Intelligent Agent process arrangement system and method based on large model driving

Abstract

The invention relates to the technical field of artificial intelligence, in particular to an intelligent Agent process arrangement system and method based on large model driving, and the system comprises a hierarchical architecture system which is composed of a module management layer, a task analysis layer and an execution optimization layer in a cooperative manner. The method has the advantages that by means of collaborative operation of a layered architecture, core advantages of a large model are deeply mined, various complex tasks can be efficiently and accurately processed, all-around optimization of the execution process is achieved, and a convenient, intelligent and efficient task processing solution is elaborately created for a user. The system can flexibly and accurately determine the type and the number of the intelligent Agents participating in task execution according to the specific characteristics, complexity and real-time requirements of the tasks, and skillfully plan the cooperation mode of the intelligent Agents, thereby showing excellent adaptive capacity to various complex, variable and cross-domain tasks.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence, and specifically to an intelligent Agent process orchestration system and method driven by a large model. Background Art

[0002] With the rapid development of artificial intelligence technology, intelligent agents have been widely used in many fields, but traditional task orchestration systems still face many technical bottlenecks:

[0003] 1). Limitations of traditional rule-driven systems. Most systems adopt static Agent scheduling strategies and lack the ability to adjust the collaboration chain in real time according to task requirements, and have poor robustness in the face of sudden exceptions (such as Agent failures and task priority changes).

[0004] 2). Challenges in the integration of large models and automation technologies. When simply relying on large models to generate process solutions, decisions that do not conform to business rules or security constraints may be generated, and the generated results are uncontrollable. Existing methods mostly use offline log analysis for process optimization, and cannot achieve real-time feedback and adaptive adjustment during the execution process. Optimization relies on manual intervention and has a high optimization cost.

[0005] 3). Difficulties in cross-domain scenario adaptation. Traditional systems need to develop customized processes separately for different domains (such as finance, manufacturing, and healthcare), with poor generality, long development cycles, and difficult domain knowledge migration. Summary of the Invention

[0006] The purpose of the present invention is to provide an intelligent Agent process orchestration system and method driven by a large model to solve the problems raised in the above background art.

[0007] To achieve the above purpose, the present invention provides the following technical solution: An intelligent Agent process orchestration system driven by a large model, including a hierarchical architecture system jointly composed of a module management layer, a task parsing layer, and an execution optimization layer:

[0008] Module management layer: Integrate mainstream pre-trained large models, and screen the base models through a dynamic evaluation mechanism covering key dimensions such as the number of parameters, domain adaptability, and inference efficiency; build a domain knowledge enhancement mechanism, inject a professional term dictionary and a domain ontology map, and adopt a two-stage fine-tuning strategy. First, initially optimize with the general capabilities of the large model, and then strengthen the design for special tasks to improve the performance of the model in a specific domain;

[0009] Task Parsing Layer: Provide a multi-modal task description input interface that supports submitting task descriptions in the forms of text, tables, and flowcharts; after receiving the task description, the large model module applies the MECE principle to decompose the task objectives exhaustively and mutually exclusively, constructs a directed acyclic graph to represent the task topology, and performs critical path analysis based on the PERT-based duration estimation model to screen out the optimal execution path;

[0010] Execution Optimization Layer: Introduce a dynamic optimization mechanism, rely on the real-time resource monitoring feedback loop to monitor the usage status of system computing resources, memory resources, and network resources, and combine with a scheduling strategy based on reinforcement learning to dynamically adjust the resource allocation plan according to task priorities, subtask resource requirements, and system load conditions; adopt dynamic batch processing technology to batch process task requests and present the processing results in a structured form through a standardized output interface.

[0011] Preferably, the Module Management Layer includes a model processing module for:

[0012] Model Screening: Integrate mainstream pre-trained large models, and comprehensively consider the models from multiple key dimensions such as parameter magnitude, domain adaptability, and inference efficiency through an advanced dynamic evaluation mechanism to screen out the base model that best suits the task requirements;

[0013] Knowledge Enhancement: Build a domain knowledge enhancement mechanism, fully mine the professional knowledge in the domain by injecting a professional term dictionary and a domain ontology graph, and integrate it into the large model to significantly improve the performance of the model in a specific domain;

[0014] Fine-tuning Optimization: Adopt a two-stage fine-tuning strategy. In the first stage, use the general capabilities of the large model for preliminary optimization, and in the second stage, carry out intensive design for special tasks to make the model accurately adapt to various complex task scenarios.

