An extensible operation and maintenance intelligent agent system, method and related device
By using a scalable operations and maintenance intelligent agent system and leveraging natural language request parsing and user confirmation mechanisms, the scalability and security issues of existing operations and maintenance technologies are resolved, enabling secure and controllable execution of operations and maintenance tasks and improving operational efficiency and adaptability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CLP GREAT WALL INTERNET SYST APPL GUANGDONG CO LTD
- Filing Date
- 2026-01-29
- Publication Date
- 2026-06-09
AI Technical Summary
Existing operation and maintenance automation technologies are insufficient in terms of scalability, task identification capabilities, and security. They are difficult to adapt to rapidly changing system architectures and business needs, and lack effective pre-execution verification mechanisms, which can easily lead to misoperation and security risks.
A scalable operation and maintenance intelligent agent system is adopted. The system receives natural language requests through the user interaction module, performs semantic parsing in conjunction with the task classification and recognition module, generates task categories, introduces them into the user confirmation module, generates a visual operation preview, and generates task execution permission based on user confirmation. The system self-update module dynamically updates the task classification prompt dictionary and operation and maintenance tools.
It enables secure, controllable, and automated execution of operation and maintenance tasks, improves operation and maintenance efficiency, reliability, and system scalability, reduces reliance on human experience, enhances the ease of use and security of operation and maintenance operations, and strengthens adaptability to new operation and maintenance needs and complex scenarios.
Smart Images
Figure CN122173357A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of automated operation and maintenance technology, and in particular to a scalable operation and maintenance intelligent agent system, method and related equipment. Background Technology
[0002] In related technologies, with the widespread application of cloud computing platforms, microservice architectures, and distributed information systems, the scale and complexity of enterprise information systems continue to increase. Operations and maintenance (O&M) has become a key technical means to ensure stable system operation and business continuity. Existing O&M systems typically monitor, diagnose, and maintain servers, network devices, application services, and system resources through automated O&M platforms, scripting tools, or workflow orchestration engines. These systems often rely on predefined O&M scripts, rule templates, or fixed workflows, which are configured by O&M personnel based on their existing experience, thereby reducing manual operation costs and improving O&M efficiency to some extent.
[0003] However, existing automated operations and maintenance (O&M) technologies still have many shortcomings in practical applications. On the one hand, O&M tasks are mostly based on predefined scripts or fixed processes. Adding or changing tasks requires manual development and deployment, resulting in poor scalability and difficulty in adapting to rapid changes in system architecture and business needs. On the other hand, in the process of parsing O&M requests and matching tasks, existing solutions mostly use static rules or simple classification models, lacking the ability to effectively identify complex semantics and unknown tasks. This can easily lead to inaccurate task identification or failure to cover new O&M scenarios, thus affecting O&M response efficiency. In addition, existing automated O&M systems often lack a sound pre-execution verification and user confirmation mechanism in the task execution phase. Once an O&M command is triggered, it is executed directly, which can easily lead to misoperation due to misunderstanding of the command or configuration errors, posing a high risk to system operation and security.
[0004] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention
[0005] The main objective of this application is to propose a scalable operation and maintenance intelligent agent system, method, and related equipment to achieve safe, controllable, and automated execution of operation and maintenance tasks, thereby improving operation and maintenance efficiency, reliability, and system scalability.
[0006] To achieve the above objectives, one aspect of this application proposes a scalable operation and maintenance intelligent agent system, the system comprising:
[0007] The user interaction module is used to receive maintenance requests input by users in natural language form, and output the task identification result, execution preview information and confirmation information corresponding to the maintenance request. The task classification and recognition module is used to perform semantic analysis on the operation and maintenance request based on a preset task classification prompt word library, and determine the task category corresponding to the operation and maintenance request or determine it as an unknown task. The user confirmation module is used to generate corresponding operation and maintenance operation preview information after the task category is determined, and to generate task execution permission information after receiving the user confirmation instruction. The task execution module is used to invoke pre-registered operation and maintenance tools to execute corresponding operation and maintenance tasks based on the task execution license information and the task category, and output the task execution results; The system self-update module is used to dynamically update the task classification prompt dictionary and the operation and maintenance tools based on the task execution results.
[0008] In some embodiments, the task classification and recognition module includes: A semantic parsing unit is used to perform semantic parsing on the operation and maintenance request and generate a vectorized representation; The similarity calculation unit is used to calculate the similarity between the vectorized representation and the task vector in the task classification prompt word library using a lightweight fine-tuning model. The classification and determination unit is used to determine whether the similarity score results meet the preset threshold conditions in order to determine the corresponding operation and maintenance task category.
[0009] In some embodiments, the semantic parsing of the maintenance request to generate a vectorized representation includes: The original natural language text of the maintenance request is segmented and semantically parsed to extract semantic feature information that represents the maintenance intent; Based on the preset big oracle model, the semantic feature information is encoded and the operation and maintenance request is mapped to a unified semantic space to generate a corresponding request vector. The request vector is subjected to dimensional normalization and semantic alignment to generate a vectorized representation.
