Network slice management method and device based on intention, equipment and storage medium

Abstract

The invention provides an intention-based network slice management method and device, equipment and a storage medium, and the method comprises the steps: automatically converting the natural language intention of a user into structured information capable of executing operation control through introducing the semantic understanding and reasoning capability of a large language model; full-process automation from intention analysis to slice deployment, adjustment and recovery is achieved, and manual configuration and script dependence are remarkably reduced. Task planning and resource solving are automatically performed according to the structured information and the execution log so as to determine a task execution sequence for network slice management and an optimal resource mapping scheme, resource allocation and performance strategies are automatically optimized according to the physical network topology data, and the utilization efficiency and service quality of network slices are improved. A multi-step invoked task execution sequence is packaged into a single-step invoked coarse-grained application programming interface, so that the invoking frequency is reduced, and the integrity and accuracy of a management process are ensured.

Description

Technical Field

[0001] This application relates to the field of network intelligent management technology, and in particular to an intent-based network slicing management method, apparatus, device and storage medium. Background Technology

[0002] The network management system failed to fully consider the actual operating habits and cognitive load of end users during its design, increasing the user's operational burden and the probability of errors. In the network management model, changes from network resource scheduling to the launch of new equipment often require cumbersome manual intervention and long-cycle management processes, resulting in problems such as difficult user input, high learning barriers, high manual configuration costs, and insufficient exploratory nature in intelligent network management. Summary of the Invention

[0003] In view of this, the purpose of this application is to propose an intent-based network slicing management method, apparatus, device and storage medium. By constructing an intent-driven network management architecture, the entire process of network management can be automatically realized based on the structured information corresponding to the user's natural language intent, thereby achieving simple and efficient network management.

[0004] To achieve the above objectives, this application provides an intent-based network slicing management method, comprising: The user's natural language intent is obtained, and the natural language intent is processed in a structured manner using a large language model to obtain structured information; Based on the structured information and execution logs, task planning is performed to obtain the task execution sequence; Based on the task execution sequence and physical network topology data, resource calculation is performed to obtain the optimal resource mapping scheme; Based on the task execution sequence and the optimal resource mapping scheme, the coarse-grained application programming interface is invoked to perform control operations on the network slices, and the operation results are obtained; the execution log is updated based on the operation results until the structured information is completed.

[0005] Optionally, the step of structuring the natural language intent using a large language model to obtain structured information includes: The key intent elements are extracted from the natural language intent by calling a large language model. The key intent elements are mapped to a structured network configuration description to obtain the structured information.

[0006] Optionally, the step of performing task planning based on the structured information and execution logs to obtain a task execution sequence includes: The structured information is used to extract the target, resulting in the control object, control type, and task objective. The structured information is subjected to constraint mapping and priority assignment to obtain control constraints; Under the constraints of the control, an initial task sequence is obtained by chaining thought based on the control type and the task objective. The current state of the controlled object is determined based on the execution log, and the initial task sequence is modified for feasibility based on the current state to obtain the task execution sequence.

[0007] Optionally, the step of obtaining an initial task sequence by chaining based on the control type and the task objective under the constraints of the control measures includes: The task objective is decomposed into multiple operational subtasks according to the control type; The action sequence constraints and optimal execution path of the operation subtasks are determined based on the control constraints. The operation subtasks are arranged according to the optimal execution path and the action order constraints to obtain the initial task sequence.

[0008] Optionally, the step of performing feasibility correction on the initial task sequence based on the current state to obtain the task execution sequence includes: Perceive the resource requirements for executing the initial task sequence; In response to the current state satisfying the resource requirements, the initial task sequence is determined as the task execution sequence; In response to the current state not meeting the resource requirements, a feasibility correction task is planned based on the current state and the resource requirements of the current state. The feasibility correction task is then embedded into the initial task sequence to obtain the task execution sequence.

[0009] Optionally, the step of solving for resources based on the task execution sequence and physical network topology data to obtain the optimal resource mapping scheme includes: Determine the available resource mapping scheme based on the task execution sequence; The physical network topology data is modeled as a physical network; According to the available resource mapping scheme, the virtual nodes of the network slice are mapped to the physical nodes of the physical network, and the virtual links of the network slice are mapped to the physical paths of the physical network, so as to obtain the set of available nodes and the set of available paths. The optimal resource mapping scheme is obtained by solving the optimal configuration of the available node set and the available path set based on the reinforcement learning algorithm.

[0010] Optionally, based on the task execution sequence and the optimal resource mapping scheme, a coarse-grained application programming interface is invoked to perform control operations on the network slices, obtaining operation results, including: Based on the optimal resource mapping scheme, the operation subtasks in the task execution sequence are encapsulated into coarse-grained application programming interfaces. The coarse-grained application programming interface is invoked to execute all operation subtasks, and the operation result is obtained.

[0011] Based on the same inventive concept, this application also provides an intent-based network slicing management device, comprising: The intent parsing module is configured to: acquire the user's natural language intent, and perform structured processing on the natural language intent using a large language model to obtain structured information; The decision planning module is configured to: perform task planning based on the structured information and execution logs to obtain a task execution sequence; The resource solving module is configured to: solve for resources based on the task execution sequence and physical network topology data to obtain the optimal resource mapping scheme; The slice management module is configured to: call a coarse-grained application programming interface to perform control operations on network slices according to the task execution sequence and the optimal resource mapping scheme, and obtain the operation results; update the execution log according to the operation results until the structured information is completed.

[0012] Based on the same inventive concept, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor implements the method described above when executing the computer program.

