Operation and maintenance management method and device based on large model, equipment and storage medium

Abstract

This invention provides a large-scale model-based operation and maintenance management method, device, equipment, and storage medium. The device receives various operation and maintenance task requests uniformly at the application service layer, and the AI ​​agent layer performs intelligent task planning and instruction generation, achieving centralized and intelligent task scheduling. The data pipeline layer responds to instructions, simultaneously executing multi-protocol real-time data acquisition and vectorized knowledge base-based relational knowledge retrieval, effectively breaking down data silos and providing complete information input that integrates real-time status and historical experience for decision-making. This device fundamentally changes the traditional operation and maintenance model, freeing manual labor from tedious and repetitive inspection, troubleshooting, and reporting work, significantly improving fault response speed, processing accuracy, and knowledge reuse efficiency, and realizing automated, intelligent, and knowledge-based closed-loop management of operation and maintenance work.

Description

Technical Field

[0001] This invention relates to the field of operation and maintenance system technology, and in particular to an operation and maintenance management method, device, equipment and storage medium based on a large model. Background Technology

[0002] As the complexity of enterprise information systems continues to rise, traditional IT operations and maintenance (O&M) models face severe challenges. Existing technical solutions primarily rely on manual operation and experience-based judgment by O&M personnel, and their limitations are becoming increasingly apparent. First, in the fault handling phase, from alarm occurrence to root cause location, manual login to multiple independent systems for cross-referencing logs and metrics is often required, resulting in long response chains, low processing efficiency, and high dependence on personnel experience. Second, O&M data exists in a "siloed" state; real-time monitoring data, historical fault records, emergency plans, system architecture diagrams, and other knowledge are scattered across different platforms and documents, with heterogeneous formats and a lack of effective unified governance and correlation analysis methods, leading to fragmented knowledge assets that cannot be quickly retrieved and applied in emergency scenarios. Furthermore, repetitive tasks such as daily inspections, capacity assessments, and report preparation consume significant manpower and are prone to human error. Although some automated scripting tools exist, their flexibility is insufficient to handle complex and unexpected O&M scenarios, and they lack the ability to perform intelligent reasoning and decision-making from massive amounts of data and knowledge.

[0003] In summary, the problems existing in the current technology urgently need to be solved. Summary of the Invention

[0004] This invention provides a large-scale model-based operation and maintenance management method, device, equipment, and storage medium to address the shortcomings of existing technologies and achieve automated, intelligent, and knowledge-based closed-loop management of operation and maintenance work.

[0005] This invention provides an operation and maintenance management system based on a large model, comprising: The application service layer is used to receive operation and maintenance task requests from clients; The AI ​​agent layer, connected to the application service layer, is used to plan task steps and generate task execution instructions according to the type of the operation and maintenance task request. The data pipeline layer is used to respond to the task execution command to obtain real-time operation and maintenance data streams related to the task and related knowledge related to the task; and input the real-time operation and maintenance data streams and related knowledge into the large model application layer; The large model application layer is used to generate decision results based on the real-time operation and maintenance data stream and the associated knowledge, and return them to the AI ​​agent layer. The AI ​​agent layer is also used to invoke the corresponding functional agent to generate executable instructions or structured reports based on the decision results, and deliver them through the application service layer.

[0006] According to the large-model-based operation and maintenance management system provided by the present invention, the AI ​​agent layer includes: A task planning engine is used to parse the type of the operation and maintenance task request and generate a sequence of task steps containing at least one atomic operation based on predefined task decomposition rules. The instruction generation module is used to construct task execution instructions to be sent to the data pipeline layer based on the task step sequence. The task execution instructions include data acquisition parameters and knowledge retrieval parameters corresponding to each atomic operation.

[0007] According to the large-model-based operation and maintenance management system provided by the present invention, the data pipeline layer includes: The Multiprotocol Communication (MCP) module is used to respond to the task execution command by acquiring the real-time operation and maintenance data stream from at least one external monitoring system via the MCP communication protocol. The Knowledge Retrieval Enhancement (RAG) module is used to retrieve the task-related associated knowledge from a vectorized knowledge base in response to the task execution instruction.

[0008] According to the present invention, an operation and maintenance management system based on a large model is provided, wherein the large model application layer includes: The receiving module is used to receive the real-time operation and maintenance data stream and the associated knowledge input from the data pipeline layer, and to fuse the real-time operation and maintenance data stream and the associated knowledge. The collaborative reasoning engine is used to comprehensively analyze the fused real-time operation and maintenance data stream and the associated knowledge, and generate decision results including problem diagnosis, root cause inference and handling suggestions based on natural language understanding and preset operation and maintenance logic rules. The output module is used to return the decision results in a structured manner to the AI ​​agent layer, so that it can call the corresponding functional agent to execute.

[0009] According to the large-model-based operation and maintenance management system provided by the present invention, the AI ​​agent layer further includes: The agent scheduling module is used to match and call the corresponding functional agents from the pre-configured agent library according to the task type and execution strategy indicated in the decision result; The parameter passing module is used to pass the description of the problem to be processed, the root cause analysis conclusions and the suggested operations included in the decision results as input parameters to the called functional agent. The result generation module is used to drive the functional agent to perform at least one of the following operations based on the input parameters: Generate executable instructions containing specific operation commands or scripts; Generate a structured report that includes statistical charts, root cause analysis summaries, and treatment steps; The AI ​​agent layer delivers the executable instructions or structured reports to the corresponding client through the interface provided by the application service layer.