[0010] In some embodiments, employing a lightweight fine-tuning model to calculate the similarity between the vectorized representation and the task vectors in the task classification prompt vocabulary includes: Lightweight fine-tuning models are built based on pre-trained large language models; Obtain the structured prompt text corresponding to each task category in the task classification prompt word library; The structured prompt word text is semantically encoded based on a preset large language model encoder to generate a corresponding task vector; The vectorized representation and the task vector are aligned in the vector space to perform a unified semantic space mapping. Within the unified semantic space, based on the lightweight fine-tuning model, similarity measurement operations are performed on the vectorized representation and each of the task vectors to generate corresponding similarity scores.
[0011] In some embodiments, determining whether the similarity score meets a preset threshold condition to determine the corresponding operation and maintenance task category includes: Obtain the similarity score result; The similarity scores are sorted to determine the candidate task category corresponding to the highest similarity score; Determine whether the highest similarity score is greater than or equal to the preset threshold; When the highest similarity score is determined to meet the preset threshold condition, the candidate task category is determined as the target operation and maintenance task category corresponding to the operation and maintenance request, and the corresponding task identification result is output. When the highest similarity score is determined not to meet the preset threshold condition, the maintenance request is marked as an unknown task, and abnormal identification information is generated to prompt the user or trigger the system self-update module.
[0012] In some embodiments, the step of invoking a pre-registered operation and maintenance tool to execute a corresponding operation and maintenance task based on the task execution license information and the task category, and outputting the task execution result, includes: Based on the task execution license information and the task category, the target operation and maintenance tool is determined by matching from the pre-registered operation and maintenance tool set; Before invoking the target operation and maintenance tool, perform a tool validity and parameter compliance check on the target operation and maintenance tool; If the verification passes, the target operation and maintenance tool is invoked to perform the corresponding operation and maintenance operation based on preset security constraints; the security constraints include execution step limit, execution time limit and tool permission verification. During the operation and maintenance process, the execution status is monitored, and the execution is terminated when the preset termination conditions are met. After the operation and maintenance is completed, the task execution result returned by the target operation and maintenance tool is obtained and output.
[0013] In some embodiments, a security enhancement design module is also included for implementing hierarchical security controls during the execution of the operation and maintenance tasks.
[0014] In some embodiments, the security enhancement design module includes: The risk assessment unit is used to assess the risk level of the operation and maintenance task based on the risk level information associated with the operation and maintenance task in the task classification prompt word library. The confirmation control unit is used to generate a corresponding user confirmation strategy and perform user confirmation based on the risk level; the user confirmation strategy includes one of single confirmation, double confirmation, or approval chain confirmation. The permission verification unit is used to verify the operation permissions of the executing entity and the calling permissions of the target operation and maintenance tool after the user confirms the approval. The security execution unit is used to constrain the task execution module to perform operation and maintenance tasks under preset security conditions when the permission verification is passed, and to trigger interrupt control when timeout, unauthorized access or abnormal status is detected; The audit record unit is used to record the risk assessment results, confirmation information, authorization verification results, and execution process.
[0015] To achieve the above objectives, another aspect of this application proposes a method for implementing the system described above, the method comprising the following steps: Obtain operation and maintenance requests input by users in natural language format; Based on a preset task classification prompt dictionary, semantic analysis is performed on the operation and maintenance request to determine the operation and maintenance task category corresponding to the operation and maintenance request or to determine it as an unknown task. After the operation and maintenance task category is determined, operation and maintenance operation preview information corresponding to the operation and maintenance task category is generated, and user confirmation instructions are received to generate task execution license information; Based on the task execution license information, the pre-registered operation and maintenance tools are invoked to execute the corresponding operation and maintenance tasks according to the operation and maintenance task category, and task execution results are generated. The task execution result will be output to the user. When the operation and maintenance request is determined to be an unknown task or meets the preset update conditions, dynamic update processing is performed on the task classification prompt dictionary and operation and maintenance tools.
[0016] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the system described above.
[0017] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the system described above.
[0018] To achieve the above objectives, another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the system described above.
[0019] The embodiments of this application include at least the following beneficial effects: This application provides a scalable operation and maintenance intelligent agent system, method, and related equipment. This solution receives operation and maintenance requests in natural language through a user interaction module and accurately parses the semantics of the requests in conjunction with a task classification and recognition module, achieving automatic matching of operation and maintenance needs with task categories. This reduces the reliance of traditional operation and maintenance on human experience and professional instructions, improving the ease of use and response efficiency of operation and maintenance operations. By introducing a user confirmation module before task execution, a visual preview of the operation and maintenance operation is generated, and task execution permission is formed based on user confirmation. This effectively avoids accidental operation and direct execution of high-risk instructions, significantly improving the security and controllability of the operation and maintenance process. At the same time, after obtaining the execution permission, the task execution module calls pre-registered operation and maintenance tools according to the task category to complete automated processing, realizing standardized and process-oriented execution of operation and maintenance tasks, reducing manual intervention, and improving execution consistency and stability. Furthermore, the system's self-updating module can dynamically update the task classification prompt dictionary and operation and maintenance tools according to the actual operation and maintenance scenarios, enabling the system to have continuous learning and evolution capabilities, enhancing its adaptability to new operation and maintenance needs and complex scenarios, thereby improving the overall scalability, intelligence level and long-term operating efficiency of the operation and maintenance system. Attached Figure Description
[0020] Figure 1 This is a schematic diagram of a scalable operation and maintenance intelligent agent system provided in an embodiment of this application; Figure 2 This is a schematic diagram of the task classification and recognition module provided in an embodiment of this application; Figure 3 This is a schematic diagram of the security enhancement design module provided in an embodiment of this application; Figure 4 This is a flowchart illustrating an implementation method for a scalable operation and maintenance intelligent agent system provided in an embodiment of this application. Detailed Implementation
[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.