[0013] Based on the same inventive concept, this application also provides a non-transitory computer-readable storage medium that stores computer instructions for causing a computer to perform the method described above.

[0014] As described above, the intent-based network slice management method, apparatus, device, and storage medium provided in this application acquire the user's natural language intent, perform structured processing on the natural language intent using a large language model to obtain structured information; perform task planning based on the structured information and execution logs to obtain a task execution sequence; perform resource solving based on the task execution sequence and physical network topology data to obtain an optimal resource mapping scheme; call a coarse-grained application programming interface to perform control operations on the network slice based on the task execution sequence and the optimal resource mapping scheme to obtain operation results; update the execution log based on the operation results until the structured information is complete. By introducing the semantic understanding and reasoning capabilities of a large language model, the user's natural language intent is automatically converted into structured information for executable operation control, realizing full-process automation from intent parsing to slice deployment, adjustment, and recycling, significantly reducing manual configuration and script dependence. Automatic task planning and resource solving based on structured information and execution logs determine the task execution sequence and optimal resource mapping scheme for network slice management, and automatically optimize resource allocation and performance strategies based on physical network topology data, improving the utilization efficiency and service quality of network slices. The multi-step task execution sequence is encapsulated into a single-step coarse-grained application programming interface (API), reducing the number of calls and ensuring the integrity and accuracy of the management process. The execution log is updated based on the operation results of the coarse-grained API calls, providing data support for subsequent intent-driven control. This enables controllable and interpretable closed-loop execution based on intent-driven principles, achieving simple and efficient network management. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of an intent-based network slicing management system according to an embodiment of this application; Figure 2 This is a flowchart of an intent-based network slice management method according to an embodiment of this application; Figure 3 This is a schematic diagram of task planning in an embodiment of this application; Figure 4 This is a schematic diagram of an intent-based network slice management device according to an embodiment of this application; Figure 5 This is a schematic diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0018] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0019] In this article, it is important to understand that any number of elements in the accompanying figures is for illustrative purposes and not for limitation, and any naming is for distinction only and has no limiting meaning.

[0020] Based on the above background description, the following situations also exist in the related technologies: Network management systems often fail to adequately consider end-users' actual operating habits and cognitive load during design, leading to input difficulties and unfriendly interactions. For example, some network management platforms suffer from basic user experience issues such as incompatible input methods and missing drop-down menus. Repeatedly entered data cannot be effectively remembered, resulting in repetitive work. This may stem from legacy technical architecture issues, such as outdated system versions, browser compatibility problems, or independent development of different subsystems with inconsistent data formats and interface protocols, creating "data silos" that hinder smooth information flow. Furthermore, reliance on command-line interfaces (CLI) or complex, fragmented web interfaces increases user workload and the probability of errors. Because network systems lack the ability to automatically identify and understand the intentions of upper-layer applications, they cannot intelligently allocate resources to meet dynamic application needs, thus limiting their ability to manage network slices.

[0021] Network slicing involves creating logically isolated, end-to-end virtualized networks from the same physical infrastructure. This allows for dynamic allocation of network resources on demand, supporting diverse service requirements. Each slice is typically defined as an end-to-end virtual network instance, and each network can be customized according to specific service needs (e.g., bandwidth, computing resources, isolation). Network slicing technology not only makes network configuration and expansion more flexible but also improves network resource utilization. The isolation between network slices allows it to simultaneously meet the needs of different work domains, ensuring tenant information security. Network slicing is an ongoing standardization effort, with different Standards Development Organizations (SDOs) working on each specific part of the network slicing architecture. GSMA, 3GPP, and ESTI have all established relevant standards for network slicing. Currently, most network slicing deployment processes follow the ESTI standard.

[0022] In network management models within related technologies, the process from network resource scheduling changes to the deployment of new equipment often involves cumbersome manual intervention and lengthy cycles. For example, during the network deployment phase, engineers often need to manually configure each device individually. This approach is not only time-consuming and labor-intensive but also highly susceptible to errors due to human negligence. As the network scales up, the workload of maintenance does not increase linearly but rather becomes more significant. A deeper reason lies in the fact that the network remains largely an "open-loop" control system, lacking an effective automated closed-loop mechanism. Many decisions and operational instructions rely on manual issuance, and the system itself cannot adaptively adjust based on intent or strategy. Furthermore, with numerous systems, experience in troubleshooting is difficult to effectively accumulate and reuse in subsequent maintenance, leading to repeated troubleshooting of similar problems and low efficiency.

[0023] In related technologies, a single large language model in a monolithic management system needs to simultaneously perform multiple functions such as intent parsing, task planning, API parameter generation, and resource mapping. This mixed task significantly increases the cognitive burden on the model, leading to increased illusion rates and execution chain errors.

[0024] This application's embodiments effectively separate the two levels of "high-level reasoning and decision-making" and "low-level algorithm execution" through modular division of labor and bounded collaboration mechanisms, enabling the system to perform its respective functions in different professional domains and maintain overall stability.

[0025] In some embodiments, the intent-based network slice management method is applied to, for example, Figure 1The intent-driven modular network management system shown addresses the issues of illusions, parameter errors, and long-chain inference failures that can easily occur with a single large language model in complex network management tasks. By dividing the network slice management process into several sub-tasks with clearly defined responsibilities and configuring a dedicated intelligent module for each sub-task, the system can complete the entire process control from intent understanding to network deployment in a collaborative mode, thereby achieving highly reliable and low-error-rate automated network slice management.

