Large model driven enterprise management hub multi-agent business empowerment system

Abstract

This invention discloses a large-model-driven multi-agent business empowerment system for enterprise management, comprising an enterprise data access and semantic fusion module, a large-model business understanding and task generation module, a multi-agent dynamic collaborative scheduling module, a business execution and state awareness module, a business closed-loop feedback and self-learning optimization module, and a security audit and access control module. Through unified business semantic fusion, large-model business understanding, multi-agent dynamic collaboration, and closed-loop self-learning optimization mechanisms, this invention achieves unified management of multi-source business data, intelligent collaborative execution of complex business tasks, and continuous optimization of system strategies. It effectively solves the problems of inconsistent business semantics, insufficient agent collaboration capabilities, and lack of dynamic learning mechanisms in existing technologies, thereby improving enterprise business processing efficiency, resource collaboration capabilities, and system intelligence.

Description

Technical Field

[0001] This invention relates to the field of enterprise management technology, and in particular to a large-model-driven multi-agent business empowerment system for enterprise management. Background Technology

[0002] As enterprises continue to advance their digital transformation, they are gradually deploying various business systems such as ERP, CRM, MES, SCM, OA, and financial management. The scale of data and the complexity of business operations are constantly increasing. While existing enterprise management systems can automate some processes and perform data analysis, they still face the following challenges in complex business scenarios: In existing multi-agent business systems, each agent typically operates independently using fixed logic, lacking a unified task scheduling and dynamic collaboration mechanism. In cross-departmental collaboration scenarios, this can easily lead to problems such as task duplication, resource conflicts, and execution link breaks. There are significant differences in data formats, business semantics, and interface protocols between different business systems. Existing systems lack unified semantic modeling and knowledge fusion capabilities, making it difficult for intelligent agents to accurately understand the business context of enterprises. Most existing systems operate based on static rules or fixed models, lacking a continuous learning and optimization mechanism based on business feedback, making it difficult to achieve dynamic strategy adjustments and intelligent capability evolution according to changes in the enterprise's operating status. Based on the above, this application proposes a large-scale model-driven multi-agent business empowerment system for enterprise management. Summary of the Invention

[0003] Based on the technical problems existing in the background technology, this invention proposes a large model-driven multi-agent business empowerment system for enterprise management.

[0004] The large-model-driven enterprise management center multi-agent business empowerment system proposed in this invention includes an enterprise data access and semantic fusion module, a large-model business understanding and task generation module, a multi-agent dynamic collaborative scheduling module, a business execution and state perception module, a business closed-loop feedback and self-learning optimization module, and a security audit and access control module. The enterprise data access and semantic fusion module is used to access enterprise ERP system, CRM system, MES system, OA system, financial system and external business data, and to perform standardized processing, semantic parsing and knowledge graph construction on multi-source heterogeneous data to generate a unified business semantic model. The large model business understanding and task generation module is used to perform semantic reasoning on enterprise business requirements, user instructions and business context based on the large language model, and automatically generate business objectives, task execution chains and intelligent agent invocation strategies. The multi-agent dynamic collaborative scheduling module is used to dynamically allocate tasks, collaboratively orchestrate, and optimize execution paths for multiple business agents based on task priority, resource occupancy status, and business dependencies. The business execution and status awareness module is used to drive each business agent to perform corresponding business operations and collect task execution status, resource operation status and business feedback information in real time. The business closed-loop feedback and self-learning optimization module is used to continuously optimize the large model parameters, task strategies and intelligent agent collaboration rules based on task execution results, user feedback and business performance indicators, so as to realize the system's self-learning evolution. The security audit and access control module is used to perform permission verification, log auditing, and risk detection on intelligent agent invocation behavior, data access behavior, and task execution process.

[0005] Preferably, the enterprise data access and semantic fusion module includes a multi-source data access unit, a data standardization processing unit, a semantic parsing unit, a knowledge graph construction unit, and a unified business semantic modeling unit; The multi-source data access unit is used to access data from ERP, CRM, MES, OA, financial systems, and external Internet data; the data standardization processing unit is used to perform field mapping, format conversion, and outlier cleaning on data of different formats; the semantic parsing unit is used to extract semantic information of business entities, business relationships, and business events using a large model; the knowledge graph construction unit is used to establish a graph of enterprise organization, processes, resources, and business relationships; and the unified business semantic modeling unit is used to generate cross-system unified business semantic vectors and business context models. The operational logic steps of the enterprise data access and semantic fusion module are as follows: S101: Accesses structured and unstructured business data from ERP, CRM, MES, SCM, OA, financial systems, and external internet business platforms via a multi-source data access unit, and establishes a unified data collection queue. The enterprise business data set is represented as follows: ; in Represents a collection of enterprise business data. This represents the nth type of business data; S102: The data standardization processing unit performs field mapping, format unification, missing value repair, and outlier cleaning on the incoming data to generate a standardized business dataset. S103: The semantic parsing unit performs business entity recognition, event extraction, and business relationship parsing on standardized business data based on a large language model, generating a business semantic feature vector. ; in This represents the semantic feature vector corresponding to the i-th type of business data. This represents the semantic mapping function of a large language model. Indicates business data content, Indicates business context information; S104: The knowledge graph construction unit constructs an enterprise knowledge graph based on business entities and business relationships, which is used to describe the enterprise's organizational structure, business processes, and resource relationships. Enterprise knowledge graph representation is as follows: ; in Represents the enterprise knowledge graph. Represents a set of business entity nodes. It represents a set of business relationships, and establishes the association mapping relationship between enterprise business objects through an enterprise knowledge graph; S105: The unified business semantic modeling unit generates a unified business semantic model based on business semantic feature vectors and enterprise knowledge graphs. ; in Represents a unified business semantic model. This represents a semantic fusion function. Represents a set of business semantic vectors. Represents an enterprise knowledge graph; S106: The system dynamically updates the unified business semantic model based on real-time business status, task execution results, and changes in the external environment, thereby ensuring that the business semantic model remains consistent with the actual operating status of the enterprise. The update formula is as follows: ; in This represents the business semantic model at the current moment. This represents the updated business semantic model. Indicates the amount of change in business semantics. This represents the update coefficient.