[0015] Preferably, the Task Parsing Layer includes a task processing module for:

[0016] Interface Input: Provide a multi-modal task description input interface that supports users to submit task descriptions in various forms such as text, tables, and flowcharts, improving the compatibility and usability of the system;

[0017] Logical Decomposition: After receiving the task description, the large model module uses its powerful language understanding and logical reasoning capabilities to apply the MECE principle to decompose the task objectives exhaustively and mutually exclusively, ensuring the comprehensiveness and accuracy of the task decomposition;

[0018] Structure Modeling and Analysis: Construct a directed acyclic graph to represent the task topology, accurately model the dependencies between tasks, and clearly present the sequence and relationships between subtasks; perform critical path analysis based on the PERT-based duration estimation model to screen out the optimal execution path.

[0019] Preferably, the execution optimization layer includes a resource optimization module for:

[0020] Resource monitoring: Relying on the real-time resource monitoring feedback loop, continuously monitor the usage status of the system's computing resources, memory resources, and network resources;

[0021] Dynamic scheduling: Combining a scheduling strategy based on reinforcement learning, according to the priority of tasks, the resource requirements of subtasks, and the current system load, dynamically adjust the resource allocation plan to achieve efficient utilization of resources;

[0022] Batch processing acceleration: Adopting dynamic batch processing technology, reasonably group multiple task requests and process them in batches to reduce the overhead of model inference.

[0023] Preferably, it further includes an intelligent Agent scheduling, execution, and return optimization module:

[0024] Scheduling and execution: The Agent scheduling engine, based on the collaboration strategy generated by the large model module, comprehensively considers the urgency of tasks, resource requirements, and the real-time load status of intelligent Agents, and accurately schedules suitable intelligent Agents to execute tasks; during the execution of tasks by intelligent Agents, obtain task context information from the context management unit, follow the rules in the rule constraint library, use professional capabilities to complete subtasks, and feedback intermediate results to the context management unit;

[0025] Return optimization: The return optimization module, based on the task execution results and real-time feedback information, when it detects that the execution efficiency of subtasks is poor or the results do not meet the standards, starts the optimization mechanism; with the help of intelligent decision-making algorithms, based on the information in the context management unit and the rules in the rule constraint library, dynamically adjust the collaboration strategy of intelligent Agents, re-plan the type, quantity, and collaboration method of Agents, and re-schedule the execution of tasks by the Agent scheduling engine to achieve closed-loop optimization of the process.

[0026] A method for an intelligent Agent process orchestration system driven by a large model includes the following steps:

[0027] Task decomposition based on a large model, through the collaborative operation of the module management layer, task parsing layer, and execution optimization layer, complete the accurate processing and efficient execution planning of complex tasks;

[0028] Agent scheduling and execution, based on the collaboration strategy generated by the large model, accurately schedule intelligent Agents to execute tasks, and obtain context information, follow rule constraints, and feedback intermediate results during the execution process;

[0029] Closed-loop optimization of the return optimization module, based on the task execution results and real-time feedback, conducts closed-loop optimization on the overall process, dynamically adjusts the collaboration strategy, reschedules the intelligent agent to execute tasks, continuously improves the task execution effect, and ensures that the system always operates efficiently and stably;

[0030] Execute resource optimization in the execution optimization layer.

[0031] Preferably, the task decomposition based on the large model specifically includes:

[0032] Build a hierarchical architecture system, including a module management layer, a task parsing layer, and an execution optimization layer;

[0033] In the module management layer, integrate mainstream pre-trained large models, screen the base model through a dynamic evaluation mechanism covering key dimensions such as the number of parameters, domain adaptability, and inference efficiency; build a domain knowledge enhancement mechanism, inject a professional term dictionary and a domain ontology map, and adopt a two-stage fine-tuning strategy. First, preliminarily optimize with the general capabilities of the large model, and then strengthen the design for special tasks to improve the model's performance in specific domains;

[0034] The task parsing layer provides a multi-mode task description input interface, supporting the submission of task descriptions in text, table, and flowchart forms; after receiving the task description, the large model module applies the MECE principle to exhaustively and mutually exclusive decompose the task objectives, constructs a directed acyclic graph to represent the task topology structure, and conducts critical path analysis based on the PERT-based duration estimation model to screen out the optimal execution path.