[0010] According to the large-model-based operation and maintenance management system provided by the present invention, the structured report generated by the result generation module driving the functional agent includes at least one of the following forms: Statistical analysis charts based on fault data; Trend forecasting charts based on resource usage data; A failure review report that includes the root cause chain and the timeline of the handling process.

[0011] According to the large-model-based operation and maintenance management system provided by the present invention, the system further includes: a driving layer, which includes the following modules: The constraint module is used to define the mandatory and prohibited items, legal value range and reporting format of operation and maintenance tasks, and to establish a verification mechanism to verify data items, value range and data time, as well as set red line constraints to prohibit unfounded reporting, direct cover-up of anomalies and cross-permission execution. The evaluation and verification module is used to quantify and score the execution results of operation and maintenance tasks based on multiple preset dimensions and weights. The dimensions include data accuracy, process compliance, timeliness achievement rate and anomaly recall rate. It outputs a single inspection score and triggers manual review to generate an evaluation feedback score when an anomaly or inaccurate data is detected. The feedback optimization module is used to store typical cases of missed reports, false reports, and misreports into the sample library to form training data, and to automatically correct the inspection threshold based on historical data and human feedback, as well as to automatically iterate prompt words and rule logic through manual annotation to improve the accuracy of the agent. The scheduling and coordination module is used to automatically retry and implement timeout circuit breaking for failures or timeouts during task execution, and to perform delayed retries for scenarios where data is not fully acquired to ensure the integrity of data collection.

[0012] This invention also provides an operation and maintenance management method based on a large model, comprising: The application service layer receives operation and maintenance task requests from clients. The AI ​​agent layer plans the task steps based on the type of the request and generates task execution instructions. The data pipeline layer responds to the instructions, acquires real-time operation and maintenance data streams and related knowledge, and inputs them into the large model application layer; The large model application layer generates decision results based on the data and knowledge, and then returns them to the AI ​​agent layer; The AI ​​agent layer invokes the functional intelligent agent based on the decision results to generate executable instructions or structured reports. The instructions or reports are delivered through the application service layer.

[0013] The present invention 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 an operation and maintenance management system based on a large model as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an operation and maintenance management system based on a large model as described above.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements an operation and maintenance management system based on a large model as described above.

[0016] This invention provides a large-scale model-based operation and maintenance management method, device, equipment, and storage medium. The device receives various operation and maintenance task requests uniformly at the application service layer, and the AI ​​agent layer performs intelligent task planning and instruction generation, achieving centralized and intelligent task scheduling. The data pipeline layer responds to instructions, simultaneously executing multi-protocol real-time data acquisition and related knowledge retrieval based on a vectorized knowledge base, effectively breaking down data silos and providing complete information input that integrates real-time status and historical experience for decision-making. The core lies in the large-scale model application layer performing deep collaborative analysis and reasoning on the incoming real-time data stream and related knowledge, generating decision results containing root cause judgments and processing suggestions, thereby accurately applying the cognitive capabilities of the large-scale model to operation and maintenance decision points. Finally, based on this decision result, the AI ​​agent layer schedules specialized functional agents to generate directly executable operation instructions or structured reports, which are then delivered through the application service layer, forming a complete closed loop of "perception-cognition-decision-execution". This device fundamentally changes the traditional operation and maintenance model, freeing people from tedious and repetitive inspection, troubleshooting and reporting work, greatly improving fault response speed, processing accuracy and knowledge reuse efficiency, and realizing the automation, intelligence and knowledge-based closed-loop management of operation and maintenance work. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1This is a schematic diagram of the structure of the operation and maintenance management system based on a large model provided by the present invention; Figure 2 This is a flowchart illustrating the operation and maintenance management method based on a large model provided by the present invention; Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] To address the problems in existing technologies, this invention proposes a large-scale model-based operation and maintenance management system to achieve automated, intelligent, and knowledge-based closed-loop management of operation and maintenance work. The large-scale model-based operation and maintenance management system is described below, as follows... Figure 1 As shown, including but not limited to: Application service layer 110 is used to receive operation and maintenance task requests from clients; AI agent layer 120, connected to the application service layer, is used to plan task steps and generate task execution instructions according to the type of the operation and maintenance task request. The data pipeline layer 130 is used to respond to the task execution instruction to obtain real-time operation and maintenance data streams related to the task and related knowledge related to the task; and input the real-time operation and maintenance data streams and the related knowledge to the large model application layer; The large model application layer 140 is used to generate decision results based on the real-time operation and maintenance data stream and the associated knowledge, and return them to the AI ​​agent layer. The AI ​​agent layer 120 is also used to invoke the corresponding functional agent to generate executable instructions or structured reports based on the decision results, and deliver them through the application service layer.

[0021] The application service layer serves as a unified portal for interaction between the system and external users or clients. Its core function is to receive and initially process operation and maintenance task requests from various channels. In actual deployment, this layer can integrate multiple access channels such as DingTalk robots, WeChat mini-programs, and web-based management interfaces. For example, operation and maintenance personnel can directly @ the robot in a DingTalk work group and describe the problem in natural language, or submit a complex capacity prediction and analysis task through the web console. Regardless of the source or form of the request, the application service layer is responsible for standardizing it into a structured request format that the internal system can understand and accurately passing it to the next layer, namely the AI ​​agent layer.

