Method and device for making intent translation in multi-agent network interpretable

Abstract

The application discloses a kind of method and device of intention translation interpretable in multi-agent network, belong to new generation information technology, wireless communication, edge computing, artificial intelligence network arrangement and intention driven network technical field;Device includes intention perception agent, network symbol translation agent, collaborative arrangement execution agent and audit explanation agent;The method is first joint perception multidimensional information and constructs structured knowledge, then knowledge is converted into network symbol, relies on symbol generation service workflow, resource strategy and control action;Finally, through audit explanation agent, evidence chain verification, compliance audit, cause and effect interpretation and closed-loop correction are completed, the reliable, interpretable translation and arrangement of application intention to network control action are realized;The application realizes the reliable translation and interpretable arrangement of application intention to network control action, improves network operation credibility and operation efficiency.

Description

Technical Field

[0001] This invention belongs to the fields of next-generation information technology, wireless communication, edge computing, artificial intelligence network orchestration and intent-driven network technology, and particularly relates to a method and apparatus for interpretable intent translation in multi-agent networks. Background Technology

[0002] With the deep integration of artificial intelligence, large-scale models, multi-agent systems, and wireless edge networks, network services are gradually evolving from traditional connectivity provision and resource scheduling models to intent-driven intelligent autonomous control models. In this model, users, applications, or business systems can express their application intents through business objectives, quality of service requirements, or policy constraints, and the network system needs to automatically complete intent understanding, task decomposition, resource orchestration, and control action execution.

[0003] However, the introduction of artificial intelligence models and multi-agent systems has transformed the intention execution process from traditional rule-driven to model-based reasoning and multi-agent collaborative driving. While this approach enhances the automation and adaptability of networks, it also makes processes such as intention understanding, network symbol translation, resource selection, service orchestration, and control action generation more complex. Furthermore, the execution results are difficult to directly explain regarding their source, constraint satisfaction, and decision-making process.

[0004] In wireless edge networks, application intents typically require processing multiple types of information, including user context, wireless access status, edge resource status, network topology, service capabilities, and policy constraints. This information is then further transformed into machine-recognizable network symbols, service workflows, resource mapping policies, and specific control actions. Existing network orchestration methods generally focus more on network function deployment, traffic scheduling, and resource allocation, lacking sufficient support for interpretable translation between intent semantics and network control actions. This makes it difficult to explain why application intents are interpreted in a particular way, why resources are selected, why control actions are generated, and whether the execution results conform to the original intent.

[0005] Meanwhile, multi-agent collaboration further increases the interpretability of the intent execution process. Different agents may undertake tasks such as intent perception, state analysis, symbol translation, resource orchestration, anomaly handling, and feedback adjustment, with the final control action often being generated jointly by multiple agents. Without a unified intermediate representation, evidence recording, and audit interpretation mechanism, it is difficult to trace the correspondence between control actions and the original application intent, real-time network state, resource constraints, and policy rules, and it is also difficult to determine whether the outputs of multiple agents are consistent.

[0006] Furthermore, wireless edge networks are characterized by dynamic changes; link quality, user location, edge node load, service function status, and policy constraints can all change over time. Therefore, intent execution should not only generate a one-time orchestration result but should also be continuously adjusted based on operational feedback. Existing methods typically lack a closed-loop mechanism that spans the entire process of intent translation, action generation, action verification, execution feedback, and anomaly interpretation. This makes it difficult to provide timely, verifiable, and traceable adjustment guidelines when service quality deteriorates, resources become congested, or policy conflicts occur.

[0007] Therefore, existing technologies have at least the following shortcomings: the translation process from application intent to network control action lacks an interpretable intermediate representation; when multiple agents collaboratively generate control actions, there is a lack of evidence chain records and consistency verification; the resource selection, service orchestration, and action admission processes lack traceable basis; and after changes in operating state, there is a lack of a closed-loop audit mechanism to link execution feedback with the original intent and orchestrated actions.

[0008] Based on the above problems, it is necessary to propose a method and device for interpretable intent translation and orchestration in wireless edge networks based on multi-agent collaboration. Through multi-agent collaboration, the method completes intent perception, network symbol translation, service workflow generation, resource mapping, evidence chain verification, audit interpretation, and feedback correction. This enables the conversion process from application intent to network control actions to be understandable, verifiable, traceable, and auditable, thereby improving the credibility of AI-driven intent execution, service continuity assurance capabilities, and operational efficiency in wireless edge networks. Summary of the Invention

[0009] The purpose of this invention is to provide a method and apparatus for interpretable intent translation in multi-agent networks, which solves the problems of opaque semantic understanding, unclear basis for generating network control requests, difficulty in tracing multi-agent collaborative decision-making, lack of verifiable evidence for orchestrated actions, and difficulty in closed-loop interpretation of execution feedback in existing wireless edge networks after the deep involvement of artificial intelligence models and multi-agent systems in network control. This invention, starting from the interpretability of intent translation, the traceability of agent collaboration, the verifiability of orchestrated actions, and the auditability of execution feedback, comprehensively considers application objectives, quality of service requirements, user context, wireless access status, edge resource status, network topology, service functionality, virtual cell configuration, policy constraints, and operational feedback information. It provides a method and apparatus for interpretable intent translation and orchestration in wireless edge networks based on multi-agent collaboration. The aim is to achieve closed-loop processing from application intent to network symbols, from network symbols to network service requests, from network service requests to orchestrated actions, from orchestrated actions to resource execution, and from execution feedback to audit interpretation. This improves the credibility, interpretability, service continuity assurance, deterministic collaborative transmission, and resource utilization efficiency of the AI-driven intent execution process in wireless edge networks.