[0022] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”
[0023] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.
[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.
[0025] This application provides a scalable operation and maintenance intelligent agent system, method, and related equipment. This solution receives operation and maintenance requests in natural language through a user interaction module and accurately parses the request semantics using a task classification and recognition module. This enables automatic matching of operation and maintenance needs with task categories, reducing the reliance on human experience and professional instructions in traditional operation and maintenance, and improving the usability and response efficiency of operation and maintenance. By introducing a user confirmation module before task execution, a visual preview of the operation and a task execution permit are generated based on user confirmation, effectively avoiding accidental operations and the direct execution of high-risk instructions, significantly improving the security and controllability of the operation and maintenance process. Simultaneously, after obtaining the execution permit, the task execution module calls pre-registered operation and maintenance tools according to the task category to complete automated processing, achieving standardized and streamlined execution of operation and maintenance tasks, reducing manual intervention, and improving execution consistency and stability. Furthermore, the system self-updating module can dynamically update the task classification prompt dictionary and operation and maintenance tools according to the actual operation and maintenance scenario, enabling the system to continuously learn and evolve, enhancing its adaptability to new operation and maintenance needs and complex scenarios, thereby improving the overall scalability, intelligence level, and long-term operating efficiency of the operation and maintenance system.
[0026] This application provides a scalable operation and maintenance intelligent agent system, relating to the field of automated operation and maintenance technology. This scalable operation and maintenance intelligent agent system can be applied to terminals, servers, or software running on either terminal or server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing a scalable operation and maintenance intelligent agent system, but is not limited to the above forms.
[0027] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.
[0028] Figure 1 This is a schematic diagram of an optional module of an scalable operation and maintenance intelligent agent system provided in an embodiment of this application. Figure 1 The system may include, but is not limited to: The user interaction module is used to receive operation and maintenance requests input by users in natural language form, and output the task identification results, execution preview information and confirmation information corresponding to the operation and maintenance request. In this embodiment, the user interaction module serves as the unified input and output interface of the operation and maintenance intelligent agent system, responsible for collecting operation and maintenance requests, encapsulating context, and providing result feedback. When operation and maintenance personnel input operation and maintenance requests in natural language through the user interaction module, the module first performs basic parsing processing on the input content, including character normalization, sentence integrity detection, and request session identifier generation. It then associates the operation and maintenance request with user identity information, operation source identifier, and current session state to form a standardized operation and maintenance request data object, providing a unified data input format for subsequent semantic analysis.
[0029] After the maintenance request data object is generated, the user interaction module sends it to the task classification and recognition module for semantic recognition, while simultaneously maintaining the request's contextual state information locally. Upon receiving the task recognition result from the task classification and recognition module, the user interaction module parses the result and transforms the task category, confidence level, and corresponding intended operation into readable task description information. This description information is used to demonstrate the system's understanding of the maintenance request to the user, allowing them to intuitively confirm whether the system has accurately recognized their maintenance intent.
[0030] Furthermore, upon entering the user confirmation phase, the user interaction module dynamically generates execution preview information based on the task identification results and presents this preview information along with the confirmation operation entry to the user. Before the user explicitly issues a confirmation command, the user interaction module maintains the current session in a pending confirmation state and does not transmit any executable commands to the task execution module. Upon receiving a user confirmation command or cancellation command, the user interaction module generates corresponding task execution permission information or task termination information through the user confirmation module, and synchronously updates the session state with this result. This ensures that the maintenance operation proceeds into the subsequent execution process or safely terminates under human control.
[0031] The task classification and recognition module is used to perform semantic analysis on maintenance requests based on a preset task classification prompt dictionary to determine the task category corresponding to the maintenance request or to determine it as an unknown task. Among them, reference Figure 2 As shown, the task classification and recognition module includes: The semantic parsing unit is used to perform semantic parsing on operation and maintenance requests and generate vectorized representations. The similarity calculation unit is used to calculate the similarity between the vectorized representation and the task vector in the task classification prompt vocabulary using a lightweight fine-tuning model; The classification and determination unit is used to determine whether the similarity score results meet the preset threshold conditions in order to determine the corresponding operation and maintenance task category.
[0032] Specifically, semantic parsing is performed on operation and maintenance requests to generate vectorized representations, including: The original natural language text of the operation and maintenance request is segmented and semantically parsed to extract semantic feature information that represents the operation and maintenance intent; Based on the pre-defined big oracle model, semantic feature information is encoded and processed to map operation and maintenance requests to a unified semantic space, generating corresponding request vectors. The request vector is subjected to dimensional normalization and semantic alignment to generate a vectorized representation.
[0033] In this embodiment, the semantic parsing unit performs structured semantic processing on maintenance requests from the user interaction module to eliminate the impact of differences in natural language expression on the accuracy of task classification. Upon receiving the raw natural language text of the maintenance request, the semantic parsing unit first performs preprocessing operations on the text, including character normalization, noise word filtering, and sentence integrity verification. Then, based on preset maintenance semantic rules, it performs word segmentation and semantic segmentation of the text, thereby identifying the key semantic units constituting the maintenance intent, transforming the raw natural language request from a "free text form" into a set of parsable semantic elements.