[0035] Preferably, Agent scheduling and execution specifically include:

[0036] Scheduling: The Agent scheduling engine, based on the collaboration strategy generated by the large model module, comprehensively considers the urgency of the task, resource requirements, and the real-time load status of the intelligent agent, and accurately schedules the appropriate intelligent agent to participate in task execution to ensure scientific and efficient task allocation;

[0037] Execution: During the execution of tasks by the intelligent agent, continuously obtain detailed task context information such as past task execution records, current task preconditions, and user-specific requirement preferences from the context management unit; strictly follow the business rules, safety specifications, and resource limitation rules preset in the rule constraint library, and use its own professional capabilities to complete subtasks; timely feedback partial processed data and preliminary generated solution framework stage intermediate results to the context management unit to ensure the coherence and accuracy of the task process.

[0038] Preferably, the closed-loop optimization of the return optimization module specifically includes:

[0039] Monitoring Trigger: The return optimization module, based on the task execution results and real-time feedback information, quickly activates the optimization mechanism when it monitors that the execution efficiency of a certain subtask is poor, such as the processing time far exceeds the expected threshold, or the execution result does not meet the pre-set quality standards and business goals;

[0040] Strategy Adjustment: With the help of intelligent decision-making algorithms, based on the information stored in the context management unit and the relevant rules in the rule constraint library, dynamically adjust the cooperation strategy of intelligent Agents, and re-plan the types, quantities, and cooperation methods of intelligent Agents participating in the task;

[0041] Rescheduling Execution: The Agent scheduling engine re-schedules intelligent Agents to execute tasks according to the new cooperation strategy, and gradually improves the task execution effect through continuous iterative optimization to ensure the efficient and stable operation of the system.

[0042] Preferably, the resource optimization of the execution optimization layer specifically includes:

[0043] Resource Monitoring: Relying on the real-time resource monitoring feedback loop, continuously monitor the usage status of the system's computing resources, memory resources, and network resources;

[0044] Dynamic Scheduling: Combining the scheduling strategy based on reinforcement learning, dynamically adjust the resource allocation plan according to the priority of the task, the resource requirements of the subtask, and the current system load to achieve efficient resource utilization;

[0045] Batch Processing Acceleration: Adopt dynamic batch processing technology to reasonably group multiple task requests and process them in batches to reduce the model inference overhead;

[0046] Result Output: Present the processing results in a structured form through a standardized output interface for easy user understanding and use.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] The intelligent Agent process orchestration system and method driven by large models proposed in the present invention, relying on the collaborative operation of a hierarchical architecture, deeply explores the core advantages of large models, can efficiently and accurately process various complex tasks, and realizes the overall optimization of the execution process, carefully creating a convenient, intelligent and highly efficient task processing solution for users. The system can flexibly and accurately determine the types and quantities of intelligent Agents participating in task execution according to the specific characteristics, complexity and real-time requirements of tasks, and cleverly plan their cooperation methods, thus showing excellent adaptability to various complex, changeable and cross-domain tasks. This system can combine real-time feedback data to continuously iterate and optimize process strategies and intelligent Agent capability portraits, significantly enhancing the adaptive ability to cross-domain tasks and greatly reducing the dependence on manual configuration. This patent realizes a paradigm shift from static planning to dynamic evolution in complex task processing by constructing a complete closed-loop system of "perception - decision - execution - optimization", and has made breakthrough progress in key indicators such as task processing efficiency, resource utilization efficiency, and system stability, with good promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flowchart of the intelligent Agent process orchestration driven by the large model of the present invention;

[0050] Figure 2 It is a system architecture diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] In order to clearly and completely describe the objectives, technical solutions of the present invention and make the advantages clearer, the following further details the embodiments of the present invention with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are part of the embodiments of the present invention, rather than all of the embodiments, and are only used to explain the embodiments of the present invention, not to limit the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0052] Embodiment 1, please refer to Figures 1 to 2 , the present invention provides a technical solution: an intelligent Agent process orchestration system driven by a large model, as Figure 1 shown, uses a large model to decompose complex tasks into goals and perform logical reasoning, dynamically generates multi-agent Agent cooperation strategies, and realizes the closed-loop optimization of the process by combining preset rules and real-time feedback. Specifically, it includes:

[0053] The first step: Large model task decomposition

[0054] Construct an efficient and intelligent hierarchical architecture system, which is composed of a module management layer, a task parsing layer and an execution optimization layer in cooperation asFigure 2 As shown, it aims to achieve precise processing and efficient execution of complex tasks.