[0010] To achieve the above objectives, the present invention provides a method for interpretable intent translation in multi-agent networks, comprising the following steps: S1. Receive application intents through the intent awareness and status acquisition layer, perform semantic parsing, context awareness and requirement extraction, and simultaneously collect wireless edge network operation status information to generate structured knowledge representations. S2. The structured knowledge representation generated based on S1 translates application intent and network state into machine-recognizable network symbols through a multi-agent collaborative translation layer, and generates network service requests, candidate service workflows, virtual cell organization schemes, resource mapping strategies, and candidate control actions. S3. Through the evidence chain verification and audit interpretation layer, using the structured knowledge representation constructed in S1 as the source of evidence, multiple verifications are performed on the network symbols, network service requests, candidate service workflows, virtual cell organization schemes, resource mapping strategies, candidate orchestration schemes, and candidate control actions generated in S2 to construct a complete evidence chain. Based on the verification results, candidate control actions are admitted or rejected to complete causal interpretation and audit output. S4. The verified orchestration scheme is distributed to the network infrastructure for execution through the orchestration execution and feedback optimization layer. The operation status is monitored in real time and feedback data is collected. Based on the feedback results, the network symbols, service workflows, resource mapping strategies and control actions are dynamically and closed-loop corrected. S5. Perform source tracing and interpretation of the entire process of intent understanding, symbol translation, orchestration and execution, verification and auditing, and feedback and correction, and output audit interpretation results.

[0011] Preferably, the specific content of S1 is as follows: S101, Deployment Intent Awareness and Status Acquisition Layer; S102. Receive application intent in the form of natural language, semi-structured text, interface request, or policy request, wherein the application intent includes at least one of business objectives, service quality requirements, spatiotemporal constraints, service continuity requirements, task priorities, and policy constraints. S103. Perform semantic parsing on the application intent, identify business objectives, service objects, service quality indicators, policy requirements and potential constraints, and generate initial intent understanding results; S104. Obtain user context, terminal status, historical service records, and mobility information, and bind them with the initial intent understanding result to obtain a context-enhanced intent representation; S105. Collect network operation information such as wireless access status, link quality, edge computing / storage resources, network topology, service function status, virtual cell status, and policy constraints; S106. Associate the context-enhanced intent representation with network operation information to generate a structured knowledge representation, wherein the structured knowledge representation is at least one of the following: knowledge enhancement graph, knowledge graph, rule base, state table, semantic ontology, and vectorized knowledge representation. S107. Perform consistency verification on the structured knowledge representation. If there is missing or conflicting information, complete, correct or re-parse it. After the verification is passed, proceed to the network symbol translation process.

[0012] Preferably, the specific content of S2 is as follows: S201, Multi-agent collaborative translation layer pre-stores agent capability information, network symbol templates, service function information, virtual cell configuration templates, and resource status views; S202. Deploy an intent-aware agent, a state-analysis agent, a network symbol translation agent, a collaborative orchestration agent, and an agent coordination module in the multi-agent collaborative translation layer. Each agent operates according to a preset interface and collaborative relationship. S203, Network Symbol Translation: The intelligent agent converts structured knowledge representations into network symbols; S204. Based on the target symbol and constraint symbol, decompose the application intent into multiple executable subtasks, and determine the subtask dependencies, execution order and priority; S205. Combining resource symbols, status symbols, service function symbols, and virtual cell symbols, match candidate service functions, radio access resources, edge resources, transmission paths, and virtual cell configurations. S206. Generate a network service request based on network symbols, wherein the network service request includes at least one of the following: quality of service objectives, isolation requirements, service function requirements, resource binding relationships, virtual cell candidate configurations, policy constraints, and candidate control actions; S207. Collaboratively orchestrate intelligent agents to combine candidate service functions and resources, generating several candidate service workflows. The service workflows include the service function call order, intelligent agent collaboration relationship, data transmission path, resource requirements, and execution constraints. S208. Based on user location, wireless coverage, access load, link quality, and service continuity requirements, generate candidate virtual cell organization schemes. S209. Based on the candidate service workflow and virtual cell organization scheme, generate a wireless resource scheduling strategy, an edge resource mapping strategy, and candidate control actions. The candidate control actions include at least one of access control, path adjustment, service function deployment, service migration, resource allocation, load balancing, virtual cell reconstruction, and anomaly handling. S210, the agent coordination module verifies the output results of each agent. If there are symbol or action conflicts, it triggers re-translation or re-arrangement. If there are no conflicts, it outputs candidate arrangement schemes.

[0013] Preferably, S3 specifically includes: S301. Evidence Chain Verification and Auditing: The intelligent agent obtains the knowledge enhancement graph constructed by the intent perception and state acquisition layer in S1, and the network symbols, network service requests, candidate service workflows, virtual cell organization schemes, resource mapping strategies, candidate orchestration schemes, and candidate control actions obtained by the multi-agent collaborative translation layer in S2, and verifies and audits the content output by S2 based on the knowledge enhancement graph. S302. Taking each candidate orchestration scheme as the retrieval object, obtain the application intent source and intent parsing result corresponding to the candidate orchestration scheme, and retrieve the corresponding business objectives, service objects, functional requirements, service quality requirements, spatiotemporal constraints, real-time network status, network topology relationships, resource availability, service function status, virtual cell configuration and policy constraints from the knowledge enhancement graph. At the same time, obtain the service quality prediction result corresponding to the candidate orchestration scheme to form an evidence set corresponding to the candidate orchestration scheme. S303. For the service quality objectives, isolation requirements, service function requirements, resource binding relationships, virtual cell candidate configurations and policy constraints in the network service request, as well as the service function call order, data transmission path, resource requirements and execution constraints in the candidate service workflow, determine the corresponding supporting evidence from the evidence set, and establish the correspondence between the key content in the network service request and the candidate service workflow and the supporting evidence. S304. Following the processing order of application intent parsing, network symbol mapping, network service request generation, candidate service workflow generation, virtual cell organization scheme generation, resource mapping strategy generation, and candidate control action generation, the inputs, outputs, and generation basis of each processing stage are sequentially associated to form a candidate evidence chain pointing from application intent to candidate control action. S305. Perform semantic consistency verification on network service requests, and determine whether the service function requirements, service quality goals and policy constraints in the network service requests are consistent with the function requirements, service quality requirements and spatiotemporal constraints and policy constraints recorded in the knowledge enhancement graph. S306. Verify the resource feasibility of the candidate control actions and determine whether the radio resources, computing resources, storage resources, transmission resources, service function resources and virtual cell resources required by the candidate control actions meet the execution requirements. S307. Perform policy consistency verification on candidate control actions to determine whether the candidate control actions comply with service level agreements, security policies, priority rules, isolation rules, and time and space constraints. S308. Based on the service quality prediction results in the evidence set, perform service quality verification on the candidate service workflow, virtual cell organization scheme and resource mapping strategy, and determine whether they meet the requirements of latency, reliability, throughput, isolation, service continuity and coverage. S309. When semantic consistency verification, resource feasibility verification, policy consistency verification, and service quality verification all pass, generate action admission results and corresponding action admission basis; when any verification item fails, generate action rejection results including the failed verification item and its rejection reason; associate the verification basis, verification results, action admission results, and action admission basis or action rejection results of each verification with the corresponding network service request, candidate orchestration scheme, and candidate control action, and record them in the candidate evidence chain. S310. Perform integrity verification on the candidate evidence chain, and determine whether the key content in the network service request and candidate service workflow all have corresponding supporting evidence, such as the application intent source, intent parsing result, supporting evidence provided by the knowledge enhancement graph, network symbol mapping relationship, network service request, candidate service workflow, virtual cell organization scheme, resource mapping strategy, candidate orchestration scheme, candidate control action and its generation basis, various verification basis and verification results, and action admission result or action rejection result. Also determine whether the association relationship established according to the processing order described in S304 is continuous and traceable. When all evidence information and verification results are recorded and the association relationship is continuous, a complete evidence chain is formed. Otherwise, output the missing evidence information or the interrupted association relationship. When an action rejection result is generated or the candidate evidence chain is incomplete, the corresponding result is fed back to the multi-agent collaborative translation layer for reprocessing. S311. Using the application intent, network status, resource capabilities, service functions, virtual cell configuration, policy constraints and their relationships recorded in the knowledge enhancement graph as the basis for interpretation, trace back along the complete evidence chain to the generation process of network symbols, network service requests, candidate service workflows, virtual cell organization schemes, resource mapping strategies and candidate control actions, analyze the generation dependencies and constraint relationships between each processing stage, and form a causal explanation for the candidate control actions. S312. Integrate the supporting evidence retrieved from the knowledge enhancement graph, the correspondence between network service requests and supporting evidence, the basis for generating candidate control actions, the basis and results of various verifications, and the basis for action admission or reasons for rejection, and record the integrated information in the complete evidence chain.