[0006] Preferably, the large model business understanding and task generation module includes a business instruction parsing unit, a context awareness unit, a task chain generation unit, an agent invocation strategy generation unit, and a task priority evaluation unit. The business instruction parsing unit is used to parse the business requirements, management objectives, and task intentions input by the user; the context awareness unit is used to generate contextual semantics by combining the enterprise's real-time operating status, historical tasks, and business environment information; the task chain generation unit is used to generate a business execution step chain based on large model reasoning; the agent invocation strategy generation unit is used to determine the agent invocation order, collaborative relationship, and resource allocation scheme; and the task priority evaluation unit is used to generate task priorities based on the business urgency, resource occupancy rate, and scope of impact. The operational logic steps of the large-scale model business understanding and task generation module are as follows: S201: The business instruction parsing unit receives user input of business requirements, operational objectives, and management instructions, and performs semantic parsing using a unified business semantic model to generate a business requirement feature vector. ; in Represents a semantic vector of business requirements. This represents a function for parsing business requirements. This indicates that the user has entered a business command. Represents a unified business semantic model; S202: The context-aware unit combines the enterprise's real-time operating status, historical task data, and business environment information to generate business context semantic information, thereby enhancing the large model's ability to understand business scenarios. The business context model is represented as: ; in Represents the business context model. This indicates a context fusion function. Represents a semantic vector of business requirements. Represents historical business data. This represents real-time operating environment data; S203: The task chain generation unit generates business task execution chains based on business requirement semantics and context models, using a large language model. ; in Indicates the business task chain, This represents the nth business subtask; S204: The agent invocation strategy generation unit generates an agent collaborative invocation strategy based on the business task chain, agent capability tags and system resource status. S205: The task priority evaluation unit generates task priority parameters based on the urgency of the business, the resource occupancy status, and the degree of business impact, and outputs them to the multi-agent dynamic collaborative scheduling module. The task priority function is expressed as: ; in Indicates the priority of the i-th task. Indicates the urgency of the business. Indicates resource utilization rate. Indicates the degree of business impact. , , This represents the weighting parameter.

[0007] Preferably, the multi-agent dynamic collaborative scheduling module includes an agent registration and management unit, a task decomposition unit, a dynamic scheduling unit, a conflict detection unit, a collaborative communication unit, and an execution path optimization unit; The agent registration and management unit is used to maintain the capability tags, running status, and service interfaces of each agent; the task decomposition unit is used to break down complex business tasks into multiple executable subtasks; the dynamic scheduling unit is used to dynamically allocate agents according to resource status and task dependencies; the conflict detection unit is used to identify task resource conflicts, execution path conflicts, and priority conflicts; the cooperative communication unit is used to realize context sharing and state synchronization among multiple agents; and the execution path optimization unit is used to optimize the task execution order based on historical execution results. The operational logic steps of the multi-agent dynamic cooperative scheduling module are as follows: S301: The task decomposition unit breaks down the business task chain into multiple sets of independently executable subtasks. ; in Represents a set of business subtasks. This represents the nth business subtask; S302: The agent registration and management unit matches the agent's capability tags with the task requirements to determine the target agent to execute; The formula for the agent matching function is: ; in Representation and Task Matching target agent, This indicates the degree of matching between the task and the agent. Represents the j-th agent; S303: The dynamic scheduling unit schedules tasks based on task priority, resource utilization, and task dependencies, where task scheduling priority is expressed as follows: ; in Indicates task priority. Indicates the urgency of the task. Indicates resource utilization rate. Indicates the degree of task dependency. , , Indicates the weighting parameter; S304: The conflict detection unit identifies task resource conflicts and execution path conflicts, and realizes context sharing between agents through the cooperative communication unit. The formula for calculating the task conflict degree is: ; in Indicates the degree of task conflict. , This represents a set of task resources. When the conflict level exceeds a threshold, the system automatically adjusts the task execution path. S305: The execution path optimization unit optimizes the task execution order based on the historical task execution results and outputs the final scheduling scheme.

[0008] Preferably, the business execution and status awareness module includes a business execution control unit, a status acquisition unit, an anomaly identification unit, and a business feedback generation unit; The business execution control unit is used to control each intelligent agent to perform corresponding business operations; the status acquisition unit is used to collect task execution status, device status and business indicators; the anomaly identification unit is used to identify task failure, process blockage and abnormal fluctuations; the business feedback generation unit is used to generate business execution results and feedback reports. The operational logic steps of the business execution and status awareness module are as follows: S401: The service execution control unit drives the corresponding intelligent agent to execute service tasks according to the scheduling scheme. The task execution status is represented as follows: ; in Indicates the task execution status. Indicates business tasks. Indicates the executing intelligent agent; S402: The status acquisition unit collects business operation indicators, resource status, and task execution status in real time to form a business status vector. ; in Indicates the CPU resource status. Indicates memory status. Indicates network status. Indicates the operational status of the service; S403: The anomaly identification unit identifies abnormal task behaviors based on changes in business status. The formula for determining abnormal task behaviors is as follows: ; in This indicates the result of the anomaly assessment. Indicates the current state value. Represents the state mean. Indicates the standard deviation of the state. Indicates the anomaly coefficient; S404: The business feedback generation unit generates a business feedback report based on the task execution results. This report includes the task completion rate, calculated using the following formula: ; in Indicates the task completion rate. Indicates the number of successful tasks. This indicates the total number of tasks.

[0009] Preferably, the business closed-loop feedback and self-learning optimization module includes a task result evaluation unit, a user feedback analysis unit, a reinforcement learning optimization unit, a model parameter update unit, and an experience knowledge accumulation unit; The task result evaluation unit is used to analyze task completion rate, execution efficiency, and business benefits; the user feedback analysis unit is used to analyze user operation feedback, satisfaction, and suggestions for improvement; the reinforcement learning optimization unit is used to optimize the agent collaboration strategy based on the reward mechanism; the model parameter update unit is used to dynamically update the large model prompt word parameters or business rule parameters; and the experience knowledge accumulation unit is used to build an enterprise business experience library and historical case library. The operational logic steps of the business closed-loop feedback and self-learning optimization module are as follows: S501: The task result evaluation unit comprehensively evaluates task execution efficiency, resource consumption, and business benefits. The comprehensive evaluation expression is as follows: ; in This represents the overall evaluation value of the task. Indicates execution efficiency. Indicates business revenue. Indicates resource consumption. , , Indicates the weighting parameter; S502: The user feedback analysis unit analyzes user suggestions for improvement and satisfaction information, where the expression for user satisfaction analysis is: ; in Indicates user satisfaction. Indicates user ratings, Indicates the number of feedback responses; S503: The reinforcement learning optimization unit generates reward values ​​based on the task execution results and optimizes the agent's cooperative strategy. Its reward function formula is: ; in Indicates the reward value. Indicates the task completion rate. , , Indicates the weighting coefficient; S504: The model parameter update unit dynamically updates the large model parameters and scheduling rules based on the reward value. The formula used in the parameter update process is as follows: ; in Indicates the current model parameters. This indicates the updated model parameters. Indicates the learning rate. This indicates the reward gradient.