[0022] The AI ​​agent layer is directly connected to the application service layer, acting as the "intelligent scheduling center" of the entire system. When it receives a specific operation and maintenance task request (such as "analyze the CPU alarms of server A") from the application service layer, it first parses the type of the request (e.g., real-time alarm analysis). Based on its built-in task planning logic and knowledge, the AI ​​agent layer decomposes this macro-task into a series of executable atomic steps. Subsequently, it generates a detailed and machine-readable "task execution instruction" based on this step sequence. This instruction is essentially a data work order, which clearly lists the specific operations and parameters that the downstream data pipeline layer needs to perform to complete this task, such as: "query the CPU usage time-series data of server A for the past 5 minutes from the Prometheus system via the MCP protocol" and "retrieve fault handling plans and historical cases related to 'high CPU usage' from the knowledge base."

[0023] The data pipeline layer is specifically designed to respond to and execute "task execution instructions" from the AI ​​agent layer. It comprises two core functional modules: a multi-protocol communication module and a knowledge retrieval enhancement module. The multi-protocol communication module is responsible for interacting with diverse external monitoring systems and databases, dynamically adapting to different communication protocols, and acquiring real-time, structured operational data streams according to instruction requirements. The knowledge retrieval enhancement module manages a vectorized operational knowledge base, storing knowledge fragments transformed from documents such as historical fault reports, emergency plans, and system architecture diagrams. Upon receiving an instruction, this module can quickly retrieve highly relevant knowledge from a vast amount of knowledge through semantic understanding. After synchronously or asynchronously completing real-time data acquisition and relevant knowledge retrieval, the data pipeline layer merges these two types of information into a complete contextual information package and directly delivers it to the large model application layer.

[0024] The large model application layer is the system's "intelligent analysis and decision engine." It receives contextual information packages from the data pipeline layer, integrating real-time dynamics and historical experience. Based on the built-in large language model and its reasoning capabilities optimized for the operations and maintenance domain, this layer performs in-depth analysis and logical association of the input information. For example, when the information input from the data pipeline layer indicates that a business system is experiencing periodic instability in database access, accompanied by automatic restarts of multiple application nodes and triggering numerous database connection interruptions and reconnections, the collaborative reasoning engine will not view these phenomena in isolation. It will associate, compare, and comprehensively reason with this series of complex real-time monitoring data and log information, along with historical in-depth analysis cases retrieved from the knowledge base, such as "under certain conditions, performing specific operations on a large table containing compressed indexes in a certain version of the database will trigger an internal bug, leading to connection anomalies." Through logical fitting of the phenomenon chain and historical root cause knowledge, the engine can generate a structured "decision result." This result not only points out the operation suspected of triggering the internal database bug but may also be related to the specific database version, table structure characteristics, and avoidance suggestions, thereby providing operations and maintenance personnel with accurate diagnosis and action guidelines that directly target the core of the problem. For example, it compares and synthesizes current high CPU usage data and information on the main processes consuming the CPU with historical cases of "Java memory leaks causing frequent garbage collection," generating a structured "decision result." This result not only includes a judgment on the root cause of the problem but also specific handling suggestions, operational steps, and even script names for reference, and is returned to the AI ​​agent layer in a standard format.

[0025] After receiving the "decision result" from the large model application layer, the AI ​​agent layer enters the final stage of task execution. Based on the nature of the decision result (e.g., the conclusion is to execute a diagnostic script), it calls the most suitable agent from a set of pre-defined functional agents (such as script execution agents, report generation agents, question-answering agents, etc.). The AI ​​agent layer passes the decision result as input to that functional agent, driving it to generate the final deliverable. For diagnostic scenarios, this might be a specific command-line instruction or an automated script; for analytical scenarios, it might be a structured analysis report with illustrations. Finally, the AI ​​agent layer delivers the generated instructions or report precisely to the client that initially initiated the request by calling the corresponding interface of the application service layer (e.g., pushing diagnostic suggestions and pending confirmation commands back to the DingTalk group), thus completing a full intelligent closed loop from task triggering to result delivery.

[0026] To illustrate the workflow more concretely, a server CPU alarm handling scenario is used as an example: Maintenance personnel report an anomaly via DingTalk; the application service layer receives and forwards the report; the AI ​​agent layer plans the steps of "data retrieval and knowledge lookup" and issues instructions; the data pipeline layer acquires real-time monitoring indicators and similar fault contingency plans in parallel, packages them, and sends them to the large model application layer; after analysis, the large model determines it to be a memory leak and suggests executing a specific check script; the AI ​​agent layer then calls the script to execute the intelligent agent to generate specific commands, and finally, the diagnostic conclusions and operational suggestions are fed back to the maintenance personnel via DingTalk. The entire process achieves automated integration from perception, analysis, decision-making to suggestion execution, greatly improving the efficiency and intelligence of maintenance response.

[0027] As a further optional embodiment, the AI ​​agent layer includes: A task planning engine is used to parse the type of the operation and maintenance task request and generate a sequence of task steps containing at least one atomic operation based on predefined task decomposition rules. The instruction generation module is used to construct task execution instructions to be sent to the data pipeline layer based on the task step sequence. The task execution instructions include data acquisition parameters and knowledge retrieval parameters corresponding to each atomic operation.

[0028] In a more preferred embodiment, the internal structure of the AI ​​agent layer can be further refined to more clearly realize its intelligent scheduling and planning functions. Specifically, the AI ​​agent layer may include a task planning engine and an instruction generation module.

