Cloud platform operation and maintenance system, operation and maintenance method and electronic equipment

Abstract

The embodiment of the invention provides a cloud platform operation and maintenance system, an operation and maintenance method and electronic equipment. The cloud platform operation and maintenance system comprises a work assignment agent, a monitoring sentry agent and an asset inspection agent. The work assignment agent not only can automatically complete a corresponding operation and maintenance operation task by using a large language model, but also can automatically call the other two agents according to user input, so that collaborative interaction between the agents is realized. The monitoring sentry agent can automatically complete a monitoring data analysis task by using a large language model. The asset inspection agent can utilize a large language model to automatically complete an inspection task. By applying the embodiment of the invention, full-system automation of on-demand activation and collaborative reasoning is realized, the operation and maintenance efficiency is improved, and the operation and maintenance cost of the cloud platform is reduced.

Description

Technical Field

[0001] This application relates to the field of cloud platform technology, and in particular to a cloud platform operation and maintenance system, operation and maintenance method and electronic device. Background Technology

[0002] As enterprises accelerate their digital transformation and information technology infrastructure continues to expand, cloud platform architectures are becoming increasingly complex, encompassing multi-level components from underlying physical devices and virtualized resources to upper-level microservice applications. Simultaneously, more and more operations and maintenance management systems are adopting automation tools and artificial intelligence technologies, such as machine learning and big data analytics, to enhance their capabilities in detection, prediction, and response during operations and maintenance.

[0003] However, the current traditional operation and maintenance model still relies on distributed tools to achieve operation and maintenance capabilities. Various operation and maintenance functions are isolated from each other. For example, using a monitoring system to collect performance indicators and provide alarms; using a log system to store log information, etc. These operation and maintenance functions need to be implemented by separate functional modules and require a lot of manual intervention from operation and maintenance personnel, resulting in low operation and maintenance efficiency.

[0004] While some functions in existing technologies can be implemented using pre-defined rule engines or shallow machine learning models, the various operation and maintenance modules remain isolated, requiring intervention from operation and maintenance personnel, and thus, operational efficiency needs further improvement. Summary of the Invention

[0005] The purpose of this application is to provide a cloud platform operation and maintenance system, operation and maintenance method, and electronic device to improve operation and maintenance efficiency and reduce operation and maintenance costs. The specific technical solution is as follows:

[0006] In a first aspect, embodiments of this application provide a cloud platform operation and maintenance system, including: a work assignment intelligent agent, a monitoring sentinel intelligent agent, and an asset inspection intelligent agent;

[0007] The task assignment agent is used to receive a natural language description of the current maintenance task input by the user and a list of target devices corresponding to the current maintenance task selected by the user; obtain the target agent assigned in the current maintenance task description; if the target agent is itself, use a large language model to generate one or more maintenance operation commands for each target device in the target device list based on the current maintenance task description and the target device list; issue the one or more maintenance operation commands to the corresponding devices for execution; if the target agent is the monitoring sentinel agent, invoke the monitoring sentinel agent to execute the current maintenance task; if the target agent is the asset inspection agent, invoke the asset inspection agent to execute the current maintenance task.

[0008] The monitoring sentinel agent is used to acquire one or more monitoring indicator data of the monitored devices, and to perform data analysis on the monitoring data of each monitored device using a large language model to generate analysis results in natural language form; the monitored devices are the devices in the operation and maintenance equipment list, or the monitored devices specified by the user.

[0009] The asset inspection intelligent agent is used to obtain the inspection script corresponding to each device to be inspected; and to inspect the device to be inspected based on each inspection script; the device to be inspected is the device in the maintenance equipment list, or the device in the inspection requirements input by the user through natural language description.

[0010] Secondly, this application provides a cloud platform operation and maintenance method, which is applied to a cloud platform operation and maintenance system. The cloud platform operation and maintenance system includes: a work assignment intelligent agent, a monitoring sentinel intelligent agent, and an asset inspection intelligent agent.

[0011] The method includes:

[0012] The task assignment agent receives a natural language description of the current maintenance task input by the user and a list of target devices corresponding to the current maintenance task selected by the user; obtains the target agent assigned in the current maintenance task description; if the target agent is itself, it uses a large language model to generate one or more maintenance operation commands for each target device in the target device list based on the current maintenance task description and the target device list; and sends the one or more maintenance operation commands to the corresponding devices for execution.

[0013] When the target intelligent agent is the monitoring sentinel intelligent agent, the monitoring sentinel intelligent agent is invoked to execute the operation and maintenance task; the monitoring sentinel intelligent agent obtains one or more monitoring indicator data of the monitored device, uses a large language model to perform data analysis on the monitoring data of each monitored device, and generates analysis results in natural language form; the monitored device is the device in the operation and maintenance device list, or the monitored device specified by the user.

[0014] When the target intelligent agent is an asset inspection intelligent agent, the asset inspection intelligent agent is invoked to execute the operation and maintenance task; so that the asset inspection intelligent agent obtains the inspection script corresponding to each device to be inspected; based on each inspection script, the device to be inspected is inspected; the device to be inspected is the device in the operation and maintenance equipment list, or the device in the inspection requirements input by the user through natural language description.

[0015] Thirdly, embodiments of this application provide an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor is used to implement the aforementioned cloud platform operation and maintenance method steps when executing the program stored in the memory.

[0016] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method steps of the aforementioned cloud platform operation and maintenance method.

[0017] Beneficial effects of the embodiments in this application:

[0018] This application provides a cloud platform operation and maintenance system, operation and maintenance method, and electronic device. The cloud platform operation and maintenance system includes three specialized intelligent agents: a work assignment agent, a monitoring sentinel agent, and an asset inspection agent. The work assignment agent can not only automatically complete corresponding operation and maintenance tasks using a large language model, but also automatically invoke the other two agents based on user input, achieving collaborative interaction between the agents. The monitoring sentinel agent can automatically complete monitoring data analysis tasks using a large language model. The asset inspection agent can automatically complete inspection tasks using a large language model. Applying this application embodiment achieves full system automation through on-demand activation and collaborative reasoning, overcoming the limitations of existing systems such as fragmented functional modules and manual intervention, improving operation and maintenance efficiency, and reducing cloud platform operation and maintenance costs.

[0019] Of course, implementing any product or method of this application does not necessarily require achieving all of the advantages described above at the same time. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application 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 only some embodiments of this application. For those skilled in the art, other embodiments can be obtained based on these drawings.

[0021] Figure 1 A schematic diagram of the structure of an example cloud platform operation and maintenance system provided in the embodiments of this application;

[0022] Figure 2 for Figure 1 The diagram shows an example of the interactive interface provided by the human-computer interaction interface in the cloud platform operation and maintenance system.