[0034] Furthermore, the semantic parsing unit performs intent-structured processing on the set of semantic elements, mapping them into multiple semantic subsets such as action semantics, target object semantics, operation scope semantics, and environment and constraint semantics, according to the semantic dimensions required for understanding the operation and maintenance tasks. For example, when a request contains words such as "check" or "restart," it is classified as action semantics; when a request contains object descriptions such as "service," "node," or "port," it is classified as target object semantics; and when a request contains limiting conditions such as "failed node" or "specified port," it is labeled as scope constraint semantics. Through the above processing, the inherent semantic structure of the operation and maintenance request is explicitly expressed, providing clear input boundaries for subsequent unified semantic encoding.
[0035] After semantic structuring, the semantic parsing unit reorganizes the semantic elements according to a preset encoding format and inputs them into a preset large language model encoder for semantic encoding processing. This maps the maintenance requests to a unified semantic representation space, generating corresponding request vectors. To ensure the comparability of different requests in the semantic space, the semantic parsing unit performs necessary vector normalization and semantic alignment processing on the generated request vectors. This ensures that the request vectors meet the unified requirements of the system in terms of dimension, scale, and semantic distribution, thus serving as a stable and reliable vectorized representation to be passed to the similarity calculation unit.
[0036] Specifically, a lightweight fine-tuning model is used to calculate the similarity between the vectorized representation and the task vectors in the task classification prompt vocabulary, including: Lightweight fine-tuning models are built based on pre-trained large language models; Retrieve the structured prompt text corresponding to each task category in the task category prompt word library; The structured prompt word text is semantically encoded based on a pre-defined large language model encoder to generate the corresponding task vector; The vectorized representation and the task vector are aligned in the vector space to achieve a unified semantic space mapping. Within a unified semantic space, based on a lightweight fine-tuning model, similarity measurement operations are performed on the vectorized representations and each task vector to generate corresponding similarity scores.
[0037] In this embodiment, the similarity calculation unit constructs a lightweight fine-tuning model based on a pre-trained large language model to enhance its adaptability to the expression methods of the operations and maintenance domain. Specifically, the lightweight fine-tuning model can serve as an additional parameter layer or adapter layer on the encoder side, making it more sensitive to the combined semantics of operations and maintenance action words such as "check / inspection / diagnosis / restart / expansion / switch" and object words such as "service / process / port / node / cluster / environment" while maintaining the general semantic understanding capability of the large model. This lightweight fine-tuning model does not change the structured expression of the prompt lexicon, but rather improves the separability of request vectors and task vectors in the same semantic coordinate system, making different task categories form clearer boundaries in the vector space.
[0038] Furthermore, the similarity calculation unit semantically encodes the structured prompt text in the task classification prompt word library to generate task vectors corresponding to each task category. The structured prompt text not only includes the task category name and task description, but also normative fields such as "prohibited operation constraints," "list of available tools," and "output format requirements," thus encoding both the semantic boundaries and security constraints of the task into the task vector. During encoding, the similarity calculation unit calls a pre-set large language model encoder to perform vectorization processing on each structured prompt text and generate a unique vector identifier for each task category. During system operation, task vectors can be cached and a version number index can be established to support incremental reconstruction after the prompt word library is updated, avoiding the additional overhead of re-encoding the entire text for each request.
[0039] After constructing the request vector and task vector, the similarity calculation unit performs vector space alignment to ensure that the two types of vectors can be directly compared in the same semantic space. Specifically, the alignment process includes applying unified standardization rules, dimensional consistency checks, and semantic space mapping consistency checks to the request vector and task vector. When a change in the task vector version is detected due to an update to the prompt vocabulary, the similarity calculation unit can choose to project the request vector to the corresponding version's semantic space or adopt a compatible mapping strategy to ensure the comparability of similarity scores between different versions. This alignment mechanism reduces score fluctuations caused by encoder configuration differences, template version differences, or semantic drift, thereby stabilizing the classification results.
[0040] During the similarity measurement phase, the similarity calculation unit performs similarity calculations on the vectorized representations and task vectors one by one within a unified semantic space, obtaining a set of similarity scores. These scores are then bound to the corresponding task category identifiers and output. To enhance the usability of the scores, the similarity calculation unit can also output Top-K candidate task categories and their confidence intervals, and pass risk levels and tool whitelist information as accompanying fields to the classification decision unit or user confirmation module for subsequent threshold determination and confirmation strategy selection. If multiple task category scores are found to be close and difficult to distinguish, the similarity calculation unit can mark this as a "candidate conflict" and output a candidate set, providing a basis for the subsequent decision unit to trigger the "fuzzy matching / unknown task" strategy, thereby improving the robustness and scalability of the overall classification process.
[0041] Specifically, it determines whether the similarity score meets a preset threshold condition to identify the corresponding operation and maintenance task category, including: Obtain similarity score results; The similarity scores are sorted to determine the candidate task category corresponding to the highest similarity score; Determine whether the highest similarity score is greater than or equal to a preset threshold; When the highest similarity score meets the preset threshold condition, the candidate task category is determined as the target operation and maintenance task category corresponding to the operation and maintenance request, and the corresponding task identification result is output. When the highest similarity score does not meet the preset threshold, the maintenance request is marked as an unknown task, and abnormal identification information is generated to prompt the user or trigger the system self-update module.