[0029] The task planning engine, the logical core of the AI ​​agent layer, is responsible for deep analysis of raw operation and maintenance task requests received from the application service layer. It first identifies the macro-level category of the request, such as "proactive inspection," "fault alarm analysis," "capacity consultation," or "knowledge Q&A." Then, based on pre-configured task decomposition rules and logical templates for each category, the engine breaks down and arranges the identified macro-level task into a sequence of task steps consisting of at least one atomic operation. For example, for a "fault alarm analysis" request, its step sequence might be planned as follows: First, obtain the real-time performance indicators of a specific target; second, retrieve historical records of handling similar faults and related knowledge.

[0030] The instruction generation module works in conjunction with the task planning engine to transform the abstract sequence of task steps into specific operation commands that the data pipeline layer can directly recognize and execute—the task execution instructions. This module fills in the specific parameters required for the execution of each atomic operation in the step sequence. These parameters explicitly specify the source, method, and content of data acquisition (data acquisition parameters), as well as the query semantics and scope of knowledge retrieval (knowledge retrieval parameters). The resulting task execution instructions are a structured and unambiguous working guide, ensuring the accuracy and efficiency of subsequent data acquisition and knowledge retrieval operations.

[0031] As a further optional embodiment, the data pipeline layer includes: The Multiprotocol Communication (MCP) module is used to respond to the task execution command by acquiring the real-time operation and maintenance data stream from at least one external monitoring system via the MCP communication protocol. The Knowledge Retrieval Enhancement (RAG) module is used to retrieve the task-related associated knowledge from a vectorized knowledge base in response to the task execution instruction.

[0032] The multi-protocol communication module, acting as a bridge between the system and external data sources, is dedicated to responding to task execution commands from the AI ​​agent layer. Its core capability lies in dynamically adapting to and connecting to one or more heterogeneous external monitoring systems, databases, or data streaming platforms via the MCP communication protocol. When the command includes data acquisition parameters, this module can automatically select or switch the corresponding communication protocol (such as PromQL, SQL, SNMP, Kafka consumption, etc.) based on the target system type specified in the parameters (e.g., cloud monitoring platform, physical device network management, business database, etc.) and query requirements. It establishes a secure connection and executes precise data queries or streaming acquisition, thereby reliably obtaining real-time operational data streams highly relevant to the task. For example, to analyze server performance, this module may simultaneously extract CPU metrics from a time-series database and collect recent error logs from a log platform.

[0033] The knowledge retrieval enhancement module focuses on processing knowledge retrieval parameters in task execution instructions to activate and utilize the operational knowledge assets built within the system. At its core is a deeply processed vectorized knowledge base, which stores vector representations of unstructured documents such as historical fault reports, emergency plans, technical manuals, and system architecture diagrams, transformed through an embedding model. When the module receives a retrieval request, it first converts the request's semantics (e.g., "common causes of high CPU usage") into a query vector. Then, it performs efficient similarity matching and semantic search within the vector knowledge base, ultimately retrieving several knowledge fragments most relevant and valuable to the current task context as the associated knowledge. This process enables the precise location of effective information from massive amounts of disorganized documents.

[0034] In actual operation, these two modules typically work in parallel or in a pipeline manner according to task instructions: the multi-protocol communication module acquires the real-time status of the system externally, while the knowledge retrieval enhancement module mines historical experience and solutions internally. The real-time data streams and related knowledge they acquire are then integrated into a structured information set, which is uniformly transported from the data pipeline layer to the large model application layer, providing a complete information input that includes both current dynamics and rich historical experience for subsequent intelligent analysis and decision-making.

[0035] As a further optional embodiment, the large model application layer includes: The receiving module is used to receive the real-time operation and maintenance data stream and the associated knowledge input from the data pipeline layer, and to fuse the real-time operation and maintenance data stream and the associated knowledge. The collaborative reasoning engine is used to comprehensively analyze the fused real-time operation and maintenance data stream and the associated knowledge, and generate decision results including problem diagnosis, root cause inference and handling suggestions based on natural language understanding and preset operation and maintenance logic rules. The output module is used to return the decision results in a structured manner to the AI ​​agent layer, so that it can call the corresponding functional agent to execute.

[0036] As a further optional embodiment, the large model application layer can be specifically divided into a receiving module, a collaborative inference engine, and an output module, with each module working together to complete the complete chain of information fusion, intelligent analysis, and result feedback.

[0037] The receiving module, serving as the data entry point for the large model application layer, is responsible for receiving two types of key information input from the data pipeline layer: real-time operation and maintenance data streams and related knowledge. A key function of this module is the effective fusion of these two types of heterogeneous information. For example, it integrates structured or semi-structured data such as time-series monitoring data and log entries with unstructured knowledge such as retrieved textual historical cases and processing steps, into a coherent and context-rich comprehensive input material. This fusion process lays a unified information foundation for subsequent in-depth analysis.

[0038] The collaborative reasoning engine is the core of this layer's intelligence. It receives integrated information after fusion processing and drives the underlying large language model to conduct comprehensive analysis. This engine not only relies on the inherent natural language understanding and generation capabilities of the large model, but also typically incorporates logical rules and reasoning frameworks specifically pre-set or trained for the operations and maintenance domain. By comparing, correlating, and inferring causality between real-time phenomena and historical experience, the engine can simulate expert thinking and output structured decision results. These results typically include a clear diagnosis of the current problem (e.g., "service unavailable"), reasonable speculation about potential root causes (e.g., "due to database connection pool exhaustion"), and specific, actionable processing suggestions (e.g., "restart the relevant service and expand the connection pool").