[0010] Preferably, the security audit and access control module includes an identity authentication unit, an access control unit, a behavior audit unit, a risk detection unit, and a data desensitization unit; The identity authentication unit is used to verify the identity of users and intelligent agents; the access control unit is used to control the data access permissions of different intelligent agents; the behavior audit unit is used to record the intelligent agent's calling behavior and task execution logs; the risk detection unit is used to identify the risks of abnormal calls, unauthorized access and sensitive data leakage; and the data desensitization unit is used to encrypt and desensitize sensitive business data. The operational logic steps of the security audit and access control module are as follows: S601: The identity authentication unit receives user login requests and intelligent agent invocation requests, performs joint verification of user identity information, device information, access token, and intelligent agent identity identifier, and generates an identity credibility score based on historical access behavior to complete the identity authentication of the user and the intelligent agent. The identity credibility calculation formula is as follows: ; in Indicates the credibility of the identity. Indicates identity information, Indicates the authentication key. Indicates an access token; S602: The access control unit generates dynamic access control policies based on user roles, business positions, data sensitivity levels, and intelligent agent capability tags, and performs permission verification for business data access, task invocation, and intelligent agent collaborative behavior; S603: The behavior auditing unit records the entire process of user operation behavior, intelligent agent invocation behavior, data access behavior, and business task execution, and establishes a traceable audit chain. The audit log set is represented as follows: ; Where L represents the audit log set. This represents the nth audit record. Each audit record includes at least the user's identity information, agent identifier, data access object, operation time, operation type, task execution result, and risk level information. When the system detects abnormal call chains, frequent access to sensitive resources, or abnormal task jump behavior, it automatically marks the corresponding audit log as high-risk. S604: The risk detection unit establishes a risk identification model based on user behavior characteristics, agent call frequency, data access patterns, and historical security events. It performs real-time detection of abnormal access behavior, unauthorized operation behavior, and potential data leakage risks. Its risk identification model uses a risk scoring function for detection, and the function expression is as follows: ; in Indicates risk score, Indicates risk factors, The risk weights are indicated by risk factors, which include high-frequency abnormal access, illegal interface calls, abnormal data downloads, unauthorized collaboration of intelligent agents, cross-domain transmission of sensitive data, and abnormal access behavior. When the risk score exceeds the preset security threshold, the system automatically performs the following actions: interrupting the execution of the current task, freezing the agent's calling permissions, restricting access to sensitive data, triggering security alarms, and pushing abnormal behavior to the security audit center. The abnormal risk behavior is also synchronized to the business closed-loop feedback and self-learning optimization module for subsequent dynamic optimization of security policies and continuous learning of risk models.

[0011] Preferably, the enterprise data access and semantic fusion module, the large model business understanding and task generation module, the multi-agent dynamic collaborative scheduling module, the business execution and state perception module, the business closed-loop feedback and self-learning optimization module, and the security audit and access control module form a closed-loop collaborative connection relationship. The enterprise data access and semantic fusion module is connected to the large model business understanding and task generation module, and is used to output a unified business semantic model, enterprise knowledge graph and business context information to the large model business understanding and task generation module. The large model business understanding and task generation module is connected to the multi-agent dynamic collaborative scheduling module and is used to output the business task chain, agent calling strategy and task priority parameters to the multi-agent dynamic collaborative scheduling module. The multi-agent dynamic collaborative scheduling module is connected to the business execution and state awareness module, and is used to output task scheduling schemes, agent execution paths and collaborative control parameters to the business execution and state awareness module; The business execution and status awareness module is connected to the business closed-loop feedback and self-learning optimization module, and is used to output task execution results, resource operation status, business feedback information and abnormal behavior information to the business closed-loop feedback and self-learning optimization module. The business closed-loop feedback and self-learning optimization module is connected to the large model business understanding and task generation module, and is used to feed back model optimization parameters, task optimization strategies and historical business experience information to the large model business understanding and task generation module, so as to realize the dynamic optimization of the large model reasoning ability and agent collaboration strategy. The security audit and access control module is bidirectionally connected to the enterprise data access and semantic fusion module, the large model business understanding and task generation module, the multi-agent dynamic collaborative scheduling module, the business execution and state perception module, and the business closed-loop feedback and self-learning optimization module. It is used to perform access verification, behavior auditing and risk control on the data access behavior, agent calling behavior and task execution process between the modules, thereby forming a security control link that runs through the entire system.

[0012] Compared with existing technologies, the beneficial effects of this invention are: By constructing a unified business semantic model, a large-scale business understanding mechanism, and an enterprise knowledge graph fusion mechanism, unified access and semantic fusion of multi-source heterogeneous business data from ERP, CRM, MES, SCM, OA, and financial systems have been achieved. This solves the problems of large differences in data formats between different business systems, inconsistent business semantics, and the difficulty for intelligent agents to accurately understand the enterprise business context in existing technologies, thereby improving the data collaboration capability and business understanding accuracy of enterprises in complex business scenarios. By establishing a multi-agent dynamic collaborative scheduling mechanism, an automatic task chain generation mechanism, and an agent collaborative communication mechanism, the automatic decomposition, dynamic task allocation, and execution path optimization of complex business tasks are realized. This solves the problems of independent operation of agents, insufficient cross-departmental collaboration capabilities, repeated task execution, and frequent resource conflicts in existing technologies, thereby improving the efficiency of enterprise business process collaboration, resource utilization, and the stability of complex task execution. By constructing a business closed-loop feedback and self-learning optimization mechanism, and combining task execution results, user feedback information and business performance indicators, the parameters of the large model, task scheduling strategy and agent collaboration rules are continuously optimized. This solves the problems of existing technologies, such as the system's reliance on static rules, lack of dynamic learning ability and difficulty in adapting to changes in the business status of enterprises. As a result, the system's self-learning ability, dynamic decision-making ability and long-term business adaptability are improved. This invention achieves unified management of multi-source business data, intelligent collaborative execution of complex business tasks, and continuous optimization of system strategies by unifying business semantic fusion, large-scale model business understanding, dynamic collaboration of multiple agents, and closed-loop self-learning optimization mechanism. It effectively solves the problems of inconsistent business semantics, insufficient agent collaboration capabilities, and lack of dynamic learning mechanisms in existing technologies, thereby improving enterprise business processing efficiency, resource collaboration capabilities, and system intelligence level. Attached Figure Description

[0013] Figure 1 This is a block diagram of the large-model-driven enterprise management hub multi-agent business empowerment system proposed in this invention. Figure 2 The flowchart shows the operation of the large-model-driven multi-agent business empowerment system for enterprise management centralization proposed in this invention. Detailed Implementation

[0014] The present invention will be further explained below with reference to specific embodiments. Example