[0026] During system operation, users first input their natural language intent. A lightweight Large Language Model (LLM) performs semantic understanding and structured parsing of the intent, extracting elements such as slice names, operation types, and related resource requirements, transforming them into structured information describing the standardized task. Subsequently, the decision-making module, based on the parsing results and the current system state, plans the task execution sequence and calls different modules to complete the operation. The decision-making module is the core coordinator of the entire system, responsible for reading execution logs, planning task steps, and calling appropriate sub-modules to complete the operation. It does not directly perform numerical calculations or generate configurations; instead, it dynamically adjusts the execution plan based on the intent structure, system feedback, and environmental state, forming a closed-loop control mechanism of "thinking—action—feedback."

[0027] The intent-driven modular network management system consists of a central decision-making model and an access intent input module. The central decision-making model includes an intent parsing module, a (central) decision-making module, a resource solving module, and a slice management system module. The intent input module is used to obtain the user's natural language input and feedback results, and to visualize the controlled network slices to realize the interface display, serving as a window module for direct human-computer interaction with the user.

[0028] The IntentParserLLM module employs an independent large language model that focuses on extracting structured information from the natural language input by the user. It performs structured parsing of natural language intent, extracting elements such as slice name, operation type (create, modify, delete, etc.), and resource requirements (such as bandwidth and latency) from the natural language intent, and converting them into standardized task descriptions to obtain corresponding structured information, providing clear and standardized input for the central decision-making model.

[0029] The decision-making module, acting as the system coordinator and strategy planner, is responsible for task planning, module scheduling, and closed-loop control based on structured information input. It dynamically adjusts the execution plan based on the structured information after intent parsing, system status (obtained through execution logs), and environmental feedback. Through a "think-action-feedback" loop, it achieves adaptive decision-making, avoiding the limitations of static workflows. The resource solver module (VNESolver) performs Virtual Network Embedding (VNE) tasks based on reinforcement learning algorithms, mapping virtual network slices to physical resources while satisfying constraints such as bandwidth, CPU, and latency. This module offloads the complex task of precise value calculation from the decision-making module, effectively preventing the central large model from exhibiting related illusions. Based on reinforcement learning algorithms, it implements the virtual network embedding task, mapping virtual network slice nodes and links to physical network resources. Through a reward mechanism that comprehensively considers constraints such as bandwidth, CPU, and latency, it outputs the optimal resource mapping scheme or returns a failure status, thus avoiding the uncertainty brought by free text output. This module offloads the complex task of precise value calculation from the large decision-making model, effectively preventing the large model from exhibiting related illusions.

[0030] The slice management system module (System APIs) encapsulates underlying SDN controllers, NFV managers, and other components, providing coarse-grained application programming interfaces (such as creating slices and modifying configurations) to execute specific network operations. It reduces the number of model calls through highly abstract interfaces and includes a built-in structured error feedback mechanism. In other words, the slice management system module includes underlying execution components such as software-defined networking (SDN) controllers and configuration managers. These are responsible for specific network state changes, resource allocation, and configuration verification.

[0031] The modules in the intent-driven modular network management system are connected through a structured application programming interface (API), with clear input and output boundaries, maintaining functional decoupling while enabling efficient information flow. The decision-making module, as the system's central hub, is responsible for overall planning and task scheduling; the intent parsing module transforms natural language intents into structured information understandable by the computer; the resource solving module performs quantifiable resource calculations and constraint solving; and the underlying slice management system module interacts with the network control plane to execute specific configuration operations. The collaborative relationships between these modules form an intelligent "planning-execution-feedback" loop, with each module acting as an illusionary isolation unit to prevent errors from cascading through the system. This modular tool combination structurally achieves effective separation between high-level reasoning and low-level execution: the large language model in the intent parsing module determines "what to do," while the decision-making module, through specialized algorithms and control algorithms, is responsible for "how to do it." This design fully leverages the language understanding and task planning advantages of LLM, while relying on external algorithms and execution components to ensure the accuracy and reliability of the results. Through this mechanism, the embodiments of this application significantly reduce the risk of illusions or logical errors in the intent-driven modular network management system in complex tasks, improve the stability, interpretability and scalability of the system, and realize secure, intelligent and autonomous network slice management.

[0032] Specifically, in an intent-driven modular network management system, users submit a high-level natural language intent in natural language. The decision-oriented LLM receives this intent along with a set of API specifications and plans and issues a series of system-level API calls to achieve the desired result. These APIs encapsulate slice management operations, such as CRUD (Create, Read, Update, Delete). Furthermore, the LLM can not only interpret static API documentation but also perform dynamic reasoning based on execution feedback (i.e., responses to previous API calls). This allows it to adjust its behavior in real time and recover from partial failures or unexpected system states. It decouples control flow logic from implementation logic. By exposing system functionality as APIs and delegating control decisions to the LLM, it allows execution paths to be planned at runtime based on the current environment and operational history, improving the adaptability and scalability of the intent-driven modular network management system.

[0033] To support intent-driven automated workflows, the intent-driven modular network management system maintains an execution log that tracks each operation and its corresponding result. The LLM uses this execution log as an evolving context to guide its subsequent decisions. Each API call is not a one-off operation but part of a continuous awareness-action loop: the LLM responds to the API to perceive the environment and continuously refines its understanding accordingly. This architectural paradigm of intent-driven modular network management systems offers several significant advantages: User-centric operation: End users interact with the system through natural language intents, without needing to understand configuration syntax or underlying semantics. The complexity of intent parsing, planning, and execution is hidden behind the LLM control layer.

[0034] Simpler control logic: Network management systems in related technologies require developers to achieve tight coupling between input processing, decision branches, and command execution. The intent-driven modular network management system provided in this application integrates these aspects into a unified, model-driven pipeline, shifting from procedural control to declarative reasoning.