[0039] The output module is responsible for converting the decision results generated by the collaborative reasoning engine, which may contain complex logic and text, into a standardized, structured data format (such as JSON, XML, or a specific protocol buffer format), and returning it to the AI ​​agent layer via a defined application programming interface. This structured return format ensures that the AI ​​agent layer can unambiguously parse the intent, conclusion, and parameters in the decision results, thereby accurately invoking the corresponding downstream functional agents to perform specific operations.

[0040] As a further optional embodiment, the AI ​​agent layer also includes: The agent scheduling module is used to match and call the corresponding functional agents from the pre-configured agent library according to the task type and execution strategy indicated in the decision result; The parameter passing module is used to pass the description of the problem to be processed, the root cause analysis conclusions and the suggested operations included in the decision results as input parameters to the called functional agent. The result generation module is used to drive the functional agent to perform at least one of the following operations based on the input parameters: Generate executable instructions containing specific operation commands or scripts; Generate a structured report that includes statistical charts, root cause analysis summaries, and treatment steps; The AI ​​agent layer delivers the executable instructions or structured reports to the corresponding client through the interface provided by the application service layer.

[0041] As a further optional embodiment, the AI ​​agent layer may also include more refined internal functional modules to specifically implement its ability to execute and deliver decision results. These modules include an agent scheduling module, a parameter transmission module, and a result generation module, which work together to complete the transformation from intelligent decision-making to concrete output.

[0042] The agent scheduling module is responsible for parsing the decision results returned by the large model application layer, identifying the implicit or explicit task types (e.g., "immediately perform diagnostics", "generate analysis reports", or "perform capacity planning") and execution strategies. Based on this parsing result, the module predefines, matches, and calls the most suitable functional agent from the system's pre-configured agent library.

[0043] The parameter transmission module intervenes immediately after the functional agent is invoked. It is responsible for extracting key contextual information and operational guidance from the decision results, including a detailed description of the problem to be addressed, the root cause conclusions drawn from the analysis, and the operational steps suggested by the large model. Subsequently, this module uses this information as structured input parameters and accurately transmits it to the invoked functional agent, ensuring that the agent begins working in the correct context.

[0044] The result generation module is crucial for driving the functional agent to produce final value. This module is responsible for initiating and managing the operation of the functional agent, driving it to execute its encapsulated professional capabilities based on received input parameters. Specifically, the result generation module drives the agent to complete at least one of the following operations: first, generating executable instructions containing specific operation commands and scripts that can be run automatically; second, generating a structured report that integrates data statistics charts, root cause analysis essentials, and standard processing procedures.

[0045] Finally, through the coordination of the AI ​​agent layer, the generated executable instructions or structured reports will be delivered to the corresponding client that initiated the original request by calling the standardized interfaces (such as message push API, file download links, etc.) provided by the application service layer, thereby completing the end-to-end closed loop from intelligent analysis to result delivery.

[0046] As a further optional embodiment, the structured report generated by the result generation module driving the functional agent includes at least one of the following forms: Statistical analysis charts based on fault data; Trend forecasting charts based on resource usage data; A failure review report that includes the root cause chain and the timeline of the handling process.

[0047] As a further optional embodiment, the functional agent driven by the result generation module can generate various structured reports with direct business value. These reports are not simply data compilations, but rather decision support materials that have undergone intelligent analysis and formatting, specifically manifested in at least one of the following forms: Firstly, there are statistical analysis charts based on historical or real-time fault data. For example, knowledge-based question-answering or analytical prediction agents can aggregate, classify, and calculate alarm events over a period of time, generating bar charts, pie charts, or trend curves that display the fault distribution of each system and the proportion of high-frequency error types, providing quantitative insights for operation and maintenance management.

[0048] Secondly, there are trend prediction charts based on historical resource usage data. For example, predictive intelligence agents can perform time-series analysis on indicators such as server CPU, memory, and disk I / O, and use prediction algorithms to generate prediction lines and confidence intervals for resource usage over a future period, presenting them in an intuitive chart format to provide data support for capacity planning.

[0049] Thirdly, there are fault review reports for single major events, including root cause chains and processing timelines. For example, after handling a complex fault, the corresponding intelligent agent can integrate the entire process data from alarm triggering, multi-stage diagnosis, measures taken to recovery confirmation, respond to human instructions, draw a clear event timeline, and display the root cause logic chain inferred or confirmed by a large model, forming a complete document that can be used for post-event review and experience accumulation.

[0050] By generating the highly structured reports described above, the system not only automates operations but also productizes data analysis, knowledge summarization, and report compilation capabilities, significantly improving the standardization level and decision-making efficiency of operation and maintenance management.

[0051] As a further optional embodiment, the functional agent includes at least one of the following: Inspection-type intelligent agents are used to automatically perform system health status inspections based on scheduled tasks. Analytical and predictive intelligent agents are used for trend analysis and capacity prediction based on historical data; A knowledge-based question-answering agent is used to respond to users' natural language questions and generate answers from a knowledge base; Database query-type intelligent agents are used to convert natural language queries into database query statements and execute them.