[0055] In the model management layer, the system integrates mainstream pre-trained large models. Through an advanced dynamic evaluation mechanism, the model is comprehensively considered from multiple key dimensions to screen out the base model that best suits the task requirements. These evaluation dimensions cover core indicators such as the number of parameters, domain adaptability, and inference efficiency. On this basis, the system further constructs a domain knowledge enhancement mechanism. By injecting a professional term dictionary and a domain ontology graph, it fully excavates the professional knowledge in the domain and integrates it into the large model, thus significantly improving the model's performance in a specific domain. At the same time, a two-stage fine-tuning strategy is adopted. In the first stage, the general capabilities of the large model are used for preliminary optimization, and in the second stage, intensive design is carried out for special tasks, enabling the model to accurately adapt to various complex task scenarios.

[0056] The task parsing layer provides a multi-mode task description input interface, supporting users to submit task descriptions in various forms, including text, tables, flowcharts, etc., greatly improving the system's compatibility and usability. After receiving the task description, the large model module gives full play to its powerful language understanding and logical reasoning capabilities to conduct in-depth analysis of the task. During the logical reasoning process, the system applies the MECE (Mutually Exclusive and Collectively Exhaustive) principle, which is a core tool for structured thinking and problem analysis, to decompose the task objectives in an exhaustive and mutually exclusive manner to ensure the comprehensiveness and accuracy of the task decomposition. At the same time, a directed acyclic graph (DAG) is constructed to represent the task topology structure, accurately model the dependencies between tasks, and clearly present the sequence and relationships between subtasks. In addition, based on the Program Evaluation and Review Technique (PERT) duration estimation model, a critical path analysis is carried out on the task process to screen out the optimal execution path, effectively improving the task processing efficiency.

[0057] The execution optimization layer introduces a dynamic optimization mechanism, mainly relying on a real-time resource monitoring feedback loop to continuously monitor the usage of the system's computing resources, memory resources, network resources, etc. Combining with a scheduling strategy based on reinforcement learning, according to the priority of tasks, the resource requirements of subtasks, and the current system load, the resource allocation plan is dynamically adjusted to achieve efficient use of resources. To further improve the inference speed, the system adopts dynamic batch processing technology to reasonably group multiple task requests and process them in batches, reducing the overhead of model inference. Finally, the system presents the processing results in a structured form through a standardized output interface, facilitating users' understanding and use.

[0058] Through the collaborative operation of the above hierarchical architecture, the system can give full play to the advantages of the large model, achieve the efficient processing and optimized execution of complex tasks, and provide users with convenient and intelligent task processing solutions.

[0059] Step 2: Agent Scheduling and Execution

[0060] Based on the collaboration strategy generated by the large model module, the Agent scheduling engine accurately schedules and adapts intelligent Agents to execute tasks. During the scheduling process, the urgency of the task, resource requirements, and the real-time load status of the intelligent Agents are comprehensively considered to ensure the scientific and efficient task allocation.

[0061] During the task execution, the intelligent Agents will continuously obtain detailed task context information from the context management unit, including but not limited to past task execution records, preconditions of the current task, and user-specific requirement preferences. At the same time, strictly following the business rules, security specifications, resource limitation rules, etc. preset in the rule constraint library, they use their professional capabilities to methodically complete the subtasks they are responsible for. During the execution process of the subtasks, the intelligent Agents will timely feedback the phased intermediate results, such as partially processed data, initially generated solution frameworks, etc., to the context management unit to provide key references for the subsequent task progress and ensure the coherence and accuracy of the entire task process.