[0014] Preferably, the specific content of S4 is as follows: S401, Deployment, Orchestration, Execution, and Feedback Optimization Layer; S402. Analyze the verified orchestration scheme to determine the service workflow, virtual cell organization scheme, resource mapping relationship, control action set, task execution order and exception handling strategy; S403. According to the resource mapping relationship, bind the service functions and control tasks to the corresponding wireless access node, edge computing node, transmission node, core network node or cloud resource node; S404, issue execution instructions for access control, resource allocation, service deployment, link configuration, service migration, load balancing, virtual cell reconstruction, and exception handling; S405: Real-time collection of service quality indicators, wireless link status, edge resource load, node availability, user mobility status, virtual cell operation status, and abnormal events; S406. When a service quality degradation, link fluctuation, resource congestion, service anomaly, service continuity risk, virtual cell failure, or policy change is detected, a feedback event is generated. S407. Associate feedback events with original application intent, structured knowledge, network symbols, service requests, orchestration schemes, and resource mapping relationships to form feedback evidence; S408. Based on the feedback evidence, dynamically adjust the network symbols, network service requests, agent collaboration relationships, service workflows, virtual cell organization schemes, and resource mapping strategies, and resubmit the adjusted schemes for verification.

[0015] An apparatus for interpretable intent translation in a multi-agent network, employing the method described in any of the preceding claims, comprising an intent perception and state acquisition layer, a multi-agent collaborative translation layer, an evidence chain verification and audit interpretation layer, and an orchestration execution and feedback optimization layer; The intent perception and state acquisition layer includes an application intent receiving module, a context perception module, a network state acquisition module, and a structured knowledge generation module; it is used to receive and parse application intents, acquire network operating states, and generate structured knowledge representations. The multi-agent collaborative translation layer includes an intent-aware agent, a state-analysis agent, a network symbol translation agent, a collaborative orchestration agent, and an agent coordination module; it is used to convert structured knowledge into network symbols, generate network service requests, service workflows, virtual cell organization schemes, resource mapping strategies, and candidate control actions, and complete conflict verification of multi-agent output results. The evidence chain verification and audit interpretation layer includes an evidence chain construction module, an action access verification module, a consistency verification module, a causal interpretation module, and an audit output module; it is used to construct a full-process evidence chain, complete multi-dimensional verification of resources, strategies, and service quality, and realize causal interpretation and audit result output. The orchestration execution and feedback optimization layer includes a resource mapping module, a control action execution module, an operation status monitoring module, an anomaly feedback module, and a closed-loop correction module; it is used to execute verified orchestration schemes, monitor operation status, collect anomaly feedback, and complete dynamic closed-loop correction across the entire link.

[0016] Preferably, the intent perception and status acquisition layer is deployed at the service entry point, network control node, edge control node, or network management platform; the orchestration execution and feedback optimization layer is distributed and deployed in the wireless access network, edge node, core network, and cloud platform.

[0017] Preferably, the network symbols output by the network symbol translation agent include at least one of the following: target symbol, constraint symbol, state symbol, resource symbol, service function symbol, virtual cell symbol, policy symbol, and action symbol.

[0018] Preferably, the evidence chain generated by the evidence chain construction module fully records the entire process from application intent input, intent parsing, structured knowledge generation, network symbol mapping, service request generation, resource verification to control action output.

[0019] Preferably, the closed-loop correction module iteratively adjusts the network symbols, network service requests, intelligent agent collaboration logic, service workflow, virtual cell organization scheme, and resource mapping strategy based on the operation feedback events, and sends the adjustment results back to the evidence chain verification and audit interpretation layer for re-verification.

[0020] Therefore, the present invention employs the above-mentioned method and apparatus for interpretable intent translation in a multi-agent network, which has the following beneficial effects: This invention addresses the issue of insufficient interpretability of intent execution arising from the deep integration of artificial intelligence models and multi-agent systems into wireless edge network control. It provides a method and apparatus for interpretable intent translation and orchestration in wireless edge networks based on multi-agent collaboration. This method and apparatus can transform complex application intents into structured knowledge representations and machine-recognizable network symbols, further forming network service requests, resource control requests, or orchestration control requests. Through multi-agent collaboration, it generates service workflows, virtual cell organization schemes, resource mapping strategies, and candidate control actions. It verifies, traces, and audits control actions through evidence chain verification and audit interpretation mechanisms. Finally, it performs closed-loop correction of network symbols, service workflows, virtual cell organization schemes, and resource mapping strategies through operational feedback. Therefore, this invention improves the credibility, interpretability, service continuity assurance, cooperative transmission determinism, resource utilization efficiency, and network operation and maintenance efficiency of AI-driven intent execution processes in wireless edge networks.