[0052] As a further optional embodiment, the functional intelligent agents are the specific capability carriers that enable the AI ​​agent layer to perform diverse and specialized operation and maintenance tasks. These intelligent agents exist in the form of software modules or services, are pre-developed and registered in an intelligent agent library, and each encapsulates proprietary logic for a specific scenario. In a specific implementation, the functional intelligent agent library may contain at least one of the following types of intelligent agents: Inspection-type intelligent agents are designed to automatically perform comprehensive or specialized checks on the health status of designated IT infrastructure, application services, or business systems based on pre-set scheduled tasks or event triggers. They typically invoke data pipeline layers to obtain real-time metrics, determine whether the status is abnormal based on rules or simple models, and ultimately generate standardized inspection result summaries or reports.

[0053] Analytical and predictive agents specialize in in-depth mining and modeling of historical operational data. They can perform trend analysis, such as predicting future capacity demand based on past resource usage data; and they can perform pattern recognition, such as clustering fault tickets to identify common problems and optimization points. Their output is typically a structured analysis report containing predictive curves, statistical conclusions, or optimization suggestions.

[0054] This knowledge-based question-answering AI agent aims to respond to various operation and maintenance-related questions raised by users in a natural dialogue manner. When users inquire about policies, processes, technical solutions, or historical faults, the AI ​​agent coordinates the data pipeline layer to retrieve relevant information from the knowledge base and leverages the application capabilities of large models to generate easy-to-understand and coherent answers, acting as an online "operation and maintenance expert."

[0055] Database query-based intelligent agents primarily address the pain point of non-technical personnel being able to easily access business data. They can understand users' data query needs expressed in natural language (e.g., "query the number of failed transactions for service A yesterday"), accurately translate them into standardized database query statements through built-in semantic understanding and conversion components, automatically execute the query, and return the results to the user in tabular or natural language summary form.

[0056] By configuring the above-mentioned different types of functional intelligent agents, the AI ​​agent layer can flexibly and accurately respond to the full spectrum of operation and maintenance needs, from automated operation to intelligent analysis, and from information consultation to data acquisition, thereby greatly expanding the application scope and practical value of the system.

[0057] As a further optional embodiment, the system further includes: a driving layer, the driving layer comprising the following modules: The constraint module is used to define the mandatory and prohibited items, legal value range and reporting format of operation and maintenance tasks, and to establish a verification mechanism to verify data items, value range and data time, as well as set red line constraints to prohibit unfounded reporting, direct cover-up of anomalies and cross-permission execution. The evaluation and verification module is used to quantify and score the execution results of operation and maintenance tasks based on multiple preset dimensions and weights. The dimensions include data accuracy, process compliance, timeliness achievement rate and anomaly recall rate. It outputs a single inspection score and triggers manual review to generate an evaluation feedback score when an anomaly or inaccurate data is detected. The feedback optimization module is used to store typical cases of missed reports, false reports, and misreports into the sample library to form training data, and to automatically correct the inspection threshold based on historical data and human feedback, as well as to automatically iterate prompt words and rule logic through manual annotation to improve the accuracy of the agent. The scheduling and coordination module is used to automatically retry and implement timeout circuit breaking for failures or timeouts during task execution, and to perform delayed retries for scenarios where data is not fully acquired to ensure the integrity of data collection.

[0058] In this embodiment, the constraint module is used to enforce regulations on the scope, data format, and behavioral boundaries of operation and maintenance operations before or during task execution. This module first predefines a series of insurmountable constraints based on the operation and maintenance strategy, specifically including: mandatory items for inspection tasks (such as the health status of core business services must be checked) and prohibited items (such as unauthorized sensitive data must not be collected); the legal value range of all data items (such as CPU utilization should be between 0-100%, and status codes should only allow enumerated values); and standardized reporting formats (such as inspection reports must follow a fixed JSON schema). During task execution, the constraint module acts as a pre-filter, verifying whether the instructions to be issued or the data to be reported meet the above constraints: verifying the integrity of data items, the reasonableness of the numerical range, and the freshness of the data (such as data older than 5 minutes is not allowed). Meanwhile, the module clearly defines three "red lines": prohibiting unfounded reporting (i.e., any alarm or conclusion must be accompanied by evidence or reasoning), prohibiting the direct concealment of anomalies (i.e., not allowing error or alarm information to be discarded silently), and prohibiting cross-permission execution (i.e., the AI ​​agent layer must not call unauthorized high-risk operation commands). Any behavior that violates these constraints will be intercepted by the module and trigger an alarm, thereby achieving the goal of "preventing errors before they occur".

[0059] The evaluation and verification module's core responsibility is to quantitatively and multidimensionally assess the quality of completed task execution results to verify whether they meet expected standards. This module employs a 100-point scoring system with four assessment dimensions and corresponding weights: Data Accuracy (40%), specifically assessing data consistency (the degree of matching with the actual source data), completeness (whether necessary fields are missing), and logical consistency (e.g., whether alarm levels and numerical ranges are self-consistent); Process Compliance (25%), assessing whether the task is executed according to the prescribed sequence of steps and whether traceable log records are maintained throughout the process; Timeliness Compliance (20%), assessing whether the task is completed within the prescribed time limit (e.g., whether inspections are started on time and results are submitted on time); and Anomaly Recall Rate (15%), assessing whether the system can correctly detect genuine anomalies (i.e., "no alarms missed") while avoiding false alarms for normal states. After a task (e.g., an automatic inspection) is completed, the evaluation and verification module automatically calculates a single score based on the above indicators. If the score is below the preset threshold, or if the module finds obvious anomalies or inaccuracies during data verification, a manual review process will be triggered. Operations personnel will then conduct a second evaluation of the results and provide a corrected score. This manual evaluation feedback will be recorded and used for subsequent model optimization.

