processing resources to these agents. The size of the processing resources is represented by weights, which are multi-level processing resource allocation limits preset by the system. The business processing specialized agents are the sub-task processing target agents of the input task, and the IDs of each sub-task processing target agent are the IDs of the business processing specialized agents. The processing resource size is represented by a weight, which is a multi-level processing resource allocation limit preset by the system, including at least: The system's preset weights are Lv0x01, Lv0x02, and Lv0x03; Furthermore, the processing resource overhead unit limit of Lv0x01 is one-quarter of that of Lv0x02 and one-sixteenth of that of Lv0x03.

6. The task scheduling system based on a multi-agent cooperative optimization computer model as described in claim 5, characterized in that: The step of generating a subtask decomposition and placement structure based on the initial task placement information and the subtask processing standard decomposition indicator, and sending it to the overall planning agent, includes at least the following: Match the IDs of the target agents for each subtask within the initial task placement information with the IDs of the target agents for each subtask in the subtask processing standard decomposition indicator. For a successfully matched target agent ID, increase its processing weight if the processing weight has not reached the upper limit; otherwise, maintain the upper limit of the weight. For a target agent ID that does not match successfully, the ID weight remains unchanged. Record the target agent IDs and weights of the subtasks for each of the two categories of successfully matched and unmatched subtasks, and generate a subtask decomposition and arrangement structure containing the target agent IDs and weights of each of the two categories of subtasks.

7. The task scheduling system based on a multi-agent cooperative optimization computer model as described in claim 2, characterized in that: The system supports the integration of multiple large language models and external tool APIs. Through standardized communication protocols, agents can efficiently exchange information, share resources, and coordinate progress. The system also has an error handling and recovery module to perform error handling and recovery, ensuring that the execution of the overall task is not affected when a problem occurs in a single agent.

8. A task scheduling method applied to a task scheduling system based on a multi-agent cooperative optimization computer model as described in any one of claims 1-7, characterized in that: Step 1: Use the overall planning agent, which includes the first external service interface, the first business processing output interface, and the second business processing output interface, to perform task reception, understanding, decomposition, and scheduling; Step 2: The collaborative processing agent module, which includes a second external service interface, receives the accompanying task information of the input task from the second external service interface, generates an agent task decomposition structure based on the accompanying task information, and sends it to the coordinating agent; the agent task decomposition structure is used to provide the coordinating agent with a first data encapsulation standard for task decomposition. Step 3: Use the subtask allocation agent containing the internal service interface to connect to the overall planning agent based on the internal service interface and receive the input task tracing information and initial task allocation information sent by the overall planning agent. The input task tracing information includes at least the first tracing parameter of the input task, which is used to characterize the relevant information of the corresponding input task requester. The initial task allocation information is the initial load balancing result of the overall planning agent. Step 4: Further use the subtask placement agent to trace the input task tracing information based on the input task tracing information, obtain the subtask processing standard decomposition indicator of the tracing party, generate the subtask decomposition and placement structure based on the initial task placement information and the subtask processing standard decomposition indicator, and send it to the overall planning agent. Step 5: Receive the subtask placement entropy set using the task scheduling platform. Based on the subtask processing target agent ID of each subtask placement entropy information in the set, distribute the subtasks according to their respective weights to the corresponding business processing specialized agents for execution. Collect processing feedback from each business processing specialized agent to confirm whether all subtask branches of the input task have been successfully processed. The business processing specialized agent receives the subtasks distributed by the task scheduling platform, completes the subtask processing, and sends task processing feedback information to the task scheduling platform afterward.

9. A computer-readable storage medium storing a program for electronic data processing, wherein, The program enables the terminal to perform the corresponding functions of the task scheduling system based on a multi-agent cooperative optimization computer model as described in any one of claims 1-7.

Images