[0035] Scalability: Developers can add new features to the system simply by registering the new API and updating the documentation. No control logic needs to be rewritten, as the LLM learns to call new APIs through timely updates.

[0036] Furthermore, the intent-driven modular network management system in this embodiment adopts an API design strategy oriented towards illusion control. The API design of the intent-driven modular network management system is a key factor in achieving stable control and anti-illusion operation of the large language model. The interaction between the central decision-making large model and the external environment is entirely completed through the system API. The API is not only the sole means of performing slice management operations but also the channel through which the system returns status and results feedback to the model. The quality of the structured design of the API directly affects the planning accuracy of the large model and the overall robustness of the system. If the API granularity is too fine or the feedback mechanism is imperfect, it can easily lead to illusion problems such as model misunderstanding, parameter confusion, or execution chain breaks. Therefore, this embodiment significantly improves the reliability and interpretability of the large model in network slice control by abstracting the interface granularity and using a structured feedback mechanism.

[0037] Related technologies employ fine-grained APIs that correspond one-to-one with underlying system operations, such as defining interfaces for "adding nodes," "allocating resources," and "establishing links." While this design theoretically provides greater operational flexibility, it reveals significant drawbacks in practice: the model needs to maintain a long-range context across multiple calls, significantly increasing the planning burden, and it is highly susceptible to misunderstandings in intermediate states, leading to incomplete configurations or incorrect ordering.

[0038] To overcome this problem, the intent-driven modular network management system provided in this application improves the API into coarse-grained, highly abstract operation interfaces. Each API encapsulates a complete logical process, such as "create slice," "modify slice topology," and "query slice status." This design encapsulates multi-step low-level operations into single-step calls, with the system internally implementing the specific execution process, allowing the central decision-making model to focus on high-level intent planning rather than low-level implementation details. For example, the system's `create_slice` interface can complete intent parsing, resource mapping, and topology generation in a single call; the `configure_slice` interface handles configuration updates and parameter verification; and the `get_slice_config` interface returns the complete slice configuration status and metadata snapshot for the model to perform contextual reasoning. Through this encapsulation, the system significantly reduces the number of calls required for the large model to complete tasks, thereby reducing the risk of planning errors and parameter misuse. Simultaneously, each API has a structured error feedback mechanism; when execution fails, the system returns a clear failure type, reason, and suggestion. For example, when `configure_slice` fails due to missing node information, the model can infer that a topology modification operation needs to be performed first.

[0039] Using coarse-grained APIs has the following benefits: First, it reduces cognitive load by lowering the level of interface abstraction, thus reducing the number of detailed decisions that large models need to make in a single task. Second, it limits fault propagation by avoiding inconsistencies in system state caused by partial execution through atomic and stateless interface design. Third, it explicitly exposes system state by returning a complete configuration snapshot and metadata during queries, rather than fragmented results, ensuring that the model can reason based on global information. Finally, it avoids open-ended output by requiring the interface to return results in a deterministic format, preventing the model from interpreting ambiguous textual results and thus preventing the spread of inference ambiguity and illusions.

[0040] In summary, the intent-driven modular network management system provided in this application implements a "controllable interaction boundary" for large language models. While retaining the high-level language understanding capabilities of LLM, it establishes a reliable anti-illusion mechanism through structured interface design. The system significantly improves planning consistency and execution success rate in complex tasks, and significantly reduces parameter errors and logical deviations caused by illusions, thereby enabling large models to achieve stable and predictable control behavior in critical network infrastructure management scenarios.

[0041] The intent-based network slice management method provided by the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0042] In some embodiments, such as Figure 2 As shown, the intent-based network slice management method includes: Step 201: Obtain the user's natural language intent, and perform structured processing on the natural language intent through a large language model to obtain structured information.

[0043] In practice, after obtaining the user's natural language intent, a large language model and an Intent Based Network (IBN) are used for structured processing.

[0044] Intent-based networking offers a novel paradigm for network services, enhancing the capabilities of autonomous networks by enabling users to define desired network outcomes (intents) at a high level of abstraction. These intents communicate goals and constraints to the Network Management System (NMS). The NMS then generates the necessary low-level configurations to achieve the specified results. During intent translation, high-level user intents are transformed into low-level configurations that the NMS can execute. This abstraction minimizes the complexity of low-level configuration details by providing a standardized interface to express high-level user intents (called the northbound interface). The NMS's Network Service Intent (NBI) uses a human-readable language to represent structured information, such as JSON or YAML. For example, in the ETSI standard, this high-level intent is defined using a Network Service Descriptor (NSD), which is in JSON format and designed to deploy network services encompassing multiple applications.

[0045] Large Language Models (LLMs), pre-trained on massive amounts of text and code corpora, possess powerful natural language understanding and generation capabilities. These capabilities enable LLMs to extract key intent elements (such as service type, performance constraints, and topology requirements) from user-input natural language intent and automatically map them into structured network configuration descriptions, thus achieving intelligent parsing from "language understanding" to "configuration generation." Consequently, LLMs become the crucial "decision center" in the IBN system, connecting user intent with the underlying network management logic.

[0046] Specifically, the natural language intent is structured using a large language model to obtain structured information, including: Key intent elements are extracted from natural language intent by calling a large language model. Key intent elements are mapped to structured network configuration descriptions, resulting in structured information.