[0060] The feedback tuning module is responsible for transforming the quality evaluations and human feedback generated by the assessment and verification module into the system's ability for self-iteration and continuous improvement. This module first builds a sample library, categorizing and storing typical cases from historical execution results, particularly types such as "missed reports" (actual anomalies that the system failed to detect), "false alarms" (the system reports anomalies when the system is actually functioning correctly), and "incorrect reports" (the system's given reasons or suggestions are incorrect). These cases, after standardization, anonymization, and labeling, form high-quality training data suitable for fine-tuning large models or optimizing prompts. Based on accumulated historical data and human feedback, the feedback tuning module can automatically correct inspection thresholds: for example, if the alarm threshold for a certain indicator frequently leads to false alarms, the module can automatically raise or lower the threshold based on statistical distribution, reducing the workload of repeated manual adjustments. Meanwhile, for the prompt words used in the large model application layer and the rule logic of the agents in the AI ​​agent layer, this module also provides automatic iteration capability based on manual annotation. When the operation and maintenance personnel explicitly adopt, modify or reject a decision suggestion in the interactive interface, the system will capture the difference and convert it into a correction of the underlying prompt words (such as supplementing counterexamples and adjusting the instruction format), thereby continuously improving the output accuracy of each agent.

[0061] The scheduling and coordination module focuses on ensuring the stability and eventual consistency of operational tasks in the face of unreliable networks, limited resources, or temporary failures. Built into the control layer, this module manages all calls sent to the data pipeline layer or external systems. For temporary anomalies such as inspection failures (e.g., no response from the target system) and interface timeouts (e.g., monitoring queries exceeding a preset waiting time), the scheduling and coordination module automatically executes an exponential backoff retry strategy to prevent task failures due to momentary fluctuations. Simultaneously, to prevent downstream system avalanche or resource exhaustion, the module also implements a timeout circuit breaker mechanism: when the same target fails consecutively a certain number of times, the request channel is temporarily cut off and automatically restored after a cooldown period. For situations where data is incomplete (e.g., the distributed database backup has not yet been synchronized) or dependent resources are not ready (e.g., a certain metric collection point is missing), the module supports delayed retries—suspending the current task and waiting for a preset delay before attempting to acquire data again, thereby ensuring the integrity of data collection and significantly reducing manual intervention caused by environmental fluctuations. Through the above mechanism, the scheduling and coordination module enables the entire operation and maintenance center to have high fault tolerance and execution reliability in complex production environments.

[0062] In summary, the four modules of the control layer—constraint module, evaluation and verification module, feedback and optimization module, and scheduling and coordination module—constitute a complete and closed-loop quality control system from four dimensions: standard boundaries, quality quantification, self-iteration, and stability assurance. This significantly improves the reliability, accuracy, and long-term adaptive capability of the operation and maintenance management device in actual production environments.

[0063] The following describes the operation and maintenance management method based on a large model provided by this invention, such as... Figure 2 As shown, the large-model-based operation and maintenance management method described below can be referred to in correspondence with the large-model-based operation and maintenance management system described above.

[0064] A large-scale model-based operation and maintenance management method includes: Step 210: Receive maintenance task requests from clients through the application service layer; Step 220: The AI ​​agent layer plans the task steps according to the type of the request and generates task execution instructions; Step 230: Respond to the instruction through the data pipeline layer, obtain real-time operation and maintenance data streams and related knowledge, and input them into the large model application layer; Step 240: The large model application layer generates decision results based on the data and knowledge, and returns them to the AI ​​agent layer; Step 250: The AI ​​agent layer invokes the functional agent based on the decision result to generate executable instructions or structured reports, and delivers the instructions or reports through the application service layer.

[0065] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3 As shown, the electronic device may include: a processor 310, a communications interface 320, a memory 330, and a communication bus 340, wherein the processor 310, the communications interface 320, and the memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions in the memory 330 to execute a large-model-based operation and maintenance management system. The method includes: The application service layer receives operation and maintenance task requests from clients. The AI ​​agent layer plans the task steps based on the type of the request and generates task execution instructions. The data pipeline layer responds to the instructions, acquires real-time operation and maintenance data streams and related knowledge, and inputs them into the large model application layer; The large model application layer generates decision results based on the data and knowledge, and then returns them to the AI ​​agent layer; The AI ​​agent layer invokes the functional intelligent agent based on the decision results to generate executable instructions or structured reports. The instructions or reports are delivered through the application service layer.

[0066] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0067] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to execute the large-model-based operation and maintenance management system provided by the above methods, the method comprising: The application service layer receives operation and maintenance task requests from clients. The AI ​​agent layer plans the task steps based on the type of the request and generates task execution instructions. The data pipeline layer responds to the instructions, acquires real-time operation and maintenance data streams and related knowledge, and inputs them into the large model application layer; The large model application layer generates decision results based on the data and knowledge, and then returns them to the AI ​​agent layer; The AI ​​agent layer invokes the functional intelligent agent based on the decision results to generate executable instructions or structured reports. The instructions or reports are delivered through the application service layer.

[0068] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the large-model-based operation and maintenance management system provided by the methods described above, the method comprising: The application service layer receives operation and maintenance task requests from clients. The AI ​​agent layer plans the task steps based on the type of the request and generates task execution instructions. The data pipeline layer responds to the instructions, acquires real-time operation and maintenance data streams and related knowledge, and inputs them into the large model application layer; The large model application layer generates decision results based on the data and knowledge, and then returns them to the AI ​​agent layer; The AI ​​agent layer invokes the functional intelligent agent based on the decision results to generate executable instructions or structured reports. The instructions or reports are delivered through the application service layer.