[0015] Reference Figure 1-2 This embodiment proposes a large model-driven enterprise management hub multi-agent business empowerment system, including an enterprise data access and semantic fusion module, a large model business understanding and task generation module, a multi-agent dynamic collaborative scheduling module, a business execution and state perception module, a business closed-loop feedback and self-learning optimization module, and a security audit and access control module. The enterprise data access and semantic fusion module is used to access enterprise ERP system, CRM system, MES system, OA system, financial system and external business data, and to perform standardized processing, semantic parsing and knowledge graph construction on multi-source heterogeneous data to generate a unified business semantic model; The enterprise data access and semantic fusion module includes a multi-source data access unit, a data standardization processing unit, a semantic parsing unit, a knowledge graph construction unit, and a unified business semantic modeling unit. The multi-source data access unit is used to access data from ERP, CRM, MES, OA, financial systems, and external Internet data; the data standardization processing unit is used to perform field mapping, format conversion, and outlier cleaning for data of different formats; the semantic parsing unit is used to extract semantic information of business entities, business relationships, and business events using a large model; the knowledge graph construction unit is used to build a graph of enterprise organization, processes, resources, and business relationships; and the unified business semantic modeling unit is used to generate cross-system unified business semantic vectors and business context models. The large-scale model business understanding and task generation module is used to perform semantic reasoning on enterprise business requirements, user instructions and business context based on the large language model, and automatically generate business objectives, task execution chains and intelligent agent invocation strategies. The large model business understanding and task generation module includes a business instruction parsing unit, a context awareness unit, a task chain generation unit, an agent invocation strategy generation unit, and a task priority evaluation unit. The business instruction parsing unit is used to parse the business requirements, management objectives, and task intentions input by the user; the context awareness unit is used to generate contextual semantics by combining the enterprise's real-time operating status, historical tasks, and business environment information; the task chain generation unit is used to generate a business execution step chain based on large model reasoning; the agent invocation strategy generation unit is used to determine the agent invocation order, collaborative relationship, and resource allocation scheme; and the task priority evaluation unit is used to generate task priorities based on business urgency, resource utilization, and scope of impact. The multi-agent dynamic collaborative scheduling module is used to dynamically allocate tasks, coordinate and orchestrate, and optimize execution paths for multiple business agents based on task priority, resource occupancy status, and business dependencies. The multi-agent dynamic cooperative scheduling module includes an agent registration and management unit, a task decomposition unit, a dynamic scheduling unit, a conflict detection unit, a cooperative communication unit, and an execution path optimization unit; The agent registration and management unit maintains the capability tags, running status, and service interfaces of each agent; the task decomposition unit breaks down complex business tasks into multiple executable subtasks; the dynamic scheduling unit dynamically allocates agents based on resource status and task dependencies; the conflict detection unit identifies task resource conflicts, execution path conflicts, and priority conflicts; the cooperative communication unit enables context sharing and state synchronization among multiple agents; and the execution path optimization unit optimizes the task execution order based on historical execution results. The business execution and status awareness module is used to drive each business agent to perform corresponding business operations and collect task execution status, resource operation status and business feedback information in real time. The business execution and status awareness module includes a business execution control unit, a status acquisition unit, an anomaly identification unit, and a business feedback generation unit; The business execution control unit is used to control each intelligent agent to perform corresponding business operations; the status acquisition unit is used to collect task execution status, device status and business indicators; the anomaly identification unit is used to identify task failures, process blockages and abnormal fluctuations; the business feedback generation unit is used to generate business execution results and feedback reports. The business closed-loop feedback and self-learning optimization module is used to continuously optimize the parameters of the large model, task strategies and agent collaboration rules based on task execution results, user feedback and business performance indicators, so as to realize the system's self-learning evolution. The business closed-loop feedback and self-learning optimization module includes a task result evaluation unit, a user feedback analysis unit, a reinforcement learning optimization unit, a model parameter update unit, and an experience knowledge accumulation unit. The task result evaluation unit is used to analyze task completion rate, execution efficiency, and business benefits; the user feedback analysis unit is used to analyze user operation feedback, satisfaction, and suggestions for improvement; the reinforcement learning optimization unit is used to optimize the intelligent agent's collaborative strategy based on the reward mechanism; the model parameter update unit is used to dynamically update the large model's prompt word parameters or business rule parameters; and the experience knowledge accumulation unit is used to build an enterprise business experience library and historical case library. The security audit and access control module is used to perform permission verification, log auditing, and risk detection on intelligent agent calling behavior, data access behavior, and task execution process; The security audit and access control module includes an identity authentication unit, an access control unit, a behavior audit unit, a risk detection unit, and a data anonymization unit; The identity authentication unit is used to verify the identity of users and intelligent agents; the access control unit is used to control the data access permissions of different intelligent agents; the behavior audit unit is used to record the intelligent agent's calling behavior and task execution logs; the risk detection unit is used to identify the risks of abnormal calls, unauthorized access, and sensitive data leakage; and the data desensitization unit is used to encrypt and desensitize sensitive business data. The enterprise data access and semantic fusion module, the large model business understanding and task generation module, the multi-agent dynamic collaborative scheduling module, the business execution and state perception module, the business closed-loop feedback and self-learning optimization module, and the security audit and access control module form a closed-loop collaborative connection relationship. The enterprise data access and semantic fusion module is connected to the large model business understanding and task generation module, and is used to output a unified business semantic model, enterprise knowledge graph and business context information to the large model business understanding and task generation module; The large model business understanding and task generation module is connected to the multi-agent dynamic collaborative scheduling module, and is used to output the business task chain, agent calling strategy and task priority parameters to the multi-agent dynamic collaborative scheduling module. The multi-agent dynamic collaborative scheduling module is connected to the business execution and state awareness module, and is used to output task scheduling schemes, agent execution paths and collaborative control parameters to the business execution and state awareness module; The business execution and status awareness module is connected to the business closed-loop feedback and self-learning optimization module, and is used to output task execution results, resource operation status, business feedback information and abnormal behavior information to the business closed-loop feedback and self-learning optimization module. The business closed-loop feedback and self-learning optimization module is connected to the large model business understanding and task generation module. It is used to feed back model optimization parameters, task optimization strategies and historical business experience information to the large model business understanding and task generation module, so as to realize the dynamic optimization of the large model's reasoning ability and agent collaboration strategy. The security audit and access control module is bidirectionally connected to the enterprise data access and semantic fusion module, the large model business understanding and task generation module, the multi-agent dynamic collaborative scheduling module, the business execution and state perception module, and the business closed-loop feedback and self-learning optimization module. It is used to perform permission verification, behavior auditing and risk control on the data access behavior, agent calling behavior and task execution process between the modules, thereby forming a security control link that runs through the entire system.