[0042] In this embodiment, the classification and determination unit is used to centrally analyze and make decisions on multiple sets of similarity scores output by the similarity calculation unit, so as to achieve the final task classification of the operation and maintenance request. After receiving the set of similarity scores, the classification and determination unit first performs a completeness check on each score to ensure that each similarity score corresponds one-to-one with the unique task category in the task classification prompt word library, thereby avoiding erroneous judgments caused by missing data or abnormal mapping.
[0043] After verifying the similarity scores, the classification unit sorts all similarity scores to identify the candidate task category that is semantically closest to the current maintenance request. During the sorting process, the classification unit not only focuses on the magnitude of the scores but also evaluates the stability of the sorting results based on the score distribution to prevent misjudgments when multiple task categories are semantically similar, thereby improving the reliability of the task classification results.
[0044] Furthermore, the classification unit compares the highest similarity score obtained after sorting with the system's preset classification threshold. The classification threshold is set based on the system's historical recognition performance and task security requirements, and is used to distinguish between maintenance tasks that can be clearly identified by the system and semantically uncertain requests. When the highest similarity score reaches or exceeds the preset threshold, the classification unit determines that the current maintenance request has clear semantic attribution conditions within the existing task classification prompt word library.
[0045] After confirming that the highest similarity score meets the preset threshold, the classification unit determines the corresponding candidate task category as the target operation and maintenance task category and generates a standardized task identification result. The task identification result includes at least the target operation and maintenance task category identifier, similarity confidence information, and a risk level identifier associated with the category. This is used to support the subsequent user confirmation module in dynamically selecting a confirmation strategy based on the risk level and to provide a clear task context for the task execution module.
[0046] When the highest similarity score does not reach the preset threshold, the classification and determination unit classifies the current maintenance request as an unknown task and generates corresponding anomaly identification information. This anomaly identification information is used to inform the user that the current request does not currently have an executable task template. It also serves as input data for the system's self-updating module, used for subsequent analysis to determine whether to add a new task classification template or adjust the prompt text, thus forming a closed-loop technology for the continuous evolution of task recognition capabilities.
[0047] The user confirmation module is used to generate corresponding operation and maintenance operation preview information after the task category is determined, and to generate task execution permission information after receiving the user confirmation instruction. In this embodiment, the user confirmation module is positioned between the task classification and identification module and the task execution module. It is used to establish a manually controllable security confirmation process before the actual execution of an operation and maintenance task. After the task classification and identification module completes the semantic analysis of the operation and maintenance request and outputs the task category and corresponding risk level, the user confirmation module receives the task identification result and, based on the pre-registered task description information associated with the task category, structures the proposed operation and maintenance operation to generate a corresponding task confirmation object, which serves as the basic data unit for subsequent confirmation processes.
[0048] During the process of generating the task confirmation object, the user confirmation module summarizes the key elements of the task to be executed, including the target object of the task, the type of operation and maintenance tools to be invoked, and the scope of the execution impact. It then generates operation and maintenance operation preview information without triggering actual execution. This preview information reflects the system's understanding of the current operation and maintenance request, allowing users to know the intended operation and maintenance actions before task execution, thereby avoiding the risk of misoperation due to semantic misjudgment or misunderstanding.
[0049] Furthermore, the user confirmation module dynamically matches the corresponding confirmation strategy based on the risk level identified in the task confirmation object and enters a pending confirmation state. In this state, the system sets a confirmation validity period for the maintenance task and continuously monitors whether the user issues a confirmation or cancellation command within the validity period. Before the confirmation command is received, the user confirmation module uses a state control mechanism to block the task from being passed downstream to the task execution module, ensuring that no maintenance operation can be triggered without human authorization.
[0050] When the user confirmation module receives a user confirmation instruction within the confirmation validity period, the module converts the task confirmation object into task execution permission information and sends the permission information along with the task context state to the task execution module. If a cancellation instruction is received within the confirmation validity period or no confirmation instruction is received after the timeout, the user confirmation module generates a task termination flag and marks the current operation and maintenance request as not executed, thereby completing the security interception and closed-loop control of the operation and maintenance task.
[0051] The task execution module is used to call pre-registered operation and maintenance tools to execute corresponding operation and maintenance tasks based on task execution license information and task category, and output the task execution results; Specifically, based on task execution license information, the pre-registered operation and maintenance tools are invoked to execute the corresponding operation and maintenance tasks according to the task category, and the task execution results are output, including: Based on task execution license information and task category, the target operation and maintenance tool is determined by matching from the pre-registered operation and maintenance tool set; Before invoking the target operation and maintenance tool, perform a tool validity and parameter compliance check on the target operation and maintenance tool; If the verification passes, the target operation and maintenance tool is invoked to perform the corresponding operation and maintenance operation based on preset security constraints; the security constraints include the number of execution steps limit, the execution time limit, and tool permission verification. During the operation and maintenance process, the execution status is monitored, and the execution is terminated when the preset termination conditions are met. After the operation and maintenance is completed, the task execution result returned by the target operation and maintenance tool is obtained and output.
[0052] In this embodiment, the task execution module is used to perform controlled execution of identified maintenance tasks after the user completes confirmation and authorization. After receiving the task execution permission information generated by the user confirmation module, the module first verifies the validity of the permission information to confirm that it is consistent with the current maintenance request context and is within a preset valid time range, thereby ensuring that subsequent maintenance operations are triggered only under conditions of explicit user authorization.