[0021] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0022] Figure 1 A schematic diagram of the overall architecture for interpretable intent translation in a wireless edge network based on multi-agent collaboration; Figure 2 This is a schematic diagram of a wireless edge network-based interpretable intent translation and orchestration device based on multi-agent collaboration. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0024] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0025] The following is combined with Figures 1-2 The embodiments of the present invention will be described in detail below.

[0026] Example 1 like Figure 1 As shown, this invention provides a method for interpretable intent translation in multi-agent networks, comprising the following steps: S1. Receive application intents input by users, applications, business systems, or network management systems through the intent awareness and state acquisition layer. Perform semantic parsing, context awareness, and requirement extraction on the application intents to obtain information on business objectives, service quality requirements, user context, spatiotemporal constraints, service continuity requirements, and policy constraints. Simultaneously, collect information on wireless access status, link quality, edge resource status, service function status, user mobility status, virtual cell status, and network topology in the wireless edge network, and generate a structured knowledge representation oriented towards intent execution.

[0027] S2. Through the multi-agent collaborative translation layer, the structured knowledge representation generated in S1 is utilized to call the intent-aware agent, the state-analysis agent, the network symbol translation agent, and the collaborative orchestration agent to translate the application intent and network state into machine-recognizable network symbols. Based on the network symbols, network service requests, resource control requests, or orchestration control requests are generated. Based on the network service requests, candidate service workflows, virtual cell organization schemes, radio resource scheduling strategies, edge resource mapping strategies, and candidate control actions are generated.

[0028] S3, Evidence Chain Verification and Audit Interpretation: The intelligent agent acquires the structured knowledge representation constructed in S1, as well as network symbols, network service requests, candidate service workflows, virtual cell organization schemes, resource mapping strategies, candidate orchestration schemes, and candidate control actions obtained from S2. Using candidate orchestration schemes as the retrieval object, it retrieves corresponding supporting evidence from the structured knowledge representation, establishes the correspondence between key content in network service requests and candidate service workflows and supporting evidence, and forms candidate evidence according to the processing order of application intent parsing, network symbol mapping, network service request generation, candidate orchestration scheme generation, and candidate control action generation. The chain verifies semantic consistency, resource feasibility, policy consistency, and service quality of network service requests and candidate control actions. Based on the verification results, it generates action admission or rejection results and records the verification basis and results to the candidate evidence chain. When the evidence information and verification results are complete and the correlation is continuous, a complete evidence chain is formed, and causal explanation and audit explanation results are generated based on the complete evidence chain. When an action rejection result is generated or the candidate evidence chain is incomplete, the corresponding results are fed back to the multi-agent collaborative translation layer for network symbol reconstruction, network service request correction, candidate service workflow adjustment, or resource mapping reselection.

[0029] S4. Through the orchestration execution and feedback optimization layer, the validated orchestration scheme is mapped to infrastructure resources such as wireless, edge, and cloud for execution. During the execution process, service quality indicators, resource utilization, link status, changes in user mobility, virtual cell operation status, and abnormal events are continuously collected. Based on the operation feedback, network symbols, network service requests, agent collaboration relationships, service workflows, virtual cell organization schemes, and resource mapping relationships are dynamically adjusted.

[0030] S5. Through the audit interpretation mechanism, the process of application intent understanding, structured knowledge generation, network symbol translation, network service request generation, candidate action generation, evidence chain verification, resource mapping execution, and feedback correction is interpreted, and traceable audit interpretation results are output, so that the entire process from application intent to network control action is understandable, verifiable, traceable, and auditable.

[0031] Step S1 includes: S101: Construct an intent perception and status acquisition layer. The intent perception and status acquisition layer is deployed in the service entry point, network control node, edge control node or network management platform. It is used to uniformly perceive, semantically process and correlate the original application intent and network operation status before the application intent enters the orchestration process.

[0032] S102: Receive application intent input from users, applications, business systems, or network management systems. The application intent includes natural language requests, semi-structured text requests, interface-based business requests, or policy-based service requests. The application intent is used to describe one or more of the following: business objectives, service objects, service quality requirements, spatiotemporal constraints, service continuity requirements, task priorities, and policy constraints.

[0033] S103: Invoke the intent-aware intelligent agent to perform semantic parsing on the application intent, identify the business objectives, service objects, business scenarios, service quality indicators, spatiotemporal boundaries, policy requirements and potential constraints, and generate initial intent understanding results.

[0034] S104: Call the context-aware module to obtain the user context, terminal status, service environment, historical service records, mobility information and service priority information related to the application intent, and bind them with the initial intent understanding result to obtain a context-enhanced intent representation.

[0035] S105: Call the network status acquisition module to collect wireless access status, link quality, access load, edge computing resources, storage resources, cache resources, network topology, service function status, virtual cell status and policy constraint information in the wireless edge network.

[0036] S106: Invoke the structured knowledge generation module to associate the context-enhanced intent representation with network state information to generate a structured knowledge representation for intent execution; the structured knowledge representation is used to describe the relationship between application intent, network state, resource capabilities, service functions, virtual cell configuration, policy rules and operating constraints.

[0037] S107: Perform a consistency check on the structured knowledge representation to determine whether there are any missing, conflicting, or inconsistent elements among the business objectives, service quality requirements, resource status, virtual cell status, policy constraints, and context information; if there are any missing or conflicting elements, complete, correct, or re-parse them; if there are no missing or conflicting elements, proceed to the network symbol translation process.

[0038] Step S2 includes: S201: Construct a multi-agent collaborative translation layer, which is used to receive the structured knowledge representation generated in step S1 and maintain agent capability information, network symbol templates, service function information, virtual cell configuration templates and resource status views.

[0039] S202: Construct an intent-aware agent, a state-analysis agent, a network symbol translation agent, a collaborative orchestration agent, and an agent coordination module in the multi-agent collaborative translation layer; different agents register and call each other according to preset input / output interfaces, triggering conditions, and collaborative relationships.