[0016] In this embodiment, the operational logic steps of the enterprise data access and semantic fusion module are as follows: S101: Accesses structured and unstructured business data from ERP, CRM, MES, SCM, OA, financial systems, and external internet business platforms via a multi-source data access unit, and establishes a unified data collection queue. The enterprise business data set is represented as follows: ; in Represents a collection of enterprise business data. This represents the nth type of business data; S102: The data standardization processing unit performs field mapping, format unification, missing value repair, and outlier cleaning on the incoming data to generate a standardized business dataset. S103: The semantic parsing unit performs business entity recognition, event extraction, and business relationship parsing on standardized business data based on a large language model, generating a business semantic feature vector. ; in This represents the semantic feature vector corresponding to the i-th type of business data. This represents the semantic mapping function of a large language model. Indicates business data content, Indicates business context information; S104: The knowledge graph construction unit constructs an enterprise knowledge graph based on business entities and business relationships, which is used to describe the enterprise's organizational structure, business processes, and resource relationships. Enterprise knowledge graph representation is as follows: ; in Represents the enterprise knowledge graph. Represents a set of business entity nodes. It represents a set of business relationships, and establishes the association mapping relationship between enterprise business objects through an enterprise knowledge graph; S105: The unified business semantic modeling unit generates a unified business semantic model based on business semantic feature vectors and enterprise knowledge graphs. ; in Represents a unified business semantic model. This represents a semantic fusion function. Represents a set of business semantic vectors. Represents an enterprise knowledge graph; S106: The system dynamically updates the unified business semantic model based on real-time business status, task execution results, and changes in the external environment, thereby ensuring that the business semantic model remains consistent with the actual operating status of the enterprise. The update formula is as follows: ; in This represents the business semantic model at the current moment. This represents the updated business semantic model. Indicates the amount of change in business semantics. This represents the update coefficient.

[0017] In this embodiment, the operational logic steps of the large model business understanding and task generation module are as follows: S201: The business instruction parsing unit receives user input of business requirements, operational objectives, and management instructions, and performs semantic parsing using a unified business semantic model to generate a business requirement feature vector. ; in Represents a semantic vector of business requirements. This represents a function for parsing business requirements. This indicates that the user has entered a business command. Represents a unified business semantic model; S202: The context-aware unit combines the enterprise's real-time operating status, historical task data, and business environment information to generate business context semantic information, thereby enhancing the large model's ability to understand business scenarios. The business context model is represented as: ; in Represents the business context model. This indicates a context fusion function. Represents a semantic vector of business requirements. Represents historical business data. This represents real-time operating environment data; S203: The task chain generation unit generates business task execution chains based on business requirement semantics and context models, using a large language model. ; in Indicates the business task chain, This represents the nth business subtask; S204: The agent invocation strategy generation unit generates an agent collaborative invocation strategy based on the business task chain, agent capability tags and system resource status. S205: The task priority evaluation unit generates task priority parameters based on the urgency of the business, the resource occupancy status, and the degree of business impact, and outputs them to the multi-agent dynamic collaborative scheduling module. The task priority function is expressed as: ; in Indicates the priority of the i-th task. Indicates the urgency of the business. Indicates resource utilization rate. Indicates the degree of business impact. , , This represents the weighting parameter.

[0018] In this embodiment, the operational logic steps of the multi-agent dynamic collaborative scheduling module are as follows: S301: The task decomposition unit breaks down the business task chain into multiple sets of independently executable subtasks. ; in Represents a set of business subtasks. This represents the nth business subtask; S302: The agent registration and management unit matches the agent's capability tags with the task requirements to determine the target agent to execute; The formula for the agent matching function is: ; in Representation and Task Matching target agent, This indicates the degree of matching between the task and the agent. Represents the j-th agent; S303: The dynamic scheduling unit schedules tasks based on task priority, resource utilization, and task dependencies, where task scheduling priority is expressed as follows: ; in Indicates task priority. Indicates the urgency of the task. Indicates resource utilization rate. Indicates the degree of task dependency. , , Indicates the weighting parameter; S304: The conflict detection unit identifies task resource conflicts and execution path conflicts, and realizes context sharing between agents through the cooperative communication unit. The formula for calculating the task conflict degree is: ; in Indicates the degree of task conflict. , This represents a set of task resources. When the conflict level exceeds a threshold, the system automatically adjusts the task execution path. S305: The execution path optimization unit optimizes the task execution order based on the historical task execution results and outputs the final scheduling scheme.

[0019] In this embodiment, the operational logic steps of the business execution and state awareness module are as follows: S401: The service execution control unit drives the corresponding intelligent agent to execute service tasks according to the scheduling scheme. The task execution status is represented as follows: ; in Indicates the task execution status. Indicates business tasks. Indicates the executing intelligent agent; S402: The status acquisition unit collects business operation indicators, resource status, and task execution status in real time to form a business status vector. ; in Indicates the CPU resource status. Indicates memory status. Indicates network status. Indicates the operational status of the service; S403: The anomaly identification unit identifies abnormal task behaviors based on changes in business status. The formula for determining abnormal task behaviors is as follows: ; in This indicates the result of the anomaly assessment. Indicates the current state value. Represents the state mean. Indicates the standard deviation of the state. Indicates the anomaly coefficient; S404: The business feedback generation unit generates a business feedback report based on the task execution results. This report includes the task completion rate, calculated using the following formula: ; in Indicates the task completion rate. Indicates the number of successful tasks. This indicates the total number of tasks.

[0020] In this embodiment, the operational logic steps of the business closed-loop feedback and self-learning optimization module are as follows: S501: The task result evaluation unit comprehensively evaluates task execution efficiency, resource consumption, and business benefits. The comprehensive evaluation expression is as follows: ; in This represents the overall evaluation value of the task. Indicates execution efficiency. Indicates business revenue. Indicates resource consumption. , , Indicates the weighting parameter; S502: The user feedback analysis unit analyzes user suggestions for improvement and satisfaction information. The expression for user satisfaction analysis is as follows: ; in Indicates user satisfaction. Indicates user ratings, Indicates the number of feedback responses; S503: The reinforcement learning optimization unit generates reward values ​​based on the task execution results and optimizes the agent's cooperative strategy. Its reward function formula is: ; in Indicates the reward value. Indicates the task completion rate. , , Indicates the weighting coefficient; S504: The model parameter update unit dynamically updates the large model parameters and scheduling rules based on the reward value. The formula used in the parameter update process is as follows: ; in Indicates the current model parameters. This indicates the updated model parameters. Indicates the learning rate. This indicates the reward gradient.