[0047] In practical implementation, taking the natural language intent of transferring the computing resources of node 2 in network slice 1 to node 35 as an example, the natural language intent involves the following key elements: Slice identifier: Slice 1 (a specific network slice instance). Operation type: Modification (computation resource adjustment). Resource change: Reallocating the computing resources (such as CPU and memory) of node 2 to node 35. Implicit requirements: Ensuring slice performance isolation and meeting resource constraints (such as bandwidth and latency not being affected).

[0048] By leveraging the semantic understanding capabilities of a large language model, structured intent information is extracted from the input natural language intent. For example, the intent type `intent_type` is `modify_slice`, indicating modification of a network slice. The slice identifier `slice_id` is `slice_1`, representing slice 1. The action is `reallocate_resource`, indicating resource reallocation. The source node of the action is `node_node`, representing node 2, and the target node is `node_35`, representing node 35. The resource type `resource_type` is `compute`, indicating computational resources. The intent information is configured as structured information that standardizes the task description, completing the intent parsing. Taking JSON format as an example, the output structured information is as follows: { "intent_type": "modify_slice", "slice_id": "slice_1", "action": "reallocate_resource", "source_node": "node_2", "target_node": "node_35", "resource_type": "compute } The intent parsing process of the large language model is a preprocessing process that transforms ambiguous natural language intents into explicit machine-readable instructions, i.e., standardized task description structured information, thereby reducing the cognitive burden on the decision-making module.

[0049] Step 202: Perform task planning based on structured information and execution logs to obtain the task execution sequence.

[0050] In practice, the decision-making module combines structured information and the current system state (obtained by querying the execution log) to plan tasks, resulting in the following sub-tasks: Context loading: Query the execution log to obtain the current status of slice 1 (such as node resource utilization and topology).

[0051] Feasibility assessment: The feasibility of the operation is determined based on the execution log. For example, check whether node 2 belongs to slice 1, whether node 35 belongs to slice 1, whether node 35 is available, and whether the resource adjustment violates the isolation constraint.

[0052] Execution sequence planning: Generates a sequence of steps, that is, imposes sequential constraints on the operation subtasks, such as first calling the resource solving module to calculate the optimal resource mapping scheme, and then triggering the modification API. It can also plan alternative paths (e.g., redirecting to node 40 when node 35 is unavailable).

[0053] Dynamic adjustment methods: Real-time optimization of planning through a perception-action cycle. For example, if the execution log shows that node 35 has been under high load recently, the model may prioritize querying the node status of node 35 before making subsequent decisions.

[0054] Data flow: Planning results are delivered in the form of a sequence of tool calls, represented as a sequence of task executions, such as calls to VNESolver and system APIs.

[0055] Step 203: Solve for resources based on the task execution sequence and physical network topology data to obtain the optimal resource mapping scheme.

[0056] In practice, the decision-making module calls the resource-solving module to perform resource solving and optimization mapping. The resource-solving module processes resource allocation based on reinforcement learning algorithms. Input processing: Receives resource constraints (such as computational resource requirements and inter-node link requirements) and combines them with the physical network state determined by the physical network topology data (obtained from the SDN controller).

[0057] Optimize computation: Use reinforcement learning algorithms (such as deep Q-networks) to simulate resource reallocation scenarios and optimize objectives (such as minimizing latency and balancing load) using reward functions. Output the optimal mapping scheme, for example, transferring 50% of the CPU resources of node 2 to node 35 while adjusting the link bandwidth.

[0058] Constraint verification: Ensure that the solution does not violate the Service Level Agreement (SLA), such as isolation, and return the error reason when it fails.

[0059] Reinforcement learning algorithms are used to replace the numerical computation of the central large model, sharing the precise computational tasks of the decision-making module and preventing the central decision-making large model from exhibiting illusions during numerical optimization. The output is a deterministic optimal resource mapping scheme, rather than natural text, and further, quantifiable output is used to prevent illusions.

[0060] For example, the resource resolution module simulates a resource reallocation scenario based on a reinforcement learning algorithm, considering the available resources of target node 35 and the SLA requirements of slice 1 (such as latency limits). It outputs the optimal mapping scheme: for example, migrating 50% of the CPU resources of node 2 to node 35 while ensuring balanced link bandwidth. If the resolution fails (e.g., node 35 has insufficient resources), a structured error is returned (e.g., {"status": "error", "reason": "target_node_overloaded"}). This specialized reinforcement learning algorithm avoids large models directly handling numerical calculations, preventing illusions (such as unreasonable resource allocation).

[0061] Step 204: Based on the task execution sequence and the optimal resource mapping scheme, call the coarse-grained application programming interface to perform control operations on the network slices and obtain the operation results; update the execution log based on the operation results until the structured information is completed.

[0062] In practice, the decision-making module calls a coarse-grained API (such as modify_slice_config) to perform actual network operations, and the process of obtaining the operation results is shown in the following example.

[0063] In some embodiments, a coarse-grained application programming interface is invoked to perform control operations on network slices based on the task execution sequence and the optimal resource mapping scheme, and the operation results are obtained, including: Based on the optimal resource mapping scheme, the operation subtasks in the task execution sequence are encapsulated into coarse-grained application programming interfaces; Call the coarse-grained application programming interface to execute all operation subtasks and obtain the operation results.