[0040] S203: The intelligent agent converts structured knowledge representation into machine-recognizable network symbols through network symbol translation. The network symbols include one or more of the following: target symbols, constraint symbols, state symbols, resource symbols, service function symbols, virtual cell symbols, policy symbols, and action symbols.

[0041] S204: Based on the target symbol and constraint symbol, the application intent is split into multiple executable subtasks, and the dependencies, execution order and priority between each subtask are determined.

[0042] S205: Based on the resource symbols, status symbols, service function symbols, and virtual cell symbols, match candidate service functions, candidate radio access resources, candidate edge computing resources, candidate transmission paths, and candidate virtual cell configurations that meet the current application intent.

[0043] S206: Generate a network service request based on the network symbol. The network service request includes one or more of the following: quality of service objectives, isolation requirements, service function requirements, resource binding relationships, virtual cell candidate configurations, policy constraints, and candidate control actions.

[0044] S207: By coordinating and orchestrating intelligent agents, candidate service functions and candidate resources are combined to generate multiple candidate service workflows; each candidate service workflow includes the service function call order, intelligent agent collaboration relationship, data transmission path, resource requirements and execution constraints.

[0045] S208: Generate a candidate virtual cell organization scheme based on user location, wireless coverage relationship, access load, link quality, candidate virtual cell configuration and service continuity requirements; the virtual cell organization scheme is used to determine the access nodes, edge nodes, service function nodes and their cooperative relationships that provide collaborative services for the target service.

[0046] S209: Based on the candidate service workflow and candidate virtual cell organization scheme, generate a wireless resource scheduling strategy, an edge resource mapping strategy, and candidate control actions; the candidate control actions include one or more of the following: access control, path adjustment, service function deployment, service migration, resource allocation, load balancing, virtual cell reconstruction, and anomaly handling.

[0047] S210: Align the output results of multiple agents through the agent coordination module, and determine whether there are conflicts between the target symbols, resource symbols, policy symbols, virtual cell symbols and candidate control actions generated by different agents; if there are conflicts, trigger re-translation or re-arrangement; if there are no conflicts, output candidate arrangement schemes.

[0048] Step S3 includes: S301: Evidence Chain Verification and Auditing Explanation The intelligent agent obtains the knowledge enhancement graph constructed in S1, as well as the network symbols, network service requests, candidate service workflows, virtual cell organization schemes, resource mapping strategies, candidate orchestration schemes, and candidate control actions output in S2, and verifies and audits the above-mentioned contents output in S2 based on the knowledge enhancement graph.

[0049] S302: Using each candidate orchestration scheme as the retrieval object, obtain the application intent source and intent parsing result corresponding to the candidate orchestration scheme, and retrieve the corresponding business objectives, service objects, functional requirements, service quality requirements, spatiotemporal constraints, real-time network status, network topology relationships, resource availability, service function status, virtual cell configuration and policy constraints from the knowledge enhancement graph. At the same time, obtain the service quality prediction result corresponding to the candidate orchestration scheme to form an evidence set corresponding to the candidate orchestration scheme.

[0050] S303: For the service quality objectives, isolation requirements, service function requirements, resource binding relationships, virtual cell candidate configurations and policy constraints in the network service request, as well as the service function call order, data transmission path, resource requirements and execution constraints in the candidate service workflow, determine the corresponding supporting evidence from the evidence set, and establish the correspondence between the key content in the network service request and the candidate service workflow and the supporting evidence.

[0051] S304: Based on the correspondence between the key content and supporting evidence in the network service request and candidate service workflow established in S303, and in accordance with the processing order of application intent parsing, network symbol mapping, network service request generation, candidate service workflow generation, virtual cell organization scheme generation, resource mapping strategy generation, and candidate control action generation, the inputs, outputs, and generation basis of each processing stage are sequentially associated, and the evidence set, network service request, candidate service workflow, virtual cell organization scheme, resource mapping strategy, candidate orchestration scheme, and candidate control action are organized into a candidate evidence chain pointing from application intent to candidate control action.

[0052] S305: Perform semantic consistency verification on network service requests, and determine whether the service function requirements, service quality objectives and policy constraints in the network service requests are consistent with the function requirements, service quality requirements and spatiotemporal constraints and policy constraints recorded in the knowledge enhancement graph.

[0053] S306: Perform resource feasibility verification on candidate control actions to determine whether the radio resources, computing resources, storage resources, transmission resources, service function resources, and virtual cell resources required by the candidate control actions meet the execution requirements.

[0054] S307: Perform policy consistency verification on candidate control actions to determine whether the candidate control actions comply with service level agreements, security policies, priority rules, isolation rules, and time and space constraints.

[0055] S308: Based on the service quality prediction results in the evidence set, perform service quality verification on the candidate service workflow, virtual cell organization scheme and resource mapping strategy to determine whether they meet the requirements of latency, reliability, throughput, isolation, service continuity and coverage.

[0056] S309: When semantic consistency verification, resource feasibility verification, policy consistency verification, and service quality verification all pass, generate an action admission result and the corresponding action admission basis; when any verification item fails, generate an action rejection result including the failed verification item and its rejection reason. Associate the verification basis, verification result, action admission result, and action admission basis or action rejection result of each verification with the corresponding network service request, candidate orchestration scheme, and candidate control action, and record them in the candidate evidence chain.

[0057] S310: Perform a completeness check on the candidate evidence chain, determining whether the key content in the network service request and candidate service workflow all have corresponding supporting evidence, including whether the application intent source, intent parsing result, supporting evidence provided by the knowledge augmentation graph, network symbol mapping relationship, network service request, candidate service workflow, virtual cell organization scheme, resource mapping strategy, candidate orchestration scheme, candidate control action and its generation basis, various verification basis and verification results, and action admission or action rejection results have all been recorded, and whether the association relationship established according to the processing order described in S304 is continuous and traceable. When the above evidence information and verification results are all recorded and the association relationship is continuous, a complete evidence chain is formed; otherwise, output the missing evidence information or the interrupted association relationship. When an action rejection result is generated or the candidate evidence chain is incomplete, the corresponding result is fed back to the multi-agent collaborative translation layer for re-translation or re-orchestration.