[0021] In this embodiment, the operational logic steps of the security audit and access control module are as follows: S601: The identity authentication unit receives user login requests and intelligent agent invocation requests, performs joint verification of user identity information, device information, access token, and intelligent agent identity identifier, and generates an identity credibility score based on historical access behavior to complete the identity authentication of the user and the intelligent agent. The identity credibility calculation formula is as follows: ; in Indicates the credibility of the identity. Indicates identity information, Indicates the authentication key. Indicates an access token; S602: The access control unit generates dynamic access control policies based on user roles, business positions, data sensitivity levels, and intelligent agent capability tags, and performs permission verification for business data access, task invocation, and intelligent agent collaborative behavior; S603: The behavior auditing unit records the entire process of user operation behavior, intelligent agent invocation behavior, data access behavior, and business task execution, and establishes a traceable audit chain. The audit log set is represented as follows: ; Where L represents the audit log set. This represents the nth audit record. Each audit record includes at least the user's identity information, agent identifier, data access object, operation time, operation type, task execution result, and risk level information. When the system detects abnormal call chains, frequent access to sensitive resources, or abnormal task jump behavior, it automatically marks the corresponding audit log as high-risk. S604: The risk detection unit establishes a risk identification model based on user behavior characteristics, agent call frequency, data access patterns, and historical security events. It performs real-time detection of abnormal access behavior, unauthorized operation behavior, and potential data leakage risks. Its risk identification model uses a risk scoring function for detection, and the function expression is as follows: ; in Indicates risk score, Indicates risk factors, The risk weights are indicated by risk factors, which include high-frequency abnormal access, illegal interface calls, abnormal data downloads, unauthorized collaboration of intelligent agents, cross-domain transmission of sensitive data, and abnormal access behavior. When the risk score exceeds the preset security threshold, the system automatically performs the following actions: interrupting the execution of the current task, freezing the agent's calling permissions, restricting access to sensitive data, triggering security alarms, and pushing abnormal behavior to the security audit center. The abnormal risk behavior is also synchronized to the business closed-loop feedback and self-learning optimization module for subsequent dynamic optimization of security policies and continuous learning of risk models.

[0022] In this embodiment, the operation process of the large-model-driven enterprise management central multi-agent business empowerment system includes the following steps: S1: Unified Access and Semantic Fusion of Enterprise Multi-Source Business Data: The system first accesses structured and unstructured data from ERP, CRM, MES, SCM, OA, financial systems, and external internet business platforms through the enterprise data access and semantic fusion module, and establishes a unified data acquisition channel. Subsequently, it performs field mapping, format unification, and anomaly cleaning on data from different sources to form a unified business data set. The system then performs semantic parsing of business entities, business relationships, and business events based on a large language model, and constructs an enterprise knowledge graph by combining the enterprise's organizational structure and business processes, ultimately generating a unified business semantic model. ; S2: Large-Scale Business Understanding and Task Generation: The large-scale business understanding and task generation module receives user input of business requirements, operational goals, and management instructions. It then performs business semantic parsing using a unified business semantic model to generate a business requirement semantic vector. Subsequently, the system combines the enterprise's real-time operational status, historical business data, and external environmental information to generate a business context model. Finally, it automatically generates a business task execution chain based on the large-scale language model. ; S3: Multi-agent dynamic collaborative scheduling: The multi-agent dynamic collaborative scheduling module decomposes the business task chain and generates multiple sets of independently executable business sub-tasks. Subsequently, target agents are dynamically matched based on the agent's capability tags, resource usage status, and task dependencies. The agent matching function is expressed as follows: Then, it detects task resource conflicts and execution path conflicts in real time, and dynamically adjusts the task execution path according to task priority, thereby improving the efficiency of multi-agent collaboration and avoiding resource conflicts. S4: Real-time Business Execution and Status Awareness: The business execution and status awareness module drives each business agent to execute corresponding business tasks and collects task execution status, resource operation status, and business operation metrics in real time. The task execution status is represented as follows: The system then constructs a business state vector through the state acquisition unit. Then, abnormal task behavior is identified, and business feedback reports and task completion rate information are generated, thereby realizing real-time perception and abnormal monitoring of business operation status. S5: Business Closed-Loop Feedback and Self-Learning Optimization: The business closed-loop feedback and self-learning optimization module comprehensively analyzes task execution efficiency, business benefits, resource consumption, and user feedback. Its comprehensive task evaluation value is expressed as follows: Subsequently, the reinforcement learning optimization unit generates reward values ​​based on task completion rate and business quality. Then, the model parameter update unit dynamically updates the large model parameters and task scheduling rules based on the reward feedback, thereby enabling the system to continuously learn, autonomously optimize and evolve its intelligent capabilities based on business feedback. S6: System Security Audit and Risk Control: Jointly verify user identity information, device information, and access tokens; the credibility of the identity is expressed as follows: Subsequently, a risk identification model is established based on user behavior characteristics, agent call frequency, and historical security events. Its risk scoring function is expressed as: When the risk score exceeds the preset threshold, the system automatically performs tasks to interrupt, freeze permissions, isolate sensitive data, and issue security alarms, thereby ensuring the safe and reliable operation of the enterprise management central system.

[0023] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A large-scale model-driven enterprise management hub multi-agent business empowerment system, characterized in that: It includes modules for enterprise data access and semantic fusion, large-scale model business understanding and task generation, multi-agent dynamic collaborative scheduling, business execution and state perception, business closed-loop feedback and self-learning optimization, and security audit and access control. The enterprise data access and semantic fusion module is used to access enterprise ERP system, CRM system, MES system, OA system, financial system and external business data, and to perform standardized processing, semantic parsing and knowledge graph construction on multi-source heterogeneous data to generate a unified business semantic model. The large model business understanding and task generation module is used to perform semantic reasoning on enterprise business requirements, user instructions and business context based on the large language model, and automatically generate business objectives, task execution chains and intelligent agent invocation strategies. The multi-agent dynamic collaborative scheduling module is used to dynamically allocate tasks, collaboratively orchestrate, and optimize execution paths for multiple business agents based on task priority, resource occupancy status, and business dependencies. The business execution and status awareness module is used to drive each business agent to perform corresponding business operations and collect task execution status, resource operation status and business feedback information in real time. The business closed-loop feedback and self-learning optimization module is used to continuously optimize the large model parameters, task strategies and intelligent agent collaboration rules based on task execution results, user feedback and business performance indicators, so as to realize the system's self-learning evolution. The security audit and access control module is used to perform permission verification, log auditing, and risk detection on intelligent agent invocation behavior, data access behavior, and task execution process.