[0062] Step 3: Return Optimization Module:

[0063] Based on the task execution results and real-time feedback information, the return optimization module conducts closed-loop optimization of the overall process. When it is detected that the execution efficiency of a certain subtask is poor, such as the processing time far exceeds the expected threshold, or its execution result does not meet the preset quality standards, business goals, etc., the system will quickly start the optimization mechanism. With the help of intelligent decision-making algorithms, according to the rich information stored in the context management unit and the relevant rules in the rule constraint library, the collaboration strategy of the intelligent Agents is dynamically adjusted. Re-plan the types, quantities, and collaboration methods of the intelligent Agents participating in the task, and then the Agent scheduling engine re-schedules the intelligent Agents to execute the task according to the new collaboration strategy. Through continuous iterative optimization, the task execution effect is gradually improved to ensure that the system always operates efficiently and stably.

[0064] Example 2, on the basis of Example 1, proposes a method for an intelligent Agent process orchestration system driven by a large model, including the following steps:

[0065] Task Decomposition Based on the Large Model:

[0066] Build a layered architecture system, including module management layer, task analysis layer and execution optimization layer;

[0067] At the module management level, mainstream pre-trained large models are integrated, and base models are selected through a dynamic evaluation mechanism covering key dimensions such as parameter magnitude, domain adaptability, and inference efficiency. A domain knowledge enhancement mechanism is built, incorporating a professional terminology dictionary and domain ontology graph. A two-stage fine-tuning strategy is adopted, first leveraging the general capabilities of the large model for initial optimization, and then strengthening the design for specialized tasks to improve the model's performance in specific domains.

[0068] The task parsing layer provides a multi-mode task description input interface, supporting the submission of task descriptions in the form of text, tables, flowcharts, etc.; after receiving the task description, the large model module applies the MECE principle to exhaustively and mutually exclusive decomposition of the task objectives, constructs a directed acyclic graph to represent the task topology structure, and performs critical path analysis based on the PERT duration estimation model to screen out the optimal execution path.

[0069] Agent scheduling and execution methods:

[0070] Scheduling: The Agent scheduling engine uses the collaborative strategy generated by the large model module to comprehensively consider the urgency of the task, resource requirements, and the real-time load status of the intelligent agent. It accurately schedules the appropriate intelligent agent to execute the task, ensuring scientific and efficient task allocation.

[0071] Execution: During task execution, the intelligent agent continuously obtains detailed task context information from the context management unit, including past task execution records, current task prerequisites, and user-specific needs and preferences. It strictly adheres to the pre-set business rules, security regulations, and resource constraints in the rule constraint library and uses its own professional capabilities to complete subtasks. It promptly feeds back intermediate results, such as partially processed data and preliminary solution frameworks, to the context management unit to ensure the consistency and accuracy of the task process.

[0072] Return to the closed-loop optimization of the optimization module:

[0073] Monitoring trigger: Based on task execution results and real-time feedback, the return optimization module quickly initiates the optimization mechanism when it detects poor execution efficiency of a subtask, such as when the processing time far exceeds the expected threshold, or when the execution result does not meet the pre-set quality standards or business goals.

[0074] Strategy adjustment: With the help of intelligent decision-making algorithms, the collaborative strategy of intelligent agents is dynamically adjusted based on the information stored in the context management unit and the relevant rules in the rule constraint library, and the type, number and collaboration mode of intelligent agents participating in the task are re-planned;

[0075] Rescheduled Execution: The Agent scheduling engine reschedules the intelligent Agent to execute tasks according to the new collaboration strategy. Through continuous iterative optimization, the task execution effect is gradually improved to ensure the efficient and stable operation of the system.

[0076] Resource Optimization in the Execution Optimization Layer:

[0077] Resource Monitoring: Relying on the real-time resource monitoring feedback loop, continuously monitor the usage status of the system's computing resources, memory resources, network resources, etc.

[0078] Dynamic Scheduling: Combining the scheduling strategy based on reinforcement learning, dynamically adjust the resource allocation plan according to the priority of tasks, the resource requirements of subtasks, and the current system load to achieve efficient resource utilization.

[0079] Batch Processing Acceleration: Adopt dynamic batch processing technology to reasonably group multiple task requests and process them in batches to reduce the model inference overhead.

[0080] Result Output: Present the processing results in a structured form through a standardized output interface for easy user understanding and use.