[0023] Figure 3A schematic diagram illustrating the structure of an example of a work assignment agent in a cloud platform operation and maintenance system provided in this application embodiment;

[0024] Figure 4 A flowchart illustrating an example of a work assignment agent in a cloud platform operation and maintenance system provided in this application embodiment;

[0025] Figure 5 This is a schematic diagram illustrating the structure of an example of a monitoring sentinel agent in a cloud platform operation and maintenance system provided in this application embodiment;

[0026] Figure 6 A flowchart illustrating an example of a monitoring sentinel agent in a cloud platform operation and maintenance system provided in this application embodiment;

[0027] Figure 7 A schematic diagram illustrating the structure of an example asset inspection intelligent agent in a cloud platform operation and maintenance system provided in this application embodiment;

[0028] Figure 8 A flowchart illustrating an example of an asset inspection intelligent agent in a cloud platform operation and maintenance system provided in this application embodiment;

[0029] Figure 9 A schematic diagram illustrating another example of the cloud platform operation and maintenance system provided in this application embodiment;

[0030] Figure 10 A flowchart illustrating an example of a cloud platform operation and maintenance method provided in this application embodiment;

[0031] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0032] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art based on this application are within the scope of protection of this application.

[0033] To improve operational efficiency and reduce operational costs, this application provides a cloud platform operation and maintenance system, operation and maintenance method, and electronic device.

[0034] First, the technical terms involved in the embodiments of this application will be introduced.

[0035] ICOP (Intelligent Cloud Operation Platform): It is used to uniformly manage infrastructure, platform services and application layer assets in a multi-cloud environment, and provides integrated operation and maintenance capabilities such as data collection, intelligent analysis, alarm diagnosis and automatic inspection.

[0036] An agent (intelligent agent) is an entity capable of perceiving its environment and taking actions to achieve specific goals. It can be software, hardware, or a system, possessing autonomy, adaptability, and interactivity. The agent perceives changes in the environment (e.g., through sensors or data input), makes judgments and decisions based on its learned knowledge and algorithms, and then executes actions to influence the environment or achieve predetermined goals. For example, an autonomous decision-making and execution unit built on a large language model, oriented towards a specific task, possesses data understanding, intent recognition, task planning, and result feedback capabilities.

[0037] LLM (Large Language Model): refers to a pre-trained artificial intelligence model with a large number of parameters and strong generalization and reasoning capabilities, used to support the cognitive and decision-making processes of intelligent agents.

[0038] RAG (Retrieval Augmented Generation): A generative AI (Artificial Intelligence) approach for large language models that enables generative AI applications to output more accurate answers by linking large language models to external resources.

[0039] The cloud platform operation and maintenance system provided in the embodiments of this application will be described in detail below. See also Figure 1 , Figure 1 This is a schematic diagram illustrating an example of a cloud platform operation and maintenance system provided in this application embodiment. The cloud platform operation and maintenance system includes: a work assignment agent 110, a monitoring sentinel agent 120, and an asset inspection agent 130;

[0040] The task assignment agent 110 is used to receive a natural language description of the current maintenance task input by the user and a list of target devices corresponding to the current maintenance task selected by the user; obtain the target agent assigned in the current maintenance task description; if the target agent is itself, use a large language model to generate one or more maintenance operation commands for each target device in the target device list based on the current maintenance task description and the target device list; issue the one or more maintenance operation commands to the corresponding device 100 for execution; if the target agent is the monitoring sentinel agent 120, invoke the monitoring sentinel agent 120 to execute the current maintenance task; if the target agent is the asset inspection agent 130, invoke the asset inspection agent 130 to execute the current maintenance task.

[0041] The monitoring sentinel agent 120 is used to acquire one or more monitoring indicator data of the monitored devices, and to perform data analysis on the monitoring data of each monitored device using a large language model to generate analysis results in natural language form; the monitored devices are the devices in the operation and maintenance equipment list, or the monitored devices specified by the user.

[0042] The asset inspection intelligent agent 130 is used to obtain the inspection script corresponding to each device to be inspected; and to inspect the device to be inspected based on each inspection script; the device to be inspected is the device in the maintenance equipment list, or the device in the inspection requirements input by the user through natural language description.

[0043] The large language model involved in this application embodiment refers to an open-source large language model or a large language model with operational semantic understanding and generation capabilities, or a public network API (Application Programming Interface).

[0044] The cloud platform operation and maintenance system provided in this application embodiment, with its work assignment intelligent agent 110, can not only automatically complete corresponding operation and maintenance tasks using a large language model, but also automatically call two other intelligent agents based on user input, realizing collaborative interaction between intelligent agents. The monitoring sentinel intelligent agent 120 can automatically complete monitoring data analysis tasks using a large language model. The asset inspection intelligent agent 130 can automatically complete inspection tasks using a large language model. Applying this application embodiment achieves full system automation through on-demand activation and collaborative reasoning, overcoming the limitations of existing systems such as fragmented functional modules, the need for manual troubleshooting of complex problems, and the need for manual annotation during deep learning model training. This improves operation and maintenance efficiency and reduces the manpower and time costs of cloud platform operation and maintenance.

[0045] In addition, by applying the cloud platform operation and maintenance system provided in the embodiments of this application, users can input a natural language description of the current operation and maintenance task and obtain the natural language output result, which reduces the difficulty of user input and understanding.

[0046] In some embodiments, such as Figure 1 As shown, the cloud platform operation and maintenance system also includes: an operation and maintenance database 140, a human-computer interaction interface 150, a time-series database 160, monitoring tools 170, and a unified asset management module 180. Figure 1 As shown, the human-computer interaction interface 150, time-series database 160, monitoring tool 170 and unified asset management module 180 belong to the ICOP platform and can be implemented based on relevant existing technologies.

[0047] The operations and maintenance database 140 includes a knowledge base 141 and a graph database 142. Knowledge base 141 stores various prior knowledge and is a pre-built database covering all operations and maintenance scenarios. It is primarily constructed through corpus organization, knowledge extraction, question-and-answer system construction, and retrieval strategy optimization of content such as fault case libraries, operations and maintenance asset product manuals, FAQ (Frequently Asked Questions) documents, SOP (Standard Operating Procedure) documents, and inspection script examples. Graph database 142 stores a knowledge graph of assets pre-established based on the equipment and other assets included in the cloud platform, primarily used for asset inspection.

[0048] The human-machine interface 150 is used for interaction between the system and the user. Specifically, the work assignment agent 110 can receive a natural language description of the maintenance task and a list of target devices selected by the user for that maintenance task through the human-machine interface 150. The monitoring sentinel agent 120 can receive monitoring commands triggered by the user and the specified monitored devices through the human-machine interface 150; the asset inspection agent 130 can receive inspection requests input by the user in natural language through the human-machine interface 150.

[0049] Specifically, the human-computer interaction interface 150 enables human-computer interaction by providing an interactive interface, see [link to relevant documentation]. Figure 2 , Figure 2 for Figure 1 The diagram shows an example of the human-machine interface provided by the cloud platform operation and maintenance system. As an example, this interface is used by the work assignment agent 110 to receive the natural language description of the current operation and maintenance task input by the user and the target device list corresponding to the current operation and maintenance task selected by the user. It includes two columns: the left column is used for the user to input the natural language description of the operation and maintenance task; the right column is used for the user to select target devices to generate a target device list.

[0050] In this embodiment, the monitoring sentinel agent 120 can execute maintenance tasks assigned by the work assignment agent 110 (monitoring data analysis tasks explicitly defined by the user in the current maintenance task description), or it can directly receive monitoring data analysis tasks from user-specified monitored devices. The asset inspection agent 130 can execute maintenance tasks assigned by the work assignment agent 110 (inspection tasks explicitly defined by the user in the current maintenance task description), or it can directly receive inspection requests input by the user through natural language descriptions and execute inspection tasks for the devices specified in the inspection requests.