[0069] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; 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. Those skilled in the art can understand and implement this without any creative effort.

[0070] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A large-scale model-based operation and maintenance management system, characterized in that, include: The application service layer is used to receive operation and maintenance task requests from clients; The AI ​​agent layer, connected to the application service layer, is used to plan task steps and generate task execution instructions based on the type of the operation and maintenance task request. The data pipeline layer is used to respond to the task execution command to obtain real-time operation and maintenance data streams related to the task and related knowledge related to the task, and input the real-time operation and maintenance data streams and related knowledge to the large model application layer; The large model application layer is used to generate decision results based on the real-time operation and maintenance data stream and the associated knowledge, and return them to the AI ​​agent layer; The AI ​​agent layer is also used to invoke the corresponding functional agent to generate executable instructions or structured reports based on the decision results, and deliver them through the application service layer.

2. The operation and maintenance management system based on a large model according to claim 1, characterized in that, The AI ​​agent layer includes: A task planning engine is used to parse the type of the operation and maintenance task request and generate a sequence of task steps containing at least one atomic operation based on predefined task decomposition rules. The instruction generation module is used to construct task execution instructions to be sent to the data pipeline layer based on the task step sequence. The task execution instructions include data acquisition parameters and knowledge retrieval parameters corresponding to each atomic operation.

3. The operation and maintenance management system based on a large model according to claim 1, characterized in that, The data pipeline layer includes: The multi-protocol communication (MCP) module is used to obtain the real-time operation and maintenance data stream from at least one external monitoring system in response to the task execution command via the MCP communication protocol. The Knowledge Retrieval Enhancement (RAG) module is used to retrieve the task-related associated knowledge from a vectorized knowledge base in response to the task execution instruction.

4. The operation and maintenance management system based on a large model according to claim 1, characterized in that, The large model application layer includes: The receiving module is used to receive the real-time operation and maintenance data stream and the associated knowledge input from the data pipeline layer, and to fuse the real-time operation and maintenance data stream and the associated knowledge. The collaborative reasoning engine is used to comprehensively analyze the fused real-time operation and maintenance data stream and the associated knowledge, and generate decision results including problem diagnosis, root cause inference and handling suggestions based on natural language understanding and preset operation and maintenance logic rules. The output module is used to return the decision results in a structured manner to the AI ​​agent layer, so that it can call the corresponding functional agent to execute.

5. The operation and maintenance management system based on a large model according to claim 1, characterized in that, The AI ​​agent layer also includes: The agent scheduling module is used to match and call the corresponding functional agents from the pre-configured agent library according to the task type and execution strategy indicated in the decision result; The parameter passing module is used to pass the description of the problem to be processed, the root cause analysis conclusions and the suggested operations included in the decision results as input parameters to the called functional agent. The result generation module is used to drive the functional agent to perform at least one of the following operations based on the input parameters: Generate executable instructions containing specific operation commands or scripts; Generate a structured report that includes statistical charts, root cause analysis summaries, and treatment steps; The AI ​​agent layer delivers the executable instructions or structured reports to the corresponding client through the interface provided by the application service layer.

6. The operation and maintenance management system based on a large model according to claim 5, characterized in that, The structured report generated by the result generation module driving the functional agent includes at least one of the following forms: Statistical analysis charts based on fault data; Trend forecasting charts based on resource usage data; A failure review report that includes the root cause chain and the timeline of the handling process.

7. The operation and maintenance management system based on a large model according to claim 1, characterized in that, The system further includes a driving layer, which comprises the following modules: The constraint module is used to define the mandatory and prohibited items, legal value range and reporting format of operation and maintenance tasks, and to establish a verification mechanism to verify data items, value range and data time, as well as set red line constraints to prohibit unfounded reporting, direct cover-up of anomalies and cross-permission execution. The evaluation and verification module is used to quantify and score the execution results of operation and maintenance tasks based on multiple preset dimensions and weights. The dimensions include data accuracy, process compliance, timeliness achievement rate and anomaly recall rate. It outputs a single inspection score and triggers manual review to generate an evaluation feedback score when an anomaly or inaccurate data is detected. The feedback optimization module is used to store typical cases of missed reports, false reports, and misreports into the sample library to form training data, and to automatically correct the inspection threshold based on historical data and human feedback, as well as to automatically iterate prompt words and rule logic through manual annotation to improve the accuracy of the agent. The scheduling and coordination module is used to automatically retry and implement timeout circuit breaking for failures or timeouts during task execution, and to perform delayed retries for scenarios where data is not fully acquired to ensure the integrity of data collection.

8. A large-scale model-based operation and maintenance management method, characterized in that, include: The application service layer receives operation and maintenance task requests from clients. The AI ​​agent layer plans the task steps and generates task execution instructions based on the type of the request. The data pipeline layer responds to the instructions, acquires real-time operation and maintenance data streams and related knowledge, and inputs them into the large model application layer; The large model application layer generates decision results based on the data and knowledge, and then returns them to the AI ​​agent layer; The AI ​​agent layer invokes the functional agent based on the decision result to generate executable instructions or structured reports, and delivers the instructions or reports through the application service layer.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the operation and maintenance management method based on a large model as described in claim 8.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the operation and maintenance management method based on a large model as described in claim 8.

Images