[0064] In practice, the coarse-grained API employs a highly abstract interface, encapsulating multiple low-level subtasks (such as resource quota modification, topology update, and configuration verification) in a single call. Internally, the coarse-grained API triggers the SDN controller to adjust node resource bindings and the NFV manager to reallocate virtual resources. The coarse-grained API executes all atomic operations according to sequential constraints, ensuring state consistency. For example, it might first reduce resources on node 2, then increase resources on node 35, and roll back failed operations if a subtask fails. After execution, the coarse-grained API returns a structured result, including a success status or error code; for example, a new configuration snapshot is returned on success, and the error type (e.g., resource_conflict) is returned on failure. Feedback includes suggestions to support system self-correction. The coarse-grained API design follows an illusion control strategy, completing multiple operations in a single call, reducing model decision complexity. Using the coarse-grained API reduces the number of calls to the fine-grained API, provides explicit error feedback, and avoids the error-prone issues of open text parsing.

[0065] Finally, automatic correction is achieved through feedback loop and lifecycle management. The system records operation results through execution logs, forming a closed loop. The execution log is updated to record details of the operation (such as operation ID, resource changes, and timestamp) for subsequent queries. The central decision-making model adjusts subsequent behavior based on the updated execution logs. For example, if resource adjustments are successful, a monitoring API can be triggered to verify performance; if they fail, a new task execution sequence is replanned based on the error.

[0066] In summary, the intent-based network slice management method provided in this application has significant advantages in terms of high automation, strong stability, and good scalability. By introducing the semantic understanding and reasoning capabilities of a large language model, the system can automatically convert users' natural language intents into executable operations, achieving full-process automation from intent parsing to slice deployment, adjustment, and recycling, significantly reducing manual configuration and script dependence. A modular design is adopted, dividing complex tasks into modules such as intent parsing, resource allocation solving, and system execution, coordinated and controlled by a central decision-making model. This reduces the reasoning burden on the central decision-making model and effectively mitigates the risk of illusion. The system API design follows the principles of high abstraction and structured feedback, encapsulating multi-step operations into single-step calls and returning a clear error type upon failure, achieving controllable and interpretable closed-loop execution. This design enables the system to maintain high reliability and robustness in complex network environments, automatically optimizing resource allocation and performance strategies based on the operating status, improving the utilization efficiency and service quality of network slices. The overall solution achieves intelligent, autonomous, and low-cost network slice management.

[0067] In some embodiments, such as Figure 3 As shown, task planning is performed based on structured information and execution logs to obtain the task execution sequence, including: Step 301: Extract the target from the structured information to obtain the control object, control type, and task objective.

[0068] In practical implementation, the task planning module of the decision-making module is the core of the system's intelligent decision-making. It is responsible for transforming structured information into an executable sequence of operations and achieving adaptive planning through dynamic context awareness. Located at the central hub of the system, the decision-making module receives structured information (such as JSON-formatted task descriptions) from the intent parsing module and outputs detailed task execution sequences (such as API call plans). Its core function is to decompose high-level intents into ordered low-level operational steps, ensuring logical rationality and resource feasibility. It integrates real-time system status (such as network resource utilization and operation history) for decision-making, avoiding the limitations of static planning. Through a perception-action cycle, it adapts to environmental changes, achieving closed-loop autonomy.

[0069] The task planning process is a concrete manifestation of the "intention-driven control" concept, which elevates the large language model from a pure language processor to a "decision brain" for network management.

[0070] Taking the natural language intent of allocating computing resources from node 2 in network slice 1 to node 35 as an example, the structured information corresponding to the natural language intent is as follows: { "intent_type": "modify_slice", "slice_id": "slice_1", "action": "reallocate_resource", "source_node": "node_2", "target_node": "node_35", "resource_type": "compute } By identifying the core content of the structured information (such as "resource reallocation") and ignoring minor modifiers, we can determine that the control object is the source node node_2, the target node node_35, and the control type is modifying the computational resources of a network slice. The task objective is to allocate the computational resources of node 2 to node 35, resulting in an abstract task objective, such as `reallocate_compute_resource(slice_1, node_2, node_35)`.

[0071] Step 302: Perform constraint mapping and priority assignment on the structured information to obtain control constraints.

[0072] In practical implementation, constraint mapping represents the transformation of natural language constraints into machine-readable parameters. For example, "high bandwidth" is mapped to bandwidth >= 100Mbps, and "half of the computing resources" is mapped to value = 50 or basis: half_of_source (meaning based on half of the source node). Priority is assigned based on the urgency of the intent (such as the implicit word "immediately") to determine whether the task needs to be executed first. This implements the constraint limitations for task planning.

[0073] Step 303: Under the constraints of control, perform chain thinking based on control type and task objective to obtain the initial task sequence.

[0074] In practice, an initial task sequence is generated based on the control type and task objective. A chain-of-thought technique is employed to decompose complex objectives into operational subtasks (e.g., "resource query → mapping calculation → configuration update"). The input is the output of the intent parsing module (e.g., {"intent_type": "modify_slice", "slice_id": "slice_1", ...}), which parses the operation type and constraints. The output is a list of steps, which constitutes the initial task sequence. Each step includes the action type (e.g., call_vne_solver), parameters, and expected result.

[0075] Within this framework, under control constraints, a chain-like approach is used based on the control type and task objectives to obtain the initial task sequence, which includes: The task objective is broken down into multiple operational subtasks based on the control type; Determine the action sequence constraints and optimal execution path of the operation subtasks based on the control constraints; The operation subtasks are arranged according to the optimal execution path and action sequence constraints to obtain the initial task sequence.

[0076] In practice, the first step is to decompose the actions, breaking down the task objective into multiple sub-tasks based on the control type. For example: Call get_node_status to query the resources of node 35.

[0077] Call vne_solver to calculate the optimal mapping.

[0078] The modify_slice_config function is invoked to perform the adjustments.

[0079] Next, determine the order constraints between actions (e.g., queries must be performed before calculations). Optimal execution paths are determined through path optimization, using a graph search algorithm to select the shortest path and reduce the number of calls.