[0051] In other words, the monitoring sentinel agent 120 and asset inspection agent 130 in the cloud platform operation and maintenance system provided in this application embodiment can not only be invoked by the work assignment agent 110, but also independently complete the corresponding maintenance tasks. Compared with the prior art, the way to complete operation and maintenance tasks is more flexible.

[0052] Figure 1 The time-series database 160 stores monitoring data obtained by the monitoring tool 170 from the device 100 for analysis by the monitoring sentinel agent 120. The unified asset management module 180 stores information on all devices in the system and manages all devices, allowing the asset inspection agent 130 to obtain information on devices to be inspected.

[0053] The following provides a detailed description of the work assignment agent 110, the monitoring sentinel agent 120, and the asset inspection agent 130 in the cloud platform operation and maintenance system.

[0054] The work assignment agent 110 in the cloud platform operation and maintenance system provided in this application embodiment serves as the central hub for task scheduling. It can receive users' natural language instructions, parse intentions, plan tasks, call command execution tools or other agents, and ensure safe execution.

[0055] See Figure 3 , Figure 3 This is a schematic diagram illustrating an example of a work assignment agent in a cloud platform operation and maintenance system provided in this application embodiment. The example work assignment agent includes: a task receiving module 301, a task parsing module 302, an agent invocation module 303, a command generation module 304, and a command issuing module 305.

[0056] The task receiving module 301 is used to receive the natural language description of the current operation and maintenance task input by the user and the list of target devices corresponding to the current operation and maintenance task selected by the user.

[0057] In this embodiment, the task receiving module 301 can... Figure 2The interface shown receives a natural language description of the current maintenance task and a list of target devices selected by the user. This list of target devices can include device IP addresses, login credentials, and device types.

[0058] The task parsing module 302 is used to input the natural language description of the current operation and maintenance task, the target device list, and the preset parsing prompt words into the large language model; obtain the parsing result returned by the large language model; the parsing result includes at least the target intelligent agent, task keywords, device type, and device manufacturer;

[0059] The agent invocation module 303 is used to send the parsing result to the monitoring sentinel agent and invoke the monitoring sentinel agent to execute the current operation and maintenance task when the target agent is the monitoring sentinel agent; to send the parsing result to the asset inspection agent and invoke the asset inspection agent to execute the current operation and maintenance task when the target agent is the asset inspection agent; and to receive the completion status returned by the monitoring sentinel agent or the asset inspection agent after completing the current operation and maintenance task.

[0060] The command generation module 304 is used to match the parsing results with the correspondence between multiple devices and operation and maintenance command templates stored in the knowledge base to obtain a matching operation and maintenance command template; input the parsing results, the knowledge base, and preset search-related prompts into the large language model; enable the large language model to search the knowledge base and return the target prior knowledge corresponding to the current operation and maintenance task; input the parsing results, target prior knowledge, operation and maintenance command template, and preset command selection prompts into the large language model; obtain one or more operation and maintenance commands corresponding to the current operation and maintenance task output by the large language model;

[0061] The command issuing module 305 is used to issue one or more operation and maintenance commands to the corresponding target devices to perform the corresponding operation and maintenance operations.

[0062] In this embodiment, the task parsing module 302 of the work assignment agent 110 parses the current maintenance task and target device list using a large language model to obtain the parsing result. The agent invocation module 303, when the target agent in the parsing result is the monitoring sentinel agent 120, sends the parsing result to the monitoring sentinel agent 120 to execute the current maintenance task; when the target agent is the asset inspection agent 130, the parsing result is sent to the asset inspection agent 130 to execute the current maintenance task, and the module receives the completion status returned by the monitoring sentinel agent or the asset inspection agent after completing the current maintenance task. This approach not only achieves collaborative interaction between agents but also realizes a closed-loop task flow, making the cloud platform maintenance system more complete and breaking through the limitations of fragmented functional modules in existing systems.

[0063] In some embodiments, the execution of the agent includes a three-layer memory mechanism. For example... Figure 3 As shown, in some embodiments, a three-layer memory mechanism can be implemented through a short-term memory module 306, a long-term memory module 307, and a security memory module 308.

[0064] The short-term memory module 306 caches the parsed results of the current operation and maintenance task, along with one or more generated operation and maintenance commands, as the context of the current task. This context is then deleted upon completion of the current operation and maintenance task. Specifically, it can cache the user intent, generated commands, and confirmation status from the current operation and maintenance task description. The short-term memory module 306 can dynamically cache the current task context on a session basis (per execution of one operation and maintenance task). For example, if the user's original intent is "Restart the Nginx service on the server," the generated command is "systemctl restartnginx," and the user's confirmation status is empty, the cache is valid for the entire lifecycle of a single session and is cleared upon timeout (e.g., 15 minutes by default) or task completion to avoid cross-task interference.

[0065] The long-term memory module 307 is used to store the pre-defined correspondence between multiple devices and operation and maintenance command templates in the knowledge base. This correspondence is generated in advance by the large language model after retrieving operation and maintenance product manuals, user historical operation records, commonly used devices, and high-frequency command patterns from the knowledge base. After the current operation and maintenance task is completed, the description of the current operation and maintenance task, the parsing result, and one or more generated operation and maintenance commands are stored as prior knowledge in the knowledge base.

[0066] Safety memory module 308 is used to store a database of keywords for high-risk operations.

[0067] In this embodiment, the high-risk operation keyword library is stored in a knowledge base and can be directly loaded when needed. The high-risk operation keyword library contains a blacklist of high-risk operations, such as keywords like "rm," "kill," and "shutdown." It can be used for real-time command filtering. If a user request or generated command is detected to contain a high-risk operation keyword, the system immediately interrupts the process, returns a prohibition message, and records the request in the security memory module to prevent bypassing security policies through semantics.

[0068] As can be seen, the layered agent memory management mechanism in this embodiment serves as the key knowledge support for the agent, undertaking the storage and transmission functions of context engineering. Through the three-layer memory mechanism, execution consistency and security are ensured; all processing results are structured and fed back to the knowledge base, thereby driving continuous system evolution.

[0069] like Figure 3 As shown, the work assignment agent 110 also includes a security judgment module 309 and a user confirmation module 310; the security judgment module 309 is used to match the one or more operation and maintenance operation commands with high-risk operation commands in the high-risk operation keyword library using a large language model before the command issuing module 305 issues the one or more operation and maintenance operation commands to the corresponding target device to perform the corresponding operation and maintenance operation. If a high-risk operation command is matched, a reminder and recommended safety command are output to the user.

[0070] The user confirmation module 310 is used to output one or more operation and maintenance operation commands to the user for confirmation before the command issuing module issues them to the corresponding target device to execute the corresponding operation and maintenance operation. If the user confirms, the command issuing module issues the one or more operation and maintenance operation commands to the corresponding target device to execute the corresponding operation and maintenance operation, and stores the execution result in the knowledge base. If the user refuses, the current operation and maintenance task description, the refused operation command, and the preset summary prompt are input into the large language model; the summary content output by the large language model is obtained and stored in the knowledge base; and the operation and maintenance operation command refused by the user is marked as refused in the knowledge base.