[0081] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent Agent process orchestration system driven by a large model, characterized in that: A hierarchical architecture system composed of a module management layer, a task parsing layer, and an execution optimization layer working together: Module management layer: Integrate mainstream pre-trained large models, and select the base model through a dynamic evaluation mechanism covering key dimensions such as the number of parameters, domain adaptability, and inference efficiency; Build a domain knowledge enhancement mechanism, inject a professional term dictionary and a domain ontology map, and adopt a two-stage fine-tuning strategy. First, initially optimize with the general capabilities of the large model, and then strengthen the design for specific tasks to improve the performance of the model in a specific domain; Task parsing layer: Provide a multi-mode task description input interface, supporting the submission of task descriptions in the forms of text, tables, and flowcharts; After receiving the task description, the large model module disassembles the task objective in an exhaustive and mutually exclusive manner using the MECE principle, constructs a directed acyclic graph to represent the task topology structure, and performs critical path analysis based on the PERT-based duration estimation model to select the optimal execution path; Execution optimization layer: Introduce a dynamic optimization mechanism, rely on the real-time resource monitoring feedback loop to monitor the usage status of the system's computing resources, memory resources, and network resources, and combine with a scheduling strategy based on reinforcement learning to dynamically adjust the resource allocation plan according to the task priority, subtask resource requirements, and system load conditions; Use dynamic batch processing technology to batch process task requests and present the processing results in a structured form through a standardized output interface.

2. The intelligent Agent process orchestration system based on large model drive according to claim 1, characterized in that: The module management layer includes a model processing module for: Model screening: Integrate mainstream pre-trained large models, and comprehensively consider the models from multiple key dimensions such as the number of parameters, domain adaptability, and inference efficiency through an advanced dynamic evaluation mechanism to select the base model that best suits the task requirements; Knowledge enhancement: Build a domain knowledge enhancement mechanism, and by injecting a professional term dictionary and a domain ontology map, fully explore the professional knowledge in the domain and integrate it into the large model to significantly improve the performance of the model in a specific domain; Fine-tuning optimization: Adopt a two-stage fine-tuning strategy. In the first stage, initially optimize with the general capabilities of the large model, and in the second stage, carry out intensive design for specific tasks to enable the model to accurately adapt to various complex task scenarios.

3. An intelligent Agent process orchestration system driven by a large model according to claim 2, characterized in that: The task parsing layer includes a task processing module for: Interface input: Provide a multi-mode task description input interface, supporting users to submit task descriptions in various forms such as text, tables, and flowcharts to improve the compatibility and usability of the system; Logical disassembly: After receiving the task description, the large model module uses its powerful language understanding and logical reasoning capabilities to disassemble the task objective in an exhaustive and mutually exclusive manner using the MECE principle to ensure the comprehensiveness and accuracy of the task decomposition; Structure modeling and analysis: Construct a directed acyclic graph to represent the task topology structure, accurately model the dependencies between tasks, and clearly present the sequence and association between each subtask; Perform critical path analysis based on the PERT-based duration estimation model to select the optimal execution path.

4. An intelligent Agent process orchestration system driven by a large model according to claim 3, characterized in that: The execution optimization layer includes a resource optimization module for: Resource monitoring: Rely on the real-time resource monitoring feedback loop to continuously monitor the usage status of the system's computing resources, memory resources, and network resources; Dynamic scheduling: Combine a scheduling strategy based on reinforcement learning. According to the priorities of tasks, the resource requirements of subtasks, and the current system load, dynamically adjust the resource allocation plan to achieve efficient utilization of resources; Batch processing acceleration: Adopt dynamic batch processing technology to reasonably group multiple task requests and process them in batches to reduce the overhead of model inference.

5. An intelligent Agent process orchestration system based on large model drive according to claim 4, characterized in that: It also includes an intelligent Agent scheduling, execution, and return optimization module: Scheduling and execution: The Agent scheduling engine, based on the collaboration strategy generated by the large model module, comprehensively considers the urgency of tasks, resource requirements, and the real-time load status of intelligent Agents, and accurately schedules and adapts intelligent Agents to execute tasks; During the execution of tasks by intelligent Agents, obtain task context information from the context management unit, follow the rules in the rule constraint library, use professional capabilities to complete subtasks, and feedback intermediate results to the context management unit; Return optimization: The return optimization module, based on the task execution results and real-time feedback information, activates the optimization mechanism when it detects that the execution efficiency of subtasks is poor or the results do not meet the standards; With the help of intelligent decision-making algorithms, based on the information of the context management unit and the rules of the rule constraint library, dynamically adjust the collaboration strategy of intelligent Agents, re-plan the types, quantities, and collaboration methods of Agents, and re-schedule and execute tasks by the Agent scheduling engine to achieve closed-loop optimization of the process.