[0080] Finally, the subtasks are arranged according to the optimal execution path and action order constraints to obtain the initial task sequence (such as an action list and dependency graph).

[0081] Step 304: Determine the current state of the controlled object based on the execution log, and make feasibility corrections to the initial task sequence based on the current state to obtain the task execution sequence.

[0082] In practice, the Context Manager tool is used to manage and integrate the system's real-time status. The current status is obtained by querying the execution logs (a structured historical database of operations), such as: the current resource allocation status of slice 1 (e.g., node 2's CPU utilization is 70%), recent operation records (e.g., the time of the most recent failure of node 35), and network topology snapshots (synchronized from the SDN controller). The execution logs are stored in a time-series database format, containing fields such as operation_id, slice_id, resource_changes, timestamp, and status.

[0083] The feasibility modification process is shown in the following example.

[0084] In some embodiments, the initial task sequence is modified for feasibility based on the current state to obtain a task execution sequence, including: Perceive the resource requirements for executing the initial task sequence; In response to the current state satisfying resource requirements, the initial task sequence is determined as the task execution sequence; In response to the fact that the current state does not meet the resource requirements, a feasibility correction task is planned based on the current state and the resource requirements of the state. The feasibility correction task is then embedded into the initial task sequence to obtain the task execution sequence.

[0085] In practical implementation, feasibility modifications include the following: The validator is invoked to evaluate the feasibility of the initial task sequence and prevent invalid operations. For example: syntax checks ensure parameter formats conform to API specifications (e.g., node ID existence); semantic checks verify resource constraints (e.g., whether node 35 supports computational resource adjustments); conflict detection identifies mutual exclusion with ongoing operations (e.g., avoiding simultaneous modification of the same node). Based on a rule engine and lightweight inference model, it quickly returns boolean results or correction suggestions.

[0086] During the resource constraint verification process, if the current state meets the resource requirements for executing the initial task sequence, it means that no additional resource adjustment subtasks are needed to support the execution of the initial task sequence, and the initial task sequence is directly determined as the task execution sequence. If the current state does not meet the resource requirements for executing the initial task sequence (e.g., node 35's computing resource usage exceeds 80%, and there is insufficient remaining control to adjust node 2's computing resources to node 35), it means that resource adjustment is required first (e.g., releasing some of node 35's computing resources). Based on the current state and resource requirements, a feasibility correction task is planned (e.g., releasing node 35's computing resources until the usage is below 50%), and the feasibility correction task is embedded into the initial task sequence to obtain the task execution sequence. By updating the task sequence in real time, errors during execution are automatically corrected, achieving closed-loop control.

[0087] Finally, the Executor Coordinator is invoked to schedule the actual execution of the initial task sequence: Tool Invocation: Triggers the resource solver module or system API based on the initial task sequence.

[0088] Timeout control: Set the maximum waiting time to avoid indefinite blocking.

[0089] Error handling: Capture module failure signals and initiate retry or alternative paths.

[0090] In some embodiments, resource solving is performed based on the task execution sequence and physical network topology data to obtain an optimal resource mapping scheme, including: Determine available resource mapping schemes based on task execution sequence; Model the physical network topology data as a physical network; Based on the available resource mapping scheme, the virtual nodes of the network slice are mapped to the physical nodes of the physical network, and the virtual links of the network slice are mapped to the physical paths of the physical network, thus obtaining the set of available nodes and the set of available paths. The optimal resource mapping scheme is obtained by solving the optimal configuration of the available node set and available path set based on the reinforcement learning algorithm.

[0091] In practice, the task execution sequence clearly defines the operation type (e.g., "modify slice"), resource requirements (e.g., "computing resources"), and constraints (e.g., "latency ≤ 10ms"). Physical network topology data indicates the current network status, including node resource availability and link connectivity.

[0092] The resource solving process is like receiving a "construction list" (task execution sequence) and a "construction site status map" (physical network topology data), which the solver needs to use to plan resource allocation.

[0093] The core operation of resource solving is to transform user requirements into mathematical optimization problems: Virtual node mapping: Selects which physical node in the physical network will carry the network slice virtual node. Selects which physical path in the physical network will carry the network slice virtual link. Resource allocation: Calculate the specific allocation of CPU, memory, etc.

[0094] Reinforcement learning algorithms simulate multiple mapping schemes based on the task execution sequence to obtain a set of available nodes and a set of available paths. The simulation process needs to balance factors such as resource utilization, load balancing, and cost. The optimal solution (e.g., the shortest path) is selected as the optimal resource mapping scheme.

[0095] Resource solving is completed by the decision module calling the resource solving module. The resource solving module acts as a translator between the decision module and the physical network, avoiding errors that may occur if the decision module directly handles complex calculations.

[0096] It should be noted that the method in this embodiment can be executed by a single device, such as a computer or server. The method can also be applied in a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method in this embodiment, and the multiple devices will interact with each other to complete the method described.

[0097] It should be noted that the above description describes some embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0098] Based on the same inventive concept, corresponding to any of the above embodiments, this application also provides an intent-based network slice management device.

[0099] refer to Figure 4 The intent-based network slice management device includes: The intent parsing module 10 is configured to: obtain the user's natural language intent, perform structured processing on the natural language intent through a large language model, and obtain structured information; The decision planning module 20 is configured to: perform task planning based on structured information and execution logs to obtain a task execution sequence; The resource solving module 30 is configured to: solve for resources based on the task execution sequence and physical network topology data to obtain the optimal resource mapping scheme; The slice management module 40 is configured to: call the coarse-grained application programming interface to perform control operations on the network slices according to the task execution sequence and the optimal resource mapping scheme, and obtain the operation results; update the execution log according to the operation results until the structured information is completed.