[0071] In this embodiment, the short-term memory module 306 ensures the traceability of the operation and maintenance process. The long-term memory module 307 enables dynamic updates to the knowledge base, making subsequent generated operation and maintenance commands more accurate. The security memory module 308, security judgment module 309, and user confirmation module 310 perform two filters on the generated operation and maintenance commands, further ensuring the security and reliability of the system.

[0072] In some embodiments, a specific execution flow of the work assignment agent 110 is described in [reference]. Figure 4 , Figure 4 This is a flowchart illustrating an example of a work assignment agent in a cloud platform operation and maintenance system provided in this application embodiment. It includes the following steps:

[0073] Step S401: Receive the natural language description of the current operation and maintenance task input by the user and the list of target devices corresponding to the current operation and maintenance task selected by the user.

[0074] Step S402: parse the current task description and target device list to obtain the parsing results and determine the target intelligent agent.

[0075] If the target agent is itself, proceed to step S403; if the target agent is the monitoring sentinel agent 120, proceed to step S410; if the target agent is the asset inspection agent 130, proceed to step S412.

[0076] Specifically, in this step, the natural language description of the current operation and maintenance task, the target device list, and preset parsing prompts can be input into the large language model; the parsing result returned by the large language model is then obtained. The parsing result includes at least the target agent, task keywords, and device-related information. The task keywords can be executing operation and maintenance commands, performing monitoring data analysis, or performing asset inspection. During this process, the large language model performs structured parsing of the target device list, generating connectable structured device metadata that includes device type and device manufacturer.

[0077] During user input, maintenance tasks can be described using natural language. These task descriptions need to be related to the functions of the relevant intelligent agents, and the task must be assigned to a specific agent within the description. The work assignment agent 110, after parsing the input maintenance task description and obtaining the target agent, can then automatically invoke other agents to assign specific tasks, thus achieving collaborative interaction between agents.

[0078] Specifically, this step is used to obtain detailed input for maintenance tasks and a clear list of target devices. The work assignment agent 110 provides a unified input interface for maintenance tasks corresponding to different agents. As the scheduling and task allocation center for agents, the work assignment agent 110 plays a pivotal role in the entire system.

[0079] Step S403: Generate operation and maintenance commands using the large model.

[0080] Specifically, a large language model can be used to generate one or more operation and maintenance commands for each target device in the target device list, based on the current operation and maintenance task description and the target device list.

[0081] In some embodiments, this step may include the following three specific steps:

[0082] Step 1: Based on the parsing results of step S402, match them with the correspondence between multiple devices and operation and maintenance command templates stored in the knowledge base to obtain the matching operation and maintenance command templates.

[0083] Step 2: Input the parsing results, knowledge base, and preset search-related prompts into the large language model; enable the large language model to search the knowledge base and return the target prior knowledge corresponding to the current operation and maintenance task; the target prior knowledge includes: historical operation and maintenance tasks and corresponding historical operation and maintenance commands;

[0084] Step 3: Input the parsing results, target prior knowledge, operation and maintenance command templates, and preset command selection prompts into the large language model; obtain one or more operation and maintenance commands corresponding to the current operation and maintenance task output by the large language model.

[0085] In some embodiments, subsequent task execution can be routed based on device type and vendor information through a routing mechanism. For example, if the device is of the host type and the device vendor does not have special commands, it will be directed to Linux host operation and maintenance operations, which will be matched with a preset host operation and maintenance operation template; if it is of the network type and some device vendors have special commands, then for network devices without special commands, it will be directed to general network device operation and maintenance operations, which will be matched with a preset network device operation and maintenance operation template; for network devices with special commands, it will be directed to network device operation and maintenance operations specific to that device, which will be matched with a preset network device operation and maintenance operation template specific to that device.

[0086] Specifically, this step is primarily to differentiate between different device objects to prevent failures in subsequent maintenance operations. During maintenance, different device types and manufacturers can affect the way commands are written; for example, certain commands for power checks may differ between network devices from one manufacturer and another. Users often struggle to be thoroughly familiar with commands for all device types across the entire maintenance chain, requiring significant time for learning and inevitably leading to forgetting or errors. Because large language models possess a mechanical memory capacity that humans lack, they can quickly and accurately generate corresponding maintenance operation commands given only relevant materials and experience. Therefore, this reduces the learning cost for maintenance personnel.

[0087] In addition, the correspondence between the multiple devices used in this step and the operation and maintenance command templates is pre-built and stored based on the product manuals, FAQ documents, SOP documents provided by the device manufacturers, as well as the user's operation and maintenance experience rules, while also leveraging the capabilities of large language models.

[0088] Step S404: Determine whether the operation and maintenance command is safe. If yes, proceed to step S406; otherwise, proceed to step S405.

[0089] In this step, a large language model and a high-risk operation keyword library can be used to determine whether one or more generated operation and maintenance operation commands are safe; if they are not safe, proceed to step S405; if they are safe, proceed to step S406.

[0090] Specifically, using a large language model, the one or more operation and maintenance commands are matched with high-risk operation commands in the high-risk operation keyword library. If a high-risk operation command is matched, it is unsafe, and step S405 is executed for the unsafe operation and maintenance command; if no match is found, it is safe, and step S406 is executed for the safe operation and maintenance command.

[0091] Step S405: Prohibit the execution of unsafe operation and maintenance commands.

[0092] In some embodiments, for maintenance operation commands that match high-risk operation commands, the system can obtain reminder information and recommended safety commands output by the large language model and output them to the user.

[0093] Step S406: Output secure operation and maintenance commands to the user.

[0094] Step S407: Determine whether the user confirms. If confirmed, proceed to step S408; if not confirmed, proceed to step S409.

[0095] Step S408: Execute the operation and maintenance command and record the execution result.

[0096] In this step, one or more secure operation and maintenance commands can be issued to the corresponding target devices to execute the corresponding operation and maintenance operations, and the execution results can be stored in the knowledge base. The current operation and maintenance task has been completed.

[0097] In this embodiment, operation and maintenance commands can be issued through secure channels such as SSH (Secure Shell) and API.

[0098] Step S409: Input the rejected operation and maintenance commands into the large language model for summarization.

[0099] In this step, the description of the current maintenance task, the rejected maintenance operation commands, and the preset summary prompts can be input into the large language model; the summary content output by the large language model is stored in the knowledge base; and the maintenance operation commands rejected by the user are marked as rejected in the knowledge base. The current maintenance task is then completed.

[0100] Step S410: Invoke the monitoring sentinel agent to execute the current operation and maintenance task.

[0101] Specifically, the parsing result obtained in step S402 can be sent to the monitoring sentinel agent 120, which will then execute the current maintenance task.

[0102] Step S411: Receive the completion status returned by the monitoring sentinel agent after it has completed the current task.

[0103] Step S412: Invoke the asset inspection intelligent agent to execute the current operation and maintenance task.

[0104] Specifically, the parsing result obtained in step S402 can be sent to the asset inspection intelligent agent, and the asset inspection intelligent agent 130 can be invoked to execute the current operation and maintenance task.

[0105] Step S413: Receive the completion status returned by the asset inspection intelligent agent after it has completed the current task.

[0106] In this embodiment, the work assignment agent 110 continuously tracks the execution status of tasks during the workflow. After the downstream agent completes the task, it returns the completion status to the work assignment agent 110, thus realizing a closed loop in the task workflow.