[0053] After the license information is verified, the task execution module matches and determines the target operation and maintenance tool from the pre-registered set of operation and maintenance tools based on the task category. This matching process uses the task category identifier as the main index and combines it with the execution environment scope and permission level information contained in the task execution license information to filter candidate operation and maintenance tools, thereby determining the target operation and maintenance tool that matches the current operation and maintenance task and has the conditions for execution, avoiding tool misselection or unauthorized calls.
[0054] After identifying the target operation and maintenance tool, the task execution module performs tool validity and parameter compliance checks before invoking the tool. Tool validity checks confirm that the target tool belongs to the system's pre-registered and not disabled tool set; parameter compliance checks verify whether the invocation parameters meet the target tool's preset parameter specifications and security requirements. Only if all checks pass will the task execution module allow the actual execution phase to begin.
[0055] During the execution phase, the task execution module invokes the target maintenance tool to perform corresponding maintenance operations based on preset security constraints. These security constraints include limits on the number of execution steps, execution time limits, and tool permission verification, and remain in effect throughout the entire execution process. The task execution module monitors the execution status in real time. When preset termination conditions such as execution timeout, exceeding step limits, or the tool returning an abnormal status are detected, the current maintenance operation is immediately terminated, and a safe exit process is initiated.
[0056] After the operation and maintenance is completed, the task execution module obtains the task execution result returned by the target operation and maintenance tool, and encapsulates the result into a structured task execution result information, which is then sent to the user interaction module for display. At the same time, it records key information during the execution process to complete an operation and maintenance task execution process with authorization control, security constraints and result traceability.
[0057] The system self-update module is used to dynamically update the task classification prompt dictionary and operation and maintenance tools based on the task execution results.
[0058] In this embodiment, the system self-update module is used to dynamically update the task classification prompt dictionary and the set of operation and maintenance tools without affecting the continuous operation of the system. The system self-update module is triggered during the system startup phase or according to a preset update cycle, and actively establishes a secure communication connection with a pre-configured update source to obtain remotely released system update metadata. The update metadata includes at least prompt dictionary version information, toolset version information, and corresponding compatibility identifiers, used to determine whether there are available updates for the current system.
[0059] After obtaining the update metadata, the system self-update module compares the locally stored prompt word library version number and toolset version number with the remote version information. When a remote version is detected to be higher than the local version and the compatibility check passes, the system self-update module generates update candidate objects and sends an update summary to the user interaction module for prompting. Before the user confirms the update, the system self-update module maintains its current operating state and does not affect the current system configuration, thus ensuring that the system update behavior is within the scope of human control.
[0060] Upon receiving the user's confirmation of the update command, the system self-update module initiates the update execution process, downloads the remote update package, and sequentially performs integrity checks, source trustworthiness checks, and version consistency checks. If the checks pass, the system self-update module loads the task classification prompt dictionary and tool description information from the update package into the system runtime environment and registers the new version resources as available through hot reloading. This allows the subsequent task classification and recognition module and task execution module to use the updated capability set without restarting the system.
[0061] To ensure system stability, the system self-update module synchronously retains the version snapshot information before the update during the update process and continuously monitors task categories and task execution status during operation after the update takes effect. When abnormal behavior or compatibility issues caused by the update are detected, the system self-update module automatically triggers the rollback mechanism, restoring the system prompt dictionary and toolset to the stable version before the update, and recording the corresponding update failure information to avoid adverse effects on the overall system availability.
[0062] The system also includes a security enhancement design module for implementing tiered security controls during the execution of operation and maintenance tasks.
[0063] Among them, reference Figure 3 As shown, the security enhancement design module includes: The risk assessment unit is used to assess the risk level of an operation and maintenance task based on the risk level information associated with the task in the task classification prompt dictionary. The confirmation control unit is used to generate corresponding user confirmation policies and perform user confirmation based on the risk level; the user confirmation policy includes one of single confirmation, double confirmation, or approval chain confirmation. The permission verification unit is used to verify the operation permissions of the executing entity and the calling permissions of the target operation and maintenance tool after the user confirms the permission. The security execution unit is used to constrain the task execution module to perform operation and maintenance tasks under preset security conditions when the permission verification is passed, and to trigger interrupt control when timeout, unauthorized access or abnormal status is detected. The audit record unit is used to record the risk assessment results, confirmation information, authorization verification results, and execution process.
[0064] In this embodiment, the security enhancement design module implements tiered security control during the execution of operation and maintenance tasks. It determines the risk level of each task and employs different user confirmation methods, permission verification requirements, and timeout control mechanisms based on different risk levels, thereby ensuring that operation and maintenance tasks are executed under controllable risks. The security enhancement design module intervenes in the execution process after task classification and initializes the security control process before the task actually begins execution.
[0065] Specifically, the security enhancement design module first classifies operational tasks into risk levels—low, medium, and high—using a risk assessment unit. The confirmation control unit then generates user confirmation policies corresponding to each risk level. These policies are as follows: for low-risk tasks, the system requires the user to click to confirm, performing only basic permission checks on the executing entity, with a five-minute confirmation timeout. For medium-risk tasks, the system requires a two-factor authentication process, including a confirmation button and CAPTCHA verification, while also performing intermediate-level permission checks on the executing entity, with a three-minute confirmation timeout. For high-risk tasks, the system requires user confirmation through an administrator approval chain, requiring the executing entity to have SA (Administrator) privileges and verifying the validity of approval credentials during the permission verification phase, with a one-minute confirmation timeout.