[0058] S311: Using the application intent, network status, resource capabilities, service functions, virtual cell configuration, policy constraints and their relationships recorded in the knowledge enhancement graph as the basis for interpretation, trace back along the complete evidence chain to the generation process of network symbols, network service requests, candidate service workflows, virtual cell organization schemes, resource mapping strategies and candidate control actions, analyze the generation dependencies and constraint relationships between each processing stage, and form a causal explanation for the candidate control actions.

[0059] S312: Integrate the supporting evidence retrieved from the knowledge-enhanced graph, the correspondence between network service requests and supporting evidence, the basis for generating candidate control actions, the basis and results of various verifications, and the basis for action admission or reasons for rejection, and record the integrated information in the complete evidence chain.

[0060] Step S4 includes: S401: Construct an orchestration execution and feedback optimization layer, which is distributed in the wireless access network, edge nodes, core network, cloud platform or network controller, and is used to receive orchestration schemes verified by the chain of evidence.

[0061] S402: Analyze the validated orchestration scheme to determine the service workflow, virtual cell organization scheme, resource mapping relationship, control action set, task execution order and exception handling strategy.

[0062] S403: Based on resource mapping relationships, bind service functions, business intelligence agents, or control tasks to the corresponding wireless access nodes, edge computing nodes, transmission nodes, core network nodes, or cloud resource nodes.

[0063] S404: Issue execution instructions based on the set of control actions. The execution instructions include one or more of the following: access control instructions, resource allocation instructions, service function deployment instructions, link configuration instructions, service migration instructions, load balancing instructions, virtual cell reconstruction instructions, and exception handling instructions.

[0064] S405: During execution, continuously collect service operation status, service quality indicators, radio link quality, edge resource load, node availability, user mobility status, virtual cell operation status, and abnormal events.

[0065] S406: When a service quality degradation, link fluctuation, edge resource congestion, service function abnormality, service continuity risk caused by user mobility, virtual cell configuration failure, or policy constraint change is detected, a feedback event is generated.

[0066] S407: Associate the feedback event with the original application intent, structured knowledge representation, network symbols, network service requests, service workflows, virtual cell organization schemes, resource mapping relationships, and control actions to form feedback evidence.

[0067] S408: Trigger closed-loop correction based on feedback evidence, dynamically adjust network symbols, network service requests, agent collaboration relationships, service workflows, virtual cell organization schemes or resource mapping strategies, and resubmit the adjustment results to the evidence chain verification and audit interpretation layer for verification.

[0068] Example 2 like Figure 2 As shown, the present invention provides an interpretable intent translation device in a multi-agent network, comprising an intent perception and state acquisition layer, a multi-agent collaborative translation layer, an evidence chain verification and audit interpretation layer, and an orchestration execution and feedback optimization layer, as described below: The Intent Awareness and State Acquisition Layer comprises an application intent receiving module, a context awareness module, a network state acquisition module, and a structured knowledge generation module. This layer receives application intents input from users, applications, business systems, or network management systems. It performs semantic recognition, context binding, and requirement extraction on natural language, semi-structured text, interface-based requests, or policy-based requests. Simultaneously, it acquires information on wireless access status, link quality, edge resource status, service function status, user mobility status, virtual cell status, and policy constraints. It then associates application intents with network operating status to generate a structured knowledge representation oriented towards intent execution. This structured knowledge representation includes one or more of the following: knowledge augmentation graph, knowledge graph, rule base, state table, semantic ontology, or vectorized knowledge representation. It provides a unified input basis for subsequent network symbol translation and interpretable orchestration.

[0069] The multi-agent collaborative translation layer comprises modules for intent perception, state analysis, network symbol translation, collaborative orchestration, and agent coordination. This layer decomposes application intents into tasks based on structured knowledge representations, translating business objectives, quality of service requirements, resource states, policy constraints, service capabilities, and virtual cell states into machine-recognizable network symbols. These network symbols include one or more of the following: target symbols, constraint symbols, state symbols, resource symbols, service function symbols, policy symbols, and action symbols. These network symbols can further form network service requests, resource control requests, or orchestration control requests. The multi-agent collaborative translation layer also generates service workflows, virtual cell organization schemes, radio resource scheduling strategies, edge resource mapping strategies, and candidate control actions based on the network symbols, ensuring semantic consistency and process interpretability between application intents and network control actions.

[0070] The evidence chain verification and audit interpretation layer includes modules for evidence chain construction, action access verification, consistency verification, causal interpretation, and audit output. This layer verifies network symbols, network service requests, candidate orchestration schemes, resource mapping relationships, and candidate control actions, forming an evidence chain that includes the source of application intent, intent parsing results, structured knowledge basis, network symbol mapping relationships, network service requests, real-time network status, resource availability evidence, virtual cell configuration, policy constraint verification results, service quality prediction results, and action access results. It also explains why the application intent is interpreted in a certain way, why network service requests are generated, why resources are selected, why control actions are granted access, and whether the execution results conform to the original intent, ensuring the verifiability, traceability, and auditability of the intent execution process.

[0071] The orchestration execution and feedback optimization layer comprises modules such as a resource mapping module, a control action execution module, an operational status monitoring module, an anomaly feedback module, and a closed-loop correction module. This layer distributes service workflows, virtual cell organization schemes, wireless resource scheduling strategies, and edge resource mapping strategies verified through the evidence chain to infrastructure resources such as wireless, edge, and cloud for execution. Simultaneously, it continuously collects service quality indicators, link status, edge node load, user mobility status, virtual cell operational status, resource utilization, and anomaly events during execution. Based on feedback results, it dynamically corrects network symbols, network service requests, agent collaboration relationships, service workflows, virtual cell organization schemes, resource mapping relationships, and control actions, achieving closed-loop optimization of intent translation, orchestration execution, evidence verification, and audit interpretation.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for interpretable intent translation in a multi-agent network, characterized in that, Includes the following steps: S1. Receive application intents through the intent awareness and status acquisition layer, perform semantic parsing, context awareness and requirement extraction, and simultaneously collect wireless edge network operation status information to generate structured knowledge representations. S2, the structured knowledge representation generated based on S1, through a multi-agent collaborative translation layer, translates application intent and network state into machine-recognizable network symbols, and generates network service requests, candidate service workflows, virtual cell organization schemes, resource mapping strategies, and candidate control actions. S3. Through the evidence chain verification and audit interpretation layer, using the structured knowledge representation constructed in S1 as the source of evidence, multiple verifications are performed on the network symbols, network service requests, candidate service workflows, virtual cell organization schemes, resource mapping strategies, candidate orchestration schemes, and candidate control actions generated in S2 to construct a complete evidence chain. Based on the verification results, candidate control actions are admitted or rejected to complete causal interpretation and audit output. S4. The verified orchestration scheme is distributed to the network infrastructure for execution through the orchestration execution and feedback optimization layer. The operation status is monitored in real time and feedback data is collected. Based on the feedback results, the network symbols, service workflows, resource mapping strategies and control actions are dynamically and closed-loop corrected. S5. Perform source tracing and interpretation of the entire process of intent understanding, symbol translation, orchestration and execution, verification and auditing, and feedback and correction, and output audit interpretation results.