[0107] In this embodiment, the work assignment agent introduces a Human-in-the-Loop (HIL) confirmation mechanism to ensure the safety and controllability of all maintenance operations after the maintenance operation command is generated. When the agent generates a candidate maintenance operation command, the system outputs a structured confirmation prompt to the user, including the command content, expected impact scope, risk level (which can be scored based on the security memory module), and a recommendation reason, such as "Based on historical operation records, you have executed similar commands on similar devices." The user can provide feedback through explicit instructions, such as "Confirm Execution" or "Reject." For maintenance operation commands approved by the user, the execution results, including standard output, error codes, and execution time, are saved in real time for subsequent contextual understanding and behavior optimization. For maintenance operation commands rejected by the user, the system not only terminates the command process but also marks the command and its context as "Rejected" and stores it in the knowledge base to prevent the model from repeatedly generating the same high-risk suggestions due to similar contexts.

[0108] In this embodiment, the entire confirmation and execution process supports asynchronous retry, timeout circuit breaking, and operation logging. All interaction logs are archived synchronously to meet compliance requirements and build a fully traceable security control closed loop from "generation—confirmation—execution—feedback".

[0109] The monitoring sentinel agent in the cloud platform operation and maintenance system provided in this application embodiment is an agent responsible for real-time anomaly detection, event backtracking, and a unified monitoring and analysis mode. It is mainly used to understand the current monitoring task, and can perform time-series analysis of monitoring indicator time-series data for various types of assets and equipment using a large language model. Based on real-time time-series analysis, it performs anomaly analysis to obtain a comprehensive anomaly diagnosis report, enabling rapid backtracking of abnormal events and constructing a unified analysis mode for monitoring data.

[0110] See Figure 5 , Figure 5 This is a schematic diagram illustrating an example of a monitoring sentinel agent in a cloud platform operation and maintenance system provided in this application embodiment. The monitoring sentinel agent includes: a data acquisition module 501, a time-series data preprocessing module 502, a time-series analysis module 503, and an anomaly analysis module 504.

[0111] The data acquisition module 501 is used to acquire time-series data of one or more monitoring indicators of the monitored device.

[0112] In this embodiment, the monitoring sentinel agent 120 is triggered in two ways: manual triggering and automatic triggering. Manual triggering relies on on-site users to execute manually and is mainly used to deal with sudden scenarios or special events; automatic triggering generally executes observation tasks at regular intervals, such as initiating a full data screening every 30 minutes.

[0113] In this way, the monitored device can be the target device in the target device list sent by the work assignment agent 110 when it is invoked, or it can be the target device entered by the user when it is manually triggered, or it can be all the monitored devices when it is triggered on a timed basis.

[0114] like Figure 1 As shown, the data acquisition module 501 can obtain time-series data of one or more monitoring indicators of the monitored device from the time-series database 160 in the ICOP platform. The time-series data in the time-series database 160 is obtained from the device 100 by the monitoring tool 170 in the ICOP platform.

[0115] Specifically, the data acquisition module 501 uses the device's IP address or asset identifier as an index and, with the help of relevant query syntax, batches raw indicators such as CPU utilization, memory throughput, and network packet loss rate from the time-series database 160 deployed on the ICOP platform. At the same time, it needs to retain context labels such as device type and device IP address, and filter invalid data points through the data verification module to ensure the integrity and timeliness of the input data.

[0116] The time series data preprocessing module 502 is used to input time series data of different indicators of different device types into the corresponding pre-trained machine learning models respectively; each machine learning model is used to classify a set of time series data of each indicator under the device type based on the fluctuation characteristics of the time series data segments in each preset sliding window for each preset sliding window, and obtain the fluctuation type of each time series data segment; and determine whether the time series data segment is abnormal based on the fluctuation type.

[0117] The time series analysis module 503 is used to input the abnormal time series data segment, the fluctuation type of the time series data segment, the relevant equipment labels, and the preset diagnostic prompt words into the large language model for abnormal time series data segments; and obtain the diagnostic results and time series statistical results of the abnormal equipment output by the large language model.

[0118] The anomaly analysis module 504 is used to input abnormal time-series data segments, diagnostic results and time-series statistical results of abnormal equipment, a preset anomaly pattern library and anomaly analysis prompt words into the large language model; so that the large language model can perform end-to-end anomaly discrimination and semantic attribution, and output anomaly analysis results.

[0119] In this embodiment, the monitoring sentinel agent first analyzes the time-series data using a large language model through the time-series analysis module 503 to obtain the diagnostic results and time-series statistical results of the devices with abnormal indicators. Then, through the anomaly analysis module 504, it analyzes the abnormal time-series data segments, the diagnostic results of the devices with abnormal indicators, and the time-series statistical results using a large language model, performing end-to-end anomaly identification and semantic attribution, and outputting the anomaly analysis results. Compared to existing systems that can only obtain the diagnostic results of devices with abnormal indicators, and whose root causes still require manual analysis based on user experience, this embodiment utilizes a large model to achieve end-to-end anomaly identification and semantic attribution, without relying on human experience, making it more intelligent.

[0120] like Figure 5 As shown, the timing analysis module 503 in this embodiment may include: a timing analysis process selection submodule 5031, a first timing analysis submodule 5032, and a second timing analysis submodule 5033. The anomaly analysis module 504 may include a first anomaly analysis submodule 5041 and a second anomaly analysis submodule 5042.

[0121] The time series analysis process selection submodule 5031 is used to select the first time series analysis process and call the first time series analysis submodule 5032 when the fluctuation type classification results, related equipment and indicator labels indicate that only a single equipment and single indicator analysis is involved for abnormal time series data segments; and select the second time series analysis process and call the second time series analysis submodule 5033 when the fluctuation type classification results, related equipment and indicator labels indicate that multiple indicators of a single equipment are involved.

[0122] The first time series analysis submodule 5032 inputs the abnormal time series data segment, the fluctuation type and single indicator of the time series data segment, and the preset diagnostic prompt words into the large language model; and obtains the diagnostic results and time series statistics of the single indicator abnormal device output by the large language model.

[0123] The second time series analysis submodule 5033, for each indicator, inputs the abnormal time series data segment, the fluctuation type of the time series data segment, the indicator and the preset diagnostic prompt words into multiple large language models in parallel; and obtains the diagnostic results and time series statistical results of multiple single-indicator abnormal devices output by each large language model.

[0124] The first anomaly analysis submodule 5041 is used to input the abnormal time-series data segment, the diagnostic results and time-series statistical results of the single-index abnormal device output by the first time-series analysis submodule 5032, the preset anomaly pattern library and anomaly analysis prompt words into the large language model; so that the large language model can perform end-to-end anomaly discrimination and semantic attribution, and output the single-item anomaly analysis result.

[0125] The second anomaly analysis submodule 5042 is used to input the abnormal time-series data segment, the diagnostic results and time-series statistical results of the multiple indicators of the abnormal device output by the second time-series analysis submodule 5033, the preset anomaly pattern library and anomaly analysis prompt words into multiple large language models; obtain multiple individual anomaly analysis results generated by each large language model after anomaly identification; input all individual anomaly analysis results into the global analysis large language model to generate a comprehensive anomaly diagnosis report covering the overall system status.