2. The large-model-driven enterprise management central multi-agent business empowerment system according to claim 1, characterized in that, The enterprise data access and semantic fusion module includes a multi-source data access unit, a data standardization processing unit, a semantic parsing unit, a knowledge graph construction unit, and a unified business semantic modeling unit; The multi-source data access unit is used to access data from ERP, CRM, MES, OA, financial systems, and external Internet data; the data standardization processing unit is used to perform field mapping, format conversion, and outlier cleaning on data of different formats; the semantic parsing unit is used to extract semantic information of business entities, business relationships, and business events using a large model; the knowledge graph construction unit is used to establish a graph of enterprise organization, processes, resources, and business relationships; and the unified business semantic modeling unit is used to generate cross-system unified business semantic vectors and business context models. The operational logic steps of the enterprise data access and semantic fusion module are as follows: S101: Accesses structured and unstructured business data from ERP, CRM, MES, SCM, OA, financial systems, and external internet business platforms via a multi-source data access unit, and establishes a unified data collection queue. The enterprise business data set is represented as follows: ； in Represents a collection of enterprise business data. This represents the nth type of business data; S102: The data standardization processing unit performs field mapping, format unification, missing value repair, and outlier cleaning on the incoming data to generate a standardized business dataset. S103: The semantic parsing unit performs business entity recognition, event extraction, and business relationship parsing on standardized business data based on a large language model, generating a business semantic feature vector. ； in This represents the semantic feature vector corresponding to the i-th type of business data. This represents the semantic mapping function of a large language model. Indicates business data content, Indicates business context information; S104: The knowledge graph construction unit constructs an enterprise knowledge graph based on business entities and business relationships, which is used to describe the enterprise's organizational structure, business processes, and resource relationships. Enterprise knowledge graph representation is as follows: ; in Represents an enterprise knowledge graph. Represents a set of business entity nodes. It represents a set of business relationships, and establishes the association mapping relationship between enterprise business objects through an enterprise knowledge graph; S105: The unified business semantic modeling unit generates a unified business semantic model based on business semantic feature vectors and enterprise knowledge graphs. ; in Represents a unified business semantic model. This represents a semantic fusion function. Represents a set of business semantic vectors. Represents an enterprise knowledge graph; S106: The system dynamically updates the unified business semantic model based on real-time business status, task execution results, and changes in the external environment, thereby ensuring that the business semantic model remains consistent with the actual operating status of the enterprise. The update formula is as follows: ； in This represents the business semantic model at the current moment. This represents the updated business semantic model. Indicates the amount of change in business semantics. This represents the update coefficient.

3. The large-model-driven enterprise management hub multi-agent business empowerment system according to claim 2, characterized in that, The large model business understanding and task generation module includes a business instruction parsing unit, a context awareness unit, a task chain generation unit, an agent invocation strategy generation unit, and a task priority evaluation unit. The business instruction parsing unit is used to parse the business requirements, management objectives and task intentions input by the user; the context awareness unit is used to generate context semantics by combining the enterprise's real-time operating status, historical tasks and business environment information; the task chain generation unit is used to generate a business execution step chain based on large model reasoning; and the agent invocation strategy generation unit is used to determine the agent invocation order, collaboration relationship and resource allocation scheme. The task priority assessment unit is used to generate task priorities based on business urgency, resource utilization, and scope of impact. The operational logic steps of the large-scale model business understanding and task generation module are as follows: S201: The business instruction parsing unit receives user input of business requirements, operational objectives, and management instructions, and performs semantic parsing using a unified business semantic model to generate a business requirement feature vector. ; in Represents a semantic vector of business requirements. This represents a function for parsing business requirements. This indicates that the user has entered a business command. Represents a unified business semantic model; S202: The context-aware unit combines the enterprise's real-time operating status, historical task data, and business environment information to generate business context semantic information, thereby enhancing the large model's ability to understand business scenarios. The business context model is represented as: ; in Represents the business context model. This indicates a context fusion function. Represents a semantic vector of business requirements. Represents historical business data. This represents real-time operating environment data; S203: The task chain generation unit generates business task execution chains based on business requirement semantics and context models, using a large language model. ; in Indicates the business task chain, This represents the nth business subtask; S204: The agent invocation strategy generation unit generates an agent collaborative invocation strategy based on the business task chain, agent capability tags and system resource status. S205: The task priority evaluation unit generates task priority parameters based on the urgency of the business, the resource occupancy status, and the degree of business impact, and outputs them to the multi-agent dynamic collaborative scheduling module. The task priority function is expressed as: ; in Indicates the priority of the i-th task. Indicates the urgency of the business. Indicates resource utilization rate. Indicates the degree of business impact. , , This represents the weighting parameter.

4. The large-model-driven enterprise management hub multi-agent business empowerment system according to claim 3, characterized in that, The multi-agent dynamic collaborative scheduling module includes an agent registration and management unit, a task decomposition unit, a dynamic scheduling unit, a conflict detection unit, a collaborative communication unit, and an execution path optimization unit. The agent registration and management unit is used to maintain the capability tags, running status, and service interfaces of each agent; the task decomposition unit is used to break down complex business tasks into multiple executable subtasks; the dynamic scheduling unit is used to dynamically allocate agents according to resource status and task dependencies; the conflict detection unit is used to identify task resource conflicts, execution path conflicts, and priority conflicts; the cooperative communication unit is used to realize context sharing and state synchronization among multiple agents; and the execution path optimization unit is used to optimize the task execution order based on historical execution results. The operational logic steps of the multi-agent dynamic cooperative scheduling module are as follows: S301: The task decomposition unit breaks down the business task chain into multiple sets of independently executable subtasks. ; in Represents a set of business subtasks. This represents the nth business subtask; S302: The agent registration and management unit matches the agent's capability tags with the task requirements to determine the target agent to execute; The formula for the agent matching function is: ; in Representation and Task Matching target agent, This indicates the degree of matching between the task and the agent. Represents the j-th agent; S303: The dynamic scheduling unit schedules tasks based on task priority, resource utilization, and task dependencies, where task scheduling priority is expressed as follows: ； in Indicates task priority. Indicates the urgency of the task. Indicates resource utilization rate. Indicates the degree of task dependency. , , Indicates the weighting parameter; S304: The conflict detection unit identifies task resource conflicts and execution path conflicts, and realizes context sharing between agents through the cooperative communication unit. The formula for calculating the task conflict degree is: ； in Indicates the degree of task conflict. , This represents a set of task resources. When the conflict level exceeds a threshold, the system automatically adjusts the task execution path. S305: The execution path optimization unit optimizes the task execution order based on the historical task execution results and outputs the final scheduling scheme.

5. The large-model-driven enterprise management hub multi-agent business empowerment system according to claim 4, characterized in that, The business execution and status awareness module includes a business execution control unit, a status acquisition unit, an anomaly identification unit, and a business feedback generation unit; The business execution control unit is used to control each intelligent agent to perform corresponding business operations; the status acquisition unit is used to collect task execution status, device status and business indicators; the anomaly identification unit is used to identify task failure, process blockage and abnormal fluctuations; the business feedback generation unit is used to generate business execution results and feedback reports. The operational logic steps of the business execution and status awareness module are as follows: S401: The service execution control unit drives the corresponding intelligent agent to execute service tasks according to the scheduling scheme. The task execution status is represented as follows: ; in Indicates the task execution status. Indicates business tasks. Indicates the executing intelligent agent; S402: The status acquisition unit collects business operation indicators, resource status, and task execution status in real time to form a business status vector. ; in Indicates the CPU resource status. Indicates memory status. Indicates network status. Indicates the operational status of the service; S403: The anomaly identification unit identifies abnormal task behaviors based on changes in business status. The formula for determining abnormal task behaviors is as follows: ; in This indicates the result of the anomaly assessment. Indicates the current state value. Represents the state mean. Indicates the standard deviation of the state. Indicates the anomaly coefficient; S404: The business feedback generation unit generates a business feedback report based on the task execution results. This report includes the task completion rate, calculated using the following formula: ； in Indicates the task completion rate. Indicates the number of successful tasks. This indicates the total number of tasks.