2. The method for interpretable intent translation in a multi-agent network according to claim 1, characterized in that, The specific content of S1 is as follows: S101, Deployment Intent Awareness and Status Acquisition Layer; S102. Receive application intent in the form of natural language, semi-structured text, interface request, or policy request, wherein the application intent includes at least one of business objectives, service quality requirements, spatiotemporal constraints, service continuity requirements, task priorities, and policy constraints. S103. Perform semantic parsing on the application intent, identify business objectives, service objects, service quality indicators, policy requirements and potential constraints, and generate initial intent understanding results; S104. Obtain user context, terminal status, historical service records, and mobility information, and bind them with the initial intent understanding result to obtain a context-enhanced intent representation; S105. Collect network operation information such as wireless access status, link quality, edge computing / storage resources, network topology, service function status, virtual cell status, and policy constraints; S106. Associate the context-enhanced intent representation with network operation information to generate a structured knowledge representation, wherein the structured knowledge representation is at least one of the following: knowledge enhancement graph, knowledge graph, rule base, state table, semantic ontology, and vectorized knowledge representation. S107. Perform consistency verification on the structured knowledge representation. If there is missing or conflicting information, complete, correct or re-parse it. After the verification is passed, proceed to the network symbol translation process.

3. The method for interpretable intent translation in a multi-agent network according to claim 2, characterized in that, The specific details of S2 are as follows: S201, Multi-agent collaborative translation layer pre-stores agent capability information, network symbol templates, service function information, virtual cell configuration templates, and resource status views; S202. Deploy an intent-aware agent, a state-analysis agent, a network symbol translation agent, a collaborative orchestration agent, and an agent coordination module in the multi-agent collaborative translation layer. Each agent operates according to a preset interface and collaborative relationship. S203. The network symbol translation agent converts structured knowledge representations into network symbols; S204. Based on the target symbol and constraint symbol, decompose the application intent into multiple executable subtasks, and determine the subtask dependencies, execution order and priority; S205. Combining resource symbols, status symbols, service function symbols, and virtual cell symbols, match candidate service functions, radio access resources, edge resources, transmission paths, and virtual cell configurations. S206. Generate a network service request based on network symbols, wherein the network service request includes at least one of the following: quality of service objectives, isolation requirements, service function requirements, resource binding relationships, virtual cell candidate configurations, policy constraints, and candidate control actions; S207. Collaboratively orchestrate intelligent agents to combine candidate service functions and resources, generating several candidate service workflows. The service workflows include the service function call order, intelligent agent collaboration relationship, data transmission path, resource requirements, and execution constraints. S208. Based on user location, wireless coverage, access load, link quality, and service continuity requirements, generate candidate virtual cell organization schemes. S209. Based on the candidate service workflow and virtual cell organization scheme, generate a wireless resource scheduling strategy, an edge resource mapping strategy, and candidate control actions. The candidate control actions include at least one of access control, path adjustment, service function deployment, service migration, resource allocation, load balancing, virtual cell reconstruction, and anomaly handling. S210, the agent coordination module verifies the output results of each agent. If there are symbol or action conflicts, it triggers re-translation or re-arrangement. If there are no conflicts, it outputs candidate arrangement schemes.