[0126] In this embodiment, different processes are used to analyze single indicators of single devices and multiple indicators of single devices for abnormal time-series data segments, which can adapt to different abnormal situations and improve analysis efficiency.

[0127] like Figure 5As shown, the monitoring sentinel agent also includes a result output module 505. The result output module 505 is used to output the anomaly analysis results from the large language model to the user in natural language form, and to retrieve the knowledge base and output historical data and corresponding solutions, or relevant content from standard operating procedure documents, to the user.

[0128] Specifically, the system can use a hybrid retrieval method, using information such as indicators, anomaly descriptions, device types, and device IP addresses as key elements, to perform a full-text + vector hybrid retrieval with historical cases in the knowledge base, returning the Top N related entries; at the same time, all results are synchronously written to the log, realizing a closed loop from anomaly discovery to knowledge reuse.

[0129] In some embodiments, a specific execution flow of the monitoring sentinel agent 120 is described in [reference needed]. Figure 6 , Figure 6 A flowchart illustrating an example of a monitoring sentinel agent in a cloud platform operation and maintenance system provided in this application embodiment; the flowchart includes:

[0130] Step S601: Obtain time-series data of one or more monitoring indicators of the monitored device.

[0131] Step S602: Time series data preprocessing.

[0132] In this step, time series data of different device types can be input into the corresponding pre-trained machine learning models. Each machine learning model is used to classify a set of time series data of different indicators under the device type based on the fluctuation characteristics of the time series data segments in each preset sliding window to obtain the fluctuation type of each time series data segment. Based on the fluctuation type, it is determined whether the time series data segment is abnormal.

[0133] In practical applications, historical monitoring data is pre-classified and labeled using the classification method described above, and these labels are used as training samples to train multiple machine learning models for the recognition and classification of time series.

[0134] For example, key features of a machine learning model can include device-related signal thresholds, the number of spikes within the signal, spike amplitude, second derivative of fluctuations, signal variance, interquartile range, and other statistical quantities. For a set of monitoring time-series data for a specific device, a sliding window method is used to capture multiple sets of signals. Based on device and indicator labels, these signals are submitted to the corresponding machine learning model for classification. The classification results of multiple sequences are ultimately confirmed by a voting method. The fluctuation types classified in this way can include: periodic fluctuations, continuously monotonically increasing fluctuations, continuously monotonically decreasing fluctuations, occasional spikes, large fluctuations, and sudden stops, etc. Based on the device's own operating characteristics and the fluctuation type of the time-series data segment, it is possible to determine whether the time-series data segment is an abnormal time-series segment. For example, for a certain device, during its normal operation, a certain indicator should rise and fall regularly. If it is determined that its time-series data segment is continuously monotonically increasing, then this time-series data segment is determined to be an abnormal time-series data segment.

[0135] Step S603: Determine whether the number of indicator items of the abnormal device is greater than 1. If yes, proceed to step S604; otherwise, proceed to step S606.

[0136] Specifically, this step can select an appropriate subsequent routing branch processing procedure for time series analysis based on the obtained classification results and relevant device labels. If the number of indicator items for the device with abnormal indicators is greater than 1, it indicates that multiple indicators for a single device are involved, and step S604 is executed; if the number of indicator items for the device with abnormal indicators is not greater than 1, it indicates that only the analysis of a single indicator for a single device is involved, and step S606 is executed.

[0137] For example: if the device type is CPU, the indicator type is 1, and the fluctuation type is occasional spike, then the analysis process for a single device and a single indicator will be initiated; if the device type is storage, the indicator type is 2, the fluctuation type is large fluctuation, and the number of indicators for the device is greater than 1, then the analysis process for multiple indicators will be initiated.

[0138] Specifically, a pre-stored routing decision rule table defined by operations and maintenance experts can be used in conjunction with real-time classification tags to match preset branch processing flows. For single-item analysis of a single device, pre-designed LLM prompts can be used to diagnose devices with abnormal indicators and generate time-series statistical results. For multi-item analysis of a single device, a hierarchical approach is adopted, first generating a sub-task queue, then calling multiple LLM instances in parallel to process atomic indicators, and finally aggregating them to ensure that the analysis path is strictly aligned with the device characteristics.

[0139] Step S604: Perform time series analysis on each indicator item.

[0140] This step can call multiple LLM instances in parallel to diagnose devices with abnormal indicators in time series data and generate time series statistical results, obtaining the diagnostic results and time series statistical results for each indicator.

[0141] Step S605: Perform anomaly analysis on multiple indicators. Proceed to step S608.

[0142] Specifically, each indicator item is first independently analyzed using LLM to generate individual conclusions. Then, the global analysis LLM aggregates the diagnostic results of individual indicators to generate a comprehensive final diagnostic report covering the overall system status anomaly. The final diagnostic report includes the impact assessment results of multiple indicators.

[0143] In this step, for each indicator, the time series data and contextual results are first input into the large language model. Combined with the pre-built anomaly pattern library, end-to-end anomaly detection and semantic attribution are performed, and structured single-item conclusions are output.

[0144] Specifically, in a single analysis, LLM receives a visual description of a time-series segment, such as "3 spikes in the past 5 minutes, with a peak value of 98%", and safety boundary rules, and outputs conclusions in JSON (JavaScript Object Notation) format, with fields including anomaly level, confidence level, and root cause speculation.

[0145] Then, all individual conclusions are structured and input into the global analysis LLM. The LLM aggregates the diagnostic results of individual indicators and generates a comprehensive final diagnostic report of anomalies covering the overall system status.

[0146] Step S606: Perform time series analysis on a single device and a single indicator.

[0147] This step involves using pre-designed LLM prompts and a large language model to diagnose single-indicator abnormal devices and generate time-series statistical results from time-series data.

[0148] Step S607: Perform anomaly analysis on a single device and single indicator item.

[0149] This step involves inputting time-series data and contextual results into a large language model, combining it with a pre-built anomaly pattern library to perform end-to-end anomaly detection and semantic attribution, and outputting structured single-item conclusions.

[0150] Step S608: Output the analysis results.

[0151] In this step, the agent outputs the analysis results in natural language and automatically links them to similar historical cases or SOP documents in the knowledge base to help engineers make quick decisions.

[0152] Therefore, it can be seen that the monitoring sentinel agent in this embodiment can understand the monitoring intent, adapt to the monitoring needs that change according to business changes, and realize the automatic analysis and anomaly location of key indicators.

[0153] The asset inspection agent 130 in the cloud platform operation and maintenance system provided in this application embodiment is an agent that performs in-depth inspections of physical / virtual assets by executing custom scripts.

[0154] See Figure 7 , Figure 7 This is a schematic diagram of an example of an asset inspection intelligent agent in a cloud platform operation and maintenance system provided in this application embodiment; the asset inspection intelligent agent includes: an inspection strategy matching module 701, an inspection script acquisition module 702, an inspection impact judgment module 703, and an inspection script execution module 704.

[0155] The inspection strategy matching module 701 is used to obtain the device function type of the device to be inspected corresponding to the current inspection task, and obtain the inspection strategy corresponding to the device to be inspected based on the preset correspondence between the device function type and the inspection strategy.

[0156] In this embodiment, the equipment to be inspected includes the work assignment intelligent agent, or the equipment in the inspection requirements input by the user through natural language description.