6. A method for an intelligent Agent process orchestration system driven by a large model according to claim 5, characterized in that: It includes the following steps: Task decomposition based on a large model. Through the coordinated operation of the module management layer, task parsing layer, and execution optimization layer, complete the precise processing and efficient execution planning of complex tasks; Agent scheduling and execution. Based on the collaboration strategy generated by the large model, accurately schedule intelligent Agents to execute tasks, and obtain context information, follow rules, and feedback intermediate results during the execution process; Closed-loop optimization of the return optimization module. Based on the task execution results and real-time feedback, conduct closed-loop optimization of the overall process, dynamically adjust the collaboration strategy, re-schedule intelligent Agents to execute tasks, continuously improve the task execution effect, and ensure that the system always operates efficiently and stably; Resource optimization of the execution optimization layer.

7. A method according to claim 6, characterized in that: The task decomposition based on a large model specifically includes: Build a hierarchical architecture system, including a module management layer, a task parsing layer, and an execution optimization layer; In the module management layer, integrate mainstream pre-trained large models, and select the base model through a dynamic evaluation mechanism covering key dimensions such as the number of parameters, domain adaptability, and inference efficiency; build a domain knowledge enhancement mechanism, inject a professional term dictionary and a domain ontology map, and adopt a two-stage fine-tuning strategy. First, preliminarily optimize with the general capabilities of the large model, and then strengthen the design for special tasks to improve the performance of the model in a specific domain; The task parsing layer provides a multi-mode task description input interface, supporting the submission of task descriptions in the form of text, tables, and flowcharts; after receiving the task description, the large model module applies the MECE principle to exhaustively and mutually exclusive decompose the task objectives, constructs a directed acyclic graph to represent the task topological structure, and conducts critical path analysis based on the PERT-based duration estimation model to select the optimal execution path.

8. A method according to claim 7, wherein: Agent scheduling and execution specifically includes: Scheduling: Based on the collaboration strategy generated by the large model module, the Agent scheduling engine comprehensively considers the urgency of tasks, resource requirements, and the real-time load status of intelligent Agents, and precisely schedules the appropriate intelligent Agents to execute tasks, ensuring scientific and efficient task allocation; Execution: During the execution of tasks by intelligent Agents, they continuously obtain detailed task context information such as past task execution records, current task preconditions, and user-specific requirement preferences from the context management unit; strictly follow the business rules, safety specifications, and resource limitation rules preset in the rule constraint library, and use their professional capabilities to complete subtasks; timely feedback some processed data and preliminary solution framework as phased intermediate results to the context management unit to ensure the coherence and accuracy of the task process.

9. A method according to claim 8, wherein: The closed-loop optimization of the return optimization module specifically includes: Monitoring trigger: Based on the task execution results and real-time feedback information, when the return optimization module detects that the execution efficiency of a certain subtask is poor, such as the processing time far exceeds the expected threshold, or the execution result does not meet the preset quality standards and business goals, it quickly activates the optimization mechanism; Strategy adjustment: With the help of intelligent decision-making algorithms, according to the information stored in the context management unit and relevant rules in the rule constraint library, dynamically adjust the collaboration strategy of intelligent Agents, and re-plan the types, quantities, and collaboration methods of intelligent Agents participating in tasks; Re-scheduling and execution: The Agent scheduling engine re-schedules intelligent Agents to execute tasks according to the new collaboration strategy, and gradually improves the task execution effect through continuous iterative optimization to ensure the efficient and stable operation of the system.

10. A method according to claim 9, characterized in that: The resource optimization of the execution optimization layer specifically includes: Resource monitoring: Relying on the real-time resource monitoring feedback loop, continuously monitor the usage status of the system's computing resources, memory resources, and network resources; Dynamic scheduling: Combining the scheduling strategy based on reinforcement learning, according to the priority of tasks, the resource requirements of subtasks, and the current system load, dynamically adjust the resource allocation plan to achieve efficient resource utilization; Batch processing acceleration: Adopt dynamic batch processing technology to reasonably group multiple task requests and process them in batches to reduce the model inference overhead; Result output: Present the processing results in a structured form through a standardized output interface for easy user understanding and use.

Images