[0066] After risk level classification, the security enhancement design module enters a dynamic confirmation process. When a user enters a deployment command, the system first performs risk level classification and then proceeds to the corresponding confirmation branch based on the judgment result: For low-risk tasks, the system requires the user to complete fingerprint verification or click to confirm; for medium-risk tasks, the system sends an SMS verification code to the user, requiring secondary confirmation; for high-risk tasks, the system submits the task to the administrator approval system and completes authorization confirmation through SMS authorization code or SAML-based identity federation authentication. Only if the confirmation process is completed within the corresponding timeout period is the task allowed to proceed to the subsequent execution phase.
[0067] After user confirmation, the security enhancement design module verifies the permissions of the executing entity and the target maintenance tool through the permission verification unit. Low-risk tasks only verify whether the executing entity has basic operation permissions; medium-risk tasks verify whether the executing entity has intermediate operation permissions; high-risk tasks, in addition to verifying that the executing entity has SA permissions, further verify the validity of the approval credentials and authorized identity. If any permission verification fails, the system immediately terminates the task execution process.
[0068] During task execution, the security enhancement design module continuously monitors the execution status through the permission verification unit. When confirmation timeout, permission overrun, or execution abnormality is detected, the execution of the operation and maintenance task is immediately interrupted, and the corresponding interruption reason, risk level, confirmation method, and permission verification result are recorded through the permission verification unit to achieve secure and controllable execution of the operation and maintenance task and complete audit record.
[0069] Please see Figure 4 This application also provides an implementation method for the above-described system, the method comprising the following steps: S1: Obtain the operation and maintenance request input by the user in natural language form; S2: Based on a preset task classification prompt dictionary, perform semantic analysis on the operation and maintenance request to determine the operation and maintenance task category corresponding to the operation and maintenance request or to determine it as an unknown task. S3: After the operation and maintenance task category is determined, generate operation and maintenance operation preview information corresponding to the operation and maintenance task category, and receive user confirmation instructions to generate task execution license information; S4: Based on the task execution license information, the pre-registered operation and maintenance tools are invoked to execute the corresponding operation and maintenance tasks according to the operation and maintenance task category, and the task execution results are generated; S5: Output the task execution results to the user; S6: When the operation and maintenance request is determined to be an unknown task or meets the preset update conditions, perform dynamic update processing on the task classification prompt dictionary and operation and maintenance tools.
[0070] It is understood that the content of the above system embodiments is applicable to this method embodiment. The specific functions implemented in this method embodiment are the same as those in the above system embodiments, and the beneficial effects achieved are also the same as those achieved in the above system embodiments.
[0071] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the system described above. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.
[0072] It is understood that the content of the above system embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above system embodiments, and the beneficial effects achieved are also the same as those achieved by the above system embodiments.
[0073] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the system described above.
[0074] It is understood that the content of the above system embodiments is applicable to this storage medium embodiment. The specific functions implemented by this storage medium embodiment are the same as those of the above system embodiments, and the beneficial effects achieved are also the same as those achieved by the above system embodiments.
[0075] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the system described above.
[0076] It is understood that the content of the above system embodiments is applicable to the present program product embodiments. The specific functions implemented by the present program product embodiments are the same as those of the above system embodiments, and the beneficial effects achieved are also the same as those achieved by the above system embodiments.
[0077] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
[0078] This application provides a scalable operation and maintenance intelligent agent system, method, and related equipment. This solution receives operation and maintenance requests in natural language through a user interaction module and accurately parses the request semantics using a task classification and recognition module. This enables automatic matching of operation and maintenance needs with task categories, reducing the reliance on human experience and professional instructions in traditional operation and maintenance, and improving the usability and response efficiency of operation and maintenance. By introducing a user confirmation module before task execution, a visual preview of the operation and a task execution permit are generated based on user confirmation, effectively avoiding accidental operations and the direct execution of high-risk instructions, significantly improving the security and controllability of the operation and maintenance process. Simultaneously, after obtaining the execution permit, the task execution module calls pre-registered operation and maintenance tools according to the task category to complete automated processing, achieving standardized and streamlined execution of operation and maintenance tasks, reducing manual intervention, and improving execution consistency and stability. Furthermore, the system self-updating module can dynamically update the task classification prompt dictionary and operation and maintenance tools according to the actual operation and maintenance scenario, enabling the system to continuously learn and evolve, enhancing its adaptability to new operation and maintenance needs and complex scenarios, thereby improving the overall scalability, intelligence level, and long-term operating efficiency of the operation and maintenance system.
[0079] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.
[0080] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.
[0081] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.
[0082] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.
[0083] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0084] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.
Claims
1. A scalable operation and maintenance intelligent agent system, characterized in that, The system includes: The user interaction module is used to receive maintenance requests input by users in natural language form, and output the task identification result, execution preview information and confirmation information corresponding to the maintenance request. The task classification and recognition module is used to perform semantic analysis on the operation and maintenance request based on a preset task classification prompt word library, and determine the task category corresponding to the operation and maintenance request or determine it as an unknown task. The user confirmation module is used to generate corresponding operation and maintenance operation preview information after the task category is determined, and to generate task execution permission information after receiving the user confirmation instruction. The task execution module is used to invoke pre-registered operation and maintenance tools to execute corresponding operation and maintenance tasks based on the task execution license information and the task category, and output the task execution results; The system self-update module is used to dynamically update the task classification prompt dictionary and the operation and maintenance tools based on the task execution results.