[0157] In this embodiment, the pre-defined correspondence between device function types and inspection strategies can be stored in the knowledge base.

[0158] Specifically, a list of devices to be inspected can be obtained in advance from the unified asset management module 180 of the ICOP platform. This list includes metadata such as device IP, type (e.g., server, switch), device model, and responsible person, and can be categorized accordingly. Corresponding inspection strategies can then be set for each category, including inspection time, inspection process, and so on.

[0159] The inspection script acquisition module 702 is used to acquire inspection scripts for the equipment to be inspected, including: acquiring a first inspection script for the equipment to be inspected; or acquiring a second inspection script uploaded by the user based on the on-site situation; or, inputting the inspection requirements described in natural language by the user, the knowledge base, and preset inspection script generation prompts into the large language model to acquire a third inspection script generated by the large language model; and binding the acquired inspection scripts with the equipment to be inspected to form one or more inspection task units.

[0160] The inspection script acquisition module 702 acquires the inspection script corresponding to the inspection strategy. The inspection script can be acquired through the three methods described above. In some embodiments, the inspection script acquisition module further verifies and performs security scans on the second and third inspection scripts; the inspection scripts that pass the verification and security scans are bound to the target asset object to form an inspection task unit, thereby improving the security of these two types of scripts.

[0161] The inspection impact judgment module 703 is used to input the equipment to be inspected in each inspection unit, the preset asset topology knowledge graph, and the preset asset impact assessment prompts into the big language model; so that the big language model queries the upstream and downstream equipment of the equipment to be inspected in the asset topology knowledge graph, and evaluates the impact of executing the inspection script in the current inspection unit on business services based on the upstream and downstream equipment of the equipment to be inspected, and outputs recommended adjustment information for the inspection strategy when the impact reaches a preset level.

[0162] like Figure 1 As shown, in this embodiment, the preset asset topology knowledge graph can be stored in the graph database 142. The asset topology knowledge graph is established based on the self-information and connection relationships of each device that needs to be inspected in the unified asset management module 180. It uses asset devices as nodes and the connection relationships of devices as edges, and defines the relevant attributes of nodes and edges, such as device type, device model, connection method, port and other information.

[0163] In this embodiment, the asset inspection intelligence sends a query request to the graph database server based on the device to be inspected, determines the device's position in the upstream and downstream of the entire business system, and obtains other devices most directly related to the device. After clarifying information such as dependencies, it assesses the potential impact of the current inspection. For example, a large language model can be used to score the impact, and the inspection strategy can be adjusted if the score reaches a preset threshold. For instance, high-overhead scripts such as full table scans may affect the utilization of device resources; if the impact score reaches a preset threshold, the inspection order or time can be automatically adjusted to avoid affecting business services.

[0164] The inspection script execution module 704 is used to perform inspections on the equipment to be inspected based on the inspection strategy corresponding to the current operation and maintenance task or the adjusted inspection strategy, using the inspection scripts in each inspection task unit, and obtain inspection results.

[0165] Specifically, the target device can be logged into via secure channels such as SSH or API to execute the inspection script. When the amount of data obtained from the inspection script is large, further compression processing is required. This can be achieved through prompt word enhancement and the construction of custom tools, allowing the model to capture key standard outputs, error logs, and return codes in real time during execution, and to anonymize sensitive information such as passwords and keys.

[0166] In this embodiment, the inspection script is evaluated by the inspection impact judgment module 703, and the inspection strategy can be adjusted to avoid the impact of inspection on business services and ensure the normal operation of business services.

[0167] like Figure 7 As shown, the asset inspection intelligent agent in this embodiment further includes: an inspection risk assessment module 705 and an inspection result archiving module 706. The inspection risk assessment module 705 is used to input the current inspection task, inspection results, knowledge base, and preset evaluation prompts into a large language model to obtain an inspection report output by the large language model, containing inspection conclusions, risk assessments, and actionable suggestions. The inspection result archiving module 706 is used to store the current inspection task and inspection report in the knowledge base.

[0168] In this embodiment, the inspection risk assessment module 705 can input the inspection results into the large language model, and combine them with the equipment manual, historical failure cases and best practices in the knowledge base to analyze and rate the results, including: identifying the current equipment status, such as not enabling a certain optimization parameter; assessing resource bottlenecks, such as memory usage consistently >90%; and outputting structured risk ratings and optimization suggestions.

[0169] Specifically, the system uses a hybrid retrieval method, taking information such as inspection intent, inspection results, equipment type, and equipment IP address as key elements, and performing a full-text + vector hybrid retrieval with historical cases and product manuals in the knowledge base vector library. The system returns the Top N related entries as one of the sources of analysis. With the help of expert experience-based prompts and the reasoning ability of the big model, the system forms the final inspection conclusion and actionable suggestions with evidence links.

[0170] Therefore, the asset inspection intelligent agent 130 in this embodiment can understand the user's inspection purpose, generate targeted inspection scripts and execute them automatically, provide professional and detailed inspection reports, and support multiple configuration methods, focusing on in-depth health checks of assets. This improves the intelligence of asset inspection.

[0171] In some embodiments, a specific execution flow of the asset inspection intelligent agent 130 is described in [reference]. Figure 8 , Figure 8A flowchart illustrating an example of an asset inspection intelligent agent in a cloud platform operation and maintenance system provided in this application embodiment. The process includes:

[0172] Step S801: Obtain the inspection strategy corresponding to the equipment to be inspected.

[0173] In this step, the device function type of the equipment to be inspected corresponding to the current inspection task can be obtained. Based on the preset correspondence between device function type and inspection strategy, the inspection strategy corresponding to the equipment to be inspected can be obtained.

[0174] Step S802: Obtain the inspection script.

[0175] In this step, as mentioned above, the inspection script can be obtained in three ways, and the second and third inspection scripts are verified and security scanned. The inspection scripts that pass the verification and security scan are bound to the target asset object to form an inspection task unit.

[0176] Step S803: Determine if it will affect business operations. If so, proceed to step S804; otherwise, proceed to step S805.

[0177] As shown above, the large language model can be used to query the upstream and downstream devices of the device to be inspected in the asset topology knowledge graph, and the impact of executing the inspection script in the current inspection unit on business services can be evaluated based on the upstream and downstream devices of the device to be inspected. If the impact reaches a preset level, step S804 is executed.

[0178] Step S804: Adjust the inspection strategy.

[0179] Specifically, the inspection sequence or time can be automatically adjusted to avoid affecting business services.

[0180] Step S805: Execute the inspection script.

[0181] Specifically, based on the inspection strategy corresponding to the current operation and maintenance task or the adjusted inspection strategy, the inspection scripts in each inspection task unit can be used to inspect the equipment to be inspected and obtain the inspection results.

[0182] Step S806: Conduct a risk assessment based on the knowledge base.

[0183] In this step, the current inspection task, inspection results, knowledge base, and preset evaluation prompts can be input into the large language model to obtain an inspection report output by the large language model, which includes inspection conclusions, risk assessments, and actionable suggestions.

[0184] Step S807: Archive the inspection report.

[0185] In this step, the current inspection task and inspection report are stored in the knowledge base.