6. The large-model-driven enterprise management central multi-agent business empowerment system according to claim 5, characterized in that, The business closed-loop feedback and self-learning optimization module includes a task result evaluation unit, a user feedback analysis unit, a reinforcement learning optimization unit, a model parameter update unit, and an experience knowledge accumulation unit. The task result evaluation unit is used to analyze task completion rate, execution efficiency, and business benefits; The user feedback analysis unit is used to analyze user operation feedback, satisfaction, and suggestions for improvement. Reinforcement learning optimization units are used to optimize agent cooperative strategies based on reward mechanisms; The model parameter update unit is used to dynamically update the prompt word parameters or business rule parameters of the large model; the experience knowledge accumulation unit is used to build an enterprise business experience library and historical case library; The operational logic steps of the business closed-loop feedback and self-learning optimization module are as follows: S501: The task result evaluation unit comprehensively evaluates task execution efficiency, resource consumption, and business benefits. The comprehensive evaluation expression is as follows: ； in This represents the overall evaluation value of the task. Indicates execution efficiency. Indicates business revenue. Indicates resource consumption. , , Indicates the weighting parameter; S502: The user feedback analysis unit analyzes user suggestions for improvement and satisfaction information, where the expression for user satisfaction analysis is: ; in Indicates user satisfaction. Indicates user ratings, Indicates the number of feedback responses; S503: The reinforcement learning optimization unit generates reward values ​​based on the task execution results and optimizes the agent's cooperative strategy. Its reward function formula is: ; in Indicates the reward value. Indicates the task completion rate. , , Indicates the weighting coefficient; S504: The model parameter update unit dynamically updates the large model parameters and scheduling rules based on the reward value. The formula used in the parameter update process is as follows: ； in Indicates the current model parameters. This indicates the updated model parameters. Indicates the learning rate. This indicates the reward gradient.

7. The large-model-driven enterprise management central multi-agent business empowerment system according to claim 6, characterized in that, The security audit and access control module includes an identity authentication unit, an access control unit, a behavior audit unit, a risk detection unit, and a data desensitization unit; The identity authentication unit is used to verify the identity of the user and the intelligent agent; the permission control unit is used to control the data access permissions of different intelligent agents; the behavior auditing unit is used to record the intelligent agent's calling behavior and task execution logs. The risk detection unit is used to identify risks such as abnormal calls, unauthorized access, and sensitive data leakage; the data desensitization unit is used to encrypt and desensitize sensitive business data. The operational logic steps of the security audit and access control module are as follows: S601: The identity authentication unit receives user login requests and intelligent agent invocation requests, performs joint verification of user identity information, device information, access token, and intelligent agent identity identifier, and generates an identity credibility score based on historical access behavior to complete the identity authentication of the user and the intelligent agent. The identity credibility calculation formula is as follows: ； in Indicates the credibility of the identity. Indicates identity information, Indicates the authentication key. Indicates an access token; S602: The access control unit generates dynamic access control policies based on user roles, business positions, data sensitivity levels, and intelligent agent capability tags, and performs permission verification for business data access, task invocation, and intelligent agent collaborative behavior; S603: The behavior auditing unit records the entire process of user operation behavior, intelligent agent invocation behavior, data access behavior, and business task execution, and establishes a traceable audit chain. The audit log set is represented as follows: ; Where L represents the audit log set. This represents the nth audit record. Each audit record includes at least the user's identity information, agent identifier, data access object, operation time, operation type, task execution result, and risk level information. When the system detects abnormal call chains, frequent access to sensitive resources, or abnormal task jump behavior, it automatically marks the corresponding audit log as high-risk. S604: The risk detection unit establishes a risk identification model based on user behavior characteristics, agent call frequency, data access patterns, and historical security events. It performs real-time detection of abnormal access behavior, unauthorized operation behavior, and potential data leakage risks. Its risk identification model uses a risk scoring function for detection, and the function expression is as follows: ； in Indicates risk score, Indicates risk factors, The risk weights are indicated by risk factors, which include high-frequency abnormal access, illegal interface calls, abnormal data downloads, unauthorized collaboration of intelligent agents, cross-domain transmission of sensitive data, and abnormal access behavior. When the risk score exceeds the preset security threshold, the system automatically performs the following actions: interrupting the execution of the current task, freezing the agent's calling permissions, restricting access to sensitive data, triggering security alarms, and pushing abnormal behavior to the security audit center. The abnormal risk behavior is also synchronized to the business closed-loop feedback and self-learning optimization module for subsequent dynamic optimization of security policies and continuous learning of risk models.

8. The large-model-driven enterprise management central multi-agent business empowerment system according to claim 7, characterized in that, The enterprise data access and semantic fusion module, the large model business understanding and task generation module, the multi-agent dynamic collaborative scheduling module, the business execution and state perception module, the business closed-loop feedback and self-learning optimization module, and the security audit and access control module form a closed-loop collaborative connection relationship. The enterprise data access and semantic fusion module is connected to the large model business understanding and task generation module, and is used to output a unified business semantic model, enterprise knowledge graph and business context information to the large model business understanding and task generation module. The large model business understanding and task generation module is connected to the multi-agent dynamic collaborative scheduling module and is used to output the business task chain, agent calling strategy and task priority parameters to the multi-agent dynamic collaborative scheduling module. The multi-agent dynamic collaborative scheduling module is connected to the business execution and state awareness module, and is used to output task scheduling schemes, agent execution paths and collaborative control parameters to the business execution and state awareness module; The business execution and status awareness module is connected to the business closed-loop feedback and self-learning optimization module, and is used to output task execution results, resource operation status, business feedback information and abnormal behavior information to the business closed-loop feedback and self-learning optimization module. The business closed-loop feedback and self-learning optimization module is connected to the large model business understanding and task generation module, and is used to feed back model optimization parameters, task optimization strategies and historical business experience information to the large model business understanding and task generation module, so as to realize the dynamic optimization of the large model reasoning ability and agent collaboration strategy. The security audit and access control module is bidirectionally connected to the enterprise data access and semantic fusion module, the large model business understanding and task generation module, the multi-agent dynamic collaborative scheduling module, the business execution and state perception module, and the business closed-loop feedback and self-learning optimization module. It is used to perform access verification, behavior auditing and risk control on the data access behavior, agent calling behavior and task execution process between the modules, thereby forming a security control link that runs through the entire system.

Images