[0100] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware.

[0101] The apparatus of the above embodiments is used to implement the corresponding intent-based network slice management method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0102] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the intent-based network slicing management method described in any of the above embodiments.

[0103] Figure 5 This embodiment illustrates a more specific hardware structure of an electronic device. The device may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.

[0104] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0105] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0106] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0107] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0108] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.

[0109] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0110] The electronic devices described above are used to implement the corresponding intent-based network slice management methods in any of the foregoing embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0111] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the intent-based network slice management method as described in any of the above embodiments.

[0112] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0113] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the intent-based network slice management method as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0114] Based on the same concept, corresponding to the methods of any of the above embodiments, this application also provides a computer program product, including computer program instructions, which, when run on a computer, cause the computer to execute the intent-based network slice management method as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0115] It is understood that before using the technical solutions of the various embodiments in this application, users will be informed of the type, scope of use, and usage scenarios of the personal information involved in an appropriate manner, and user authorization will be obtained.

[0116] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose, based on the prompt message, whether to provide personal information to the software or hardware such as electronic devices, applications, servers, or storage media performing the operations described in this application.

[0117] As an optional but not limited implementation, in response to a user's active request, sending a prompt message to the user can be done via a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0118] It is understood that the above notification and user authorization process is merely illustrative and does not limit the implementation of this application. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this application.

[0119] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application is limited to these examples; under the concept of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0120] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0121] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0122] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the claims of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. An intent-based network slice management method, characterized in that, include: The user's natural language intent is obtained, and the natural language intent is processed in a structured manner using a large language model to obtain structured information; Based on the structured information and execution logs, task planning is performed to obtain the task execution sequence; Based on the task execution sequence and physical network topology data, resource calculation is performed to obtain the optimal resource mapping scheme; Based on the task execution sequence and the optimal resource mapping scheme, the coarse-grained application programming interface is invoked to perform control operations on the network slices, and the operation results are obtained; the execution log is updated based on the operation results until the structured information is completed.

2. The intent-based network slice management method according to claim 1, characterized in that, The process of structuring the natural language intent using a large language model to obtain structured information includes: The key intent elements are extracted from the natural language intent by calling a large language model. The key intent elements are mapped to a structured network configuration description to obtain the structured information.

3. The intent-based network slice management method according to claim 1, characterized in that, The step of planning tasks based on the structured information and execution logs to obtain a task execution sequence includes: The structured information is used to extract the target, resulting in the control object, control type, and task objective. The structured information is subjected to constraint mapping and priority assignment to obtain control constraints; Under the constraints of the control, an initial task sequence is obtained by chaining thought based on the control type and the task objective. The current state of the controlled object is determined based on the execution log, and the initial task sequence is modified for feasibility based on the current state to obtain the task execution sequence.

4. The intent-based network slice management method according to claim 3, characterized in that, The initial task sequence is obtained by chaining thought based on the control type and the task objective under the constraints of the control. The task objective is decomposed into multiple operational subtasks according to the control type; The action sequence constraints and optimal execution path of the operation subtasks are determined based on the control constraints. The operation subtasks are arranged according to the optimal execution path and the action order constraints to obtain the initial task sequence.

5. The intent-based network slice management method according to claim 3, characterized in that, The step of performing feasibility correction on the initial task sequence based on the current state to obtain the task execution sequence includes: Perceive the resource requirements for executing the initial task sequence; In response to the current state satisfying the resource requirements, the initial task sequence is determined as the task execution sequence; In response to the current state not meeting the resource requirements, a feasibility correction task is planned based on the current state and the resource requirements of the current state. The feasibility correction task is then embedded into the initial task sequence to obtain the task execution sequence.

6. The intent-based network slice management method according to claim 1, characterized in that, The step of solving for resources based on the task execution sequence and physical network topology data to obtain the optimal resource mapping scheme includes: Determine the available resource mapping scheme based on the task execution sequence; The physical network topology data is modeled as a physical network; According to the available resource mapping scheme, the virtual nodes of the network slice are mapped to the physical nodes of the physical network, and the virtual links of the network slice are mapped to the physical paths of the physical network, so as to obtain the set of available nodes and the set of available paths. The optimal resource mapping scheme is obtained by solving the optimal configuration of the available node set and the available path set based on the reinforcement learning algorithm.

7. The intent-based network slice management method according to claim 1, characterized in that, Based on the task execution sequence and the optimal resource mapping scheme, a coarse-grained application programming interface is invoked to perform control operations on network slices, obtaining the operation results, including: Based on the optimal resource mapping scheme, the operation subtasks in the task execution sequence are encapsulated into coarse-grained application programming interfaces. The coarse-grained application programming interface is invoked to execute all operation subtasks, and the operation result is obtained.

8. An intent-based network slice management device, characterized in that, include: The intent parsing module is configured to: acquire the user's natural language intent, and perform structured processing on the natural language intent using a large language model to obtain structured information; The decision planning module is configured to: perform task planning based on the structured information and execution logs to obtain a task execution sequence; The resource solving module is configured to: solve for resources based on the task execution sequence and physical network topology data to obtain the optimal resource mapping scheme; The slice management module is configured to: call a coarse-grained application programming interface to perform control operations on network slices according to the task execution sequence and the optimal resource mapping scheme, and obtain the operation results; update the execution log according to the operation results until the structured information is completed.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 7.

Images