4. The method for interpretable intent translation in a multi-agent network according to claim 3, characterized in that, S3 specifically includes: S301. Evidence Chain Verification and Auditing: The intelligent agent obtains the knowledge enhancement graph constructed by the intent perception and state acquisition layer in S1, and the network symbols, network service requests, candidate service workflows, virtual cell organization schemes, resource mapping strategies, candidate orchestration schemes, and candidate control actions obtained by the multi-agent collaborative translation layer in S2, and verifies and audits the content output by S2 based on the knowledge enhancement graph. S302. Taking each candidate orchestration scheme as the retrieval object, obtain the application intent source and intent parsing result corresponding to the candidate orchestration scheme, and retrieve the corresponding business objectives, service objects, functional requirements, service quality requirements, spatiotemporal constraints, real-time network status, network topology relationships, resource availability, service function status, virtual cell configuration and policy constraints from the knowledge enhancement graph. At the same time, obtain the service quality prediction result corresponding to the candidate orchestration scheme to form an evidence set corresponding to the candidate orchestration scheme. S303. For the service quality objectives, isolation requirements, service function requirements, resource binding relationships, virtual cell candidate configurations and policy constraints in the network service request, as well as the service function call order, data transmission path, resource requirements and execution constraints in the candidate service workflow, determine the corresponding supporting evidence from the evidence set, and establish the correspondence between the key content in the network service request and the candidate service workflow and the supporting evidence. S304. Following the processing order of application intent parsing, network symbol mapping, network service request generation, candidate service workflow generation, virtual cell organization scheme generation, resource mapping strategy generation, and candidate control action generation, the inputs, outputs, and generation basis of each processing stage are sequentially associated to form a candidate evidence chain pointing from application intent to candidate control action. S305. Perform semantic consistency verification on network service requests, and determine whether the service function requirements, service quality goals and policy constraints in the network service requests are consistent with the function requirements, service quality requirements and spatiotemporal constraints and policy constraints recorded in the knowledge enhancement graph. S306. Verify the resource feasibility of the candidate control actions and determine whether the radio resources, computing resources, storage resources, transmission resources, service function resources and virtual cell resources required by the candidate control actions meet the execution requirements. S307. Perform policy consistency verification on candidate control actions to determine whether the candidate control actions comply with service level agreements, security policies, priority rules, isolation rules, and time and space constraints. S308. Based on the service quality prediction results in the evidence set, perform service quality verification on the candidate service workflow, virtual cell organization scheme and resource mapping strategy, and determine whether they meet the requirements of latency, reliability, throughput, isolation, service continuity and coverage. S309. When semantic consistency verification, resource feasibility verification, policy consistency verification, and service quality verification all pass, generate action admission results and corresponding action admission basis; when any verification item fails, generate action rejection results including the failed verification item and its rejection reason; associate the verification basis, verification results, action admission results, and action admission basis or action rejection results of each verification with the corresponding network service request, candidate orchestration scheme, and candidate control action, and record them in the candidate evidence chain. S310. Perform integrity verification on the candidate evidence chain, and determine whether the key content in the network service request and candidate service workflow all have corresponding supporting evidence, such as the application intent source, intent parsing result, supporting evidence provided by the knowledge enhancement graph, network symbol mapping relationship, network service request, candidate service workflow, virtual cell organization scheme, resource mapping strategy, candidate orchestration scheme, candidate control action and its generation basis, various verification basis and verification results, and action admission result or action rejection result. Also determine whether the association relationship established according to the processing order described in S304 is continuous and traceable. When all evidence information and verification results are recorded and the association relationship is continuous, a complete evidence chain is formed. Otherwise, output the missing evidence information or the interrupted association relationship. When an action rejection result is generated or the candidate evidence chain is incomplete, the corresponding result is fed back to the multi-agent collaborative translation layer for reprocessing. S311. Using the application intent, network status, resource capabilities, service functions, virtual cell configuration, policy constraints and their relationships recorded in the knowledge enhancement graph as the basis for interpretation, trace back along the complete evidence chain to the generation process of network symbols, network service requests, candidate service workflows, virtual cell organization schemes, resource mapping strategies and candidate control actions, analyze the generation dependencies and constraint relationships between each processing stage, and form a causal explanation for the candidate control actions. S312. Integrate the supporting evidence retrieved from the knowledge enhancement graph, the correspondence between network service requests and supporting evidence, the basis for generating candidate control actions, the basis and results of various verifications, and the basis for action admission or reasons for rejection, and record the integrated information in the complete evidence chain.

5. The method for interpretable intent translation in a multi-agent network according to claim 4, characterized in that, The specific details of S4 are as follows: S401, Deployment, Orchestration, Execution, and Feedback Optimization Layer; S402. Analyze the verified orchestration scheme to determine the service workflow, virtual cell organization scheme, resource mapping relationship, control action set, task execution order and exception handling strategy; S403. According to the resource mapping relationship, bind the service functions and control tasks to the corresponding wireless access node, edge computing node, transmission node, core network node or cloud resource node; S404, issue execution commands for access control, resource allocation, service deployment, link configuration, service migration, load balancing, virtual cell reconstruction, and exception handling; S405: Real-time collection of service quality indicators, wireless link status, edge resource load, node availability, user mobility status, virtual cell operation status, and abnormal events; S406. When a service quality degradation, link fluctuation, resource congestion, service anomaly, service continuity risk, virtual cell failure, or policy change is detected, a feedback event is generated. S407. Associate feedback events with the original application intent, structured knowledge, network symbols, service requests, orchestration schemes, and resource mapping relationships to form feedback evidence; S408. Based on the feedback evidence, dynamically adjust the network symbols, network service requests, agent collaboration relationships, service workflows, virtual cell organization schemes, and resource mapping strategies, and resubmit the adjusted schemes for verification.

6. An apparatus for interpretable intent translation in a multi-agent network, characterized in that, The method described in any one of claims 1-5 includes an intent perception and state acquisition layer, a multi-agent collaborative translation layer, an evidence chain verification and audit interpretation layer, and an orchestration execution and feedback optimization layer. The intent perception and state acquisition layer includes an application intent receiving module, a context perception module, a network state acquisition module, and a structured knowledge generation module. Used to receive and parse application intents, collect network operating status, and generate structured knowledge representations; The multi-agent collaborative translation layer includes an intent-aware agent, a state-analysis agent, a network symbol translation agent, a collaborative orchestration agent, and an agent coordination module. It is used to convert structured knowledge into network symbols, generate network service requests, service workflows, virtual cell organization schemes, resource mapping strategies and candidate control actions, and complete the conflict verification of multi-agent output results; The evidence chain verification and audit interpretation layer includes an evidence chain construction module, an action access verification module, a consistency verification module, a causal interpretation module, and an audit output module; it is used to construct a full-process evidence chain, complete multi-dimensional verification of resources, strategies, and service quality, and realize causal interpretation and audit result output. The orchestration execution and feedback optimization layer includes a resource mapping module, a control action execution module, a running status monitoring module, an anomaly feedback module, and a closed-loop correction module. Used to execute verified orchestration schemes, monitor operational status, collect anomaly feedback, and complete dynamic closed-loop correction across the entire chain.

7. The apparatus for interpretable intent translation in a multi-agent network according to claim 6, characterized in that, The intent perception and status acquisition layer is deployed at the service entry point, network control node, edge control node, or network management platform; the orchestration execution and feedback optimization layer is distributed and deployed in the wireless access network, edge node, core network, and cloud platform.

8. The apparatus for interpretable intent translation in a multi-agent network according to claim 7, characterized in that, The network symbols output by the network symbol translation agent include at least one of the following: target symbol, constraint symbol, state symbol, resource symbol, service function symbol, virtual cell symbol, policy symbol, and action symbol.

9. The apparatus for interpretable intent translation in a multi-agent network according to claim 8, characterized in that, The evidence chain generated by the evidence chain construction module fully records the entire process from application intent input, intent parsing, structured knowledge generation, network symbol mapping, service request generation, resource verification to control action output.

10. The apparatus for interpretable intent translation in a multi-agent network according to claim 9, characterized in that, The closed-loop correction module iteratively adjusts network symbols, network service requests, agent collaboration logic, service workflow, virtual cell organization scheme, and resource mapping strategy based on operational feedback events, and sends the adjustment results back to the evidence chain verification and audit interpretation layer for re-verification.

Images