2. The system according to claim 1, characterized in that, The task classification and recognition module includes: A semantic parsing unit is used to perform semantic parsing on the operation and maintenance request and generate a vectorized representation; The similarity calculation unit is used to calculate the similarity between the vectorized representation and the task vector in the task classification prompt word library using a lightweight fine-tuning model. The classification and determination unit is used to determine whether the similarity score results meet the preset threshold conditions in order to determine the corresponding operation and maintenance task category.
3. The system according to claim 2, characterized in that, The step of semantically parsing the maintenance request and generating a vectorized representation includes: The original natural language text of the maintenance request is segmented and semantically parsed to extract semantic feature information that represents the maintenance intent; Based on the preset big oracle model, the semantic feature information is encoded and the operation and maintenance request is mapped to a unified semantic space to generate a corresponding request vector. The request vector is subjected to dimensional normalization and semantic alignment to generate a vectorized representation.
4. The system according to claim 2, characterized in that, The step of employing a lightweight fine-tuning model to calculate the similarity between the vectorized representation and the task vectors in the task classification prompt vocabulary includes: Lightweight fine-tuning models are built based on pre-trained large language models; Obtain the structured prompt text corresponding to each task category in the task classification prompt word library; The structured prompt word text is semantically encoded based on a pre-set large language model encoder to generate a corresponding task vector; The vectorized representation and the task vector are aligned in the vector space to perform a unified semantic space mapping. Within the unified semantic space, based on the lightweight fine-tuning model, similarity measurement operations are performed on the vectorized representation and each of the task vectors to generate corresponding similarity scores.
5. The system according to claim 2, characterized in that, The determination of whether the similarity score meets the preset threshold condition to determine the corresponding operation and maintenance task category includes: Obtain the similarity score result; The similarity scores are sorted to determine the candidate task category corresponding to the highest similarity score; Determine whether the highest similarity score is greater than or equal to the preset threshold; When the highest similarity score is determined to meet the preset threshold condition, the candidate task category is determined as the target operation and maintenance task category corresponding to the operation and maintenance request, and the corresponding task identification result is output. When the highest similarity score is determined not to meet the preset threshold condition, the maintenance request is marked as an unknown task, and abnormal identification information is generated to prompt the user or trigger the system self-update module.
6. The system according to claim 1, characterized in that, The step of invoking pre-registered operation and maintenance tools to execute corresponding operation and maintenance tasks based on the task execution license information and the task category, and outputting the task execution results, includes: Based on the task execution license information and the task category, the target operation and maintenance tool is determined by matching from the pre-registered operation and maintenance tool set; Before invoking the target operation and maintenance tool, perform a tool validity and parameter compliance check on the target operation and maintenance tool; If the verification passes, the target operation and maintenance tool is invoked to perform the corresponding operation and maintenance operation based on preset security constraints; the security constraints include execution step limit, execution time limit and tool permission verification. During the operation and maintenance process, the execution status is monitored, and the execution is terminated when the preset termination conditions are met. After the operation and maintenance is completed, the task execution result returned by the target operation and maintenance tool is obtained and output.
7. The system according to claim 1, characterized in that, It also includes a security enhancement design module for implementing tiered security controls during the execution of the operation and maintenance tasks.
8. The system according to claim 7, characterized in that, The security enhancement design module includes: The risk assessment unit is used to assess the risk level of the operation and maintenance task based on the risk level information associated with the operation and maintenance task in the task classification prompt word library. The confirmation control unit is used to generate a corresponding user confirmation strategy and perform user confirmation based on the risk level; the user confirmation strategy includes one of single confirmation, double confirmation, or approval chain confirmation. The permission verification unit is used to verify the operation permissions of the executing entity and the calling permissions of the target operation and maintenance tool after the user confirms the approval. The security execution unit is used to constrain the task execution module to perform operation and maintenance tasks under preset security conditions when the permission verification is passed, and to trigger interrupt control when timeout, unauthorized access or abnormal status is detected; The audit record unit is used to record the audit results, confirmation information, authorization verification results, and execution process.
9. A method for implementing the system according to any one of claims 1-8, characterized in that, The method includes the following steps: Obtain operation and maintenance requests input by users in natural language format; Based on a preset task classification prompt dictionary, semantic analysis is performed on the operation and maintenance request to determine the operation and maintenance task category corresponding to the operation and maintenance request or to determine it as an unknown task. After the operation and maintenance task category is determined, operation and maintenance operation preview information corresponding to the operation and maintenance task category is generated, and user confirmation instructions are received to generate task execution license information; Based on the task execution license information, the pre-registered operation and maintenance tools are invoked to execute the corresponding operation and maintenance tasks according to the operation and maintenance task category, and task execution results are generated. The task execution result will be output to the user. When the operation and maintenance request is determined to be an unknown task or meets the preset update conditions, dynamic update processing is performed on the task classification prompt dictionary and operation and maintenance tools.
10. An electronic device, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the system as described in any one of claims 1 to 8.