[0186] Therefore, the cloud platform operation and maintenance system provided in this application embodiment achieves knowledge-guided reasoning and decision-making by constructing a knowledge base covering all operation and maintenance scenarios and deeply integrating it with a large model. Each monitoring, inspection, and handling result is structured and fed back to the knowledge base, forming a closed loop of "perception—analysis—execution—accumulation," solving the pain points of existing solutions such as "scattered knowledge, lack of reusability, and severe cold start," and enabling the system to have a self-learning ability that becomes increasingly intelligent with use. Moreover, each intelligent agent can understand natural language intent, generate secure commands, and execute low-value tasks. On-site users only need to confirm key steps and control quality risks, greatly reducing the effort invested in repetitive daily tasks and overcoming the drawbacks of traditional tools' passive response and heavy reliance on human labor. In addition, each intelligent agent has a security verification and secondary approval mechanism to ensure that all automated operations are traceable and interceptable, meeting the compliance requirements of the operation and maintenance site.

[0187] In summary, the technical solution provided in this application is superior to existing solutions in terms of intelligence, closed-loop performance, and security.

[0188] In some embodiments, the cloud platform operation and maintenance system can also achieve intelligent alarm diagnosis and intelligent prediction of resource requirements. For details, see [link to documentation]. Figure 9 , Figure 9 This is a schematic diagram of another example of a cloud platform operation and maintenance system provided in the embodiments of this application.

[0189] like Figure 9 As shown, the cloud platform operation and maintenance system includes, in addition to a work assignment intelligent agent 110, a monitoring sentinel intelligent agent 120, and an asset inspection intelligent agent 130, an alarm diagnosis intelligent agent 190 and a resource demand prediction intelligent agent 200. The alarm diagnosis intelligent agent 190 is used to parse and diagnose received raw alarm data based on an alarm diagnosis knowledge base and an asset database, using a large language model to obtain alarm diagnosis results. The alarm diagnosis knowledge base includes at least a historical fault case database and a troubleshooting work thought chain template. The alarm diagnosis results include at least alarm root cause analysis results. The resource demand prediction intelligent agent 200 is used to process resource allocation-related indicators of the target business system using a resource prediction model to obtain the amount of resources that need to be allocated to the target business system for a preset resource allocation. The resource prediction model is a multiple regression model constructed using historical business system resource allocation data. Figure 9 As shown, the alarm diagnosis knowledge base can be stored in the knowledge base 141, and the asset database can be stored in the unified asset management module 180.

[0190] In some embodiments, the alarm diagnosis agent 190 performs intelligent alarm diagnosis through the following steps:

[0191] Step 1: Receive the raw alarm data output by the cloud platform's alarm system;

[0192] like Figure 9 As shown, the alarm diagnosis agent 190 can receive raw alarm data generated by the monitoring tool 170 of the ICOP platform through abnormal indicators during the monitoring process.

[0193] Step 2: Based on a preset alarm diagnosis knowledge base, a preset asset database, and preset parsing prompts, the original alarm data is parsed using a large language model to obtain parsing results. The alarm diagnosis knowledge base includes at least: different alarm scenarios and their corresponding different troubleshooting work chain templates, a historical fault case database, and maintenance documents for each device in the cloud platform. The historical fault case database includes at least: the faulty device for each historical case, the alarm scenario of the faulty device, and the root cause of the alarm. The parsing results include at least: the target device that triggered the alarm and the device information of the target device, obtained based on the first search result of the preset asset database, and the target troubleshooting work chain corresponding to the alarm scenario of the target device, obtained based on the second search result of the alarm diagnosis knowledge base.

[0194] Step 3: Based on the original alarm data, the second search results, the parsing results, and the preset diagnostic prompts, use the large language model to perform alarm diagnosis and obtain the alarm diagnosis results of each target device output by the large language model. The alarm diagnosis results shall include at least the alarm root cause analysis results.

[0195] In some embodiments, the resource demand prediction agent 200 achieves intelligent prediction of resource demand through the following steps:

[0196] First, a resource prediction model is constructed, including: for each sample business system among multiple sample business systems of different types, obtaining the indicator values ​​of each indicator in a first indicator set at a first historical moment, and the resource size of the preset resources allocated at the first historical moment; wherein, the first indicator set includes: indicators representing the attribute information of the business system, and / or, indicators representing the operating status of the business system; using each indicator in the first indicator set as the independent variable and the preset resources as the dependent variable, a multiple regression model is constructed using the indicator values ​​of each indicator in the first indicator set at the first historical moment and the corresponding resource size, to obtain the resource prediction model;

[0197] Then, the resource prediction model is used to process the indicator values ​​of each indicator in the first indicator set for the target business system to obtain the amount of resources that need to be allocated to the target business system for the preset resources.

[0198] like Figure 9As shown in this embodiment, the indicator values ​​of each indicator in the indicator set required by the resource demand prediction agent 200 can be obtained from the resource information of each asset device in the unified asset management module 180 of the ICOP platform.

[0199] The cloud platform operation and maintenance system in this embodiment further integrates alarm diagnosis intelligence agent 190 and resource demand prediction intelligence agent 200, further realizing intelligent alarm diagnosis and intelligent resource demand prediction, and further improving the integration and intelligence of the cloud platform operation and maintenance system.

[0200] Corresponding to the aforementioned cloud platform operation and maintenance system, this application embodiment also provides a cloud platform operation and maintenance method, the method comprising:

[0201] The task assignment agent receives a natural language description of the current maintenance task input by the user and a list of target devices corresponding to the current maintenance task selected by the user; obtains the target agent assigned in the current maintenance task description; if the target agent is itself, it uses a large language model to generate one or more maintenance operation commands for each target device in the target device list based on the current maintenance task description and the target device list; and sends the one or more maintenance operation commands to the corresponding devices for execution.

[0202] When the target intelligent agent is the monitoring sentinel intelligent agent, the monitoring sentinel intelligent agent is invoked to execute the operation and maintenance task; the monitoring sentinel intelligent agent obtains one or more monitoring indicator data of the monitored device, uses a large language model to perform data analysis on the monitoring data of each monitored device, and generates analysis results in natural language form; the monitored device is the device in the operation and maintenance device list, or the monitored device specified by the user.

[0203] When the target intelligent agent is an asset inspection intelligent agent, the asset inspection intelligent agent is invoked to execute the operation and maintenance task; so that the asset inspection intelligent agent obtains the inspection script corresponding to each device to be inspected; based on each inspection script, the device to be inspected is inspected; the device to be inspected is the device in the operation and maintenance equipment list, or the device in the inspection requirements input by the user through natural language description.

[0204] For details, see Figure 10 , Figure 10 A flowchart illustrating an example of a cloud platform operation and maintenance method provided in this application embodiment; the process includes:

[0205] Step S1001: Receive the natural language description of the current operation and maintenance task input by the user and the list of target devices corresponding to the current operation and maintenance task selected by the user.

[0206] Step S1002: parse the current task description and target device list to obtain the parsing results and determine the target intelligent agent.

[0207] If the target agent is itself, execute steps S1003 and S1004; if the target agent is the monitoring sentinel agent 120, execute step S1005; if the target agent is the asset inspection agent 130, execute step S1006.

[0208] Step S1003: Using a large language model, based on the current operation and maintenance task description and the target device list, generate one or more operation and maintenance operation commands for each target device in the target device list.

[0209] Step S1004: Send one or more secure operation and maintenance commands to the corresponding target devices to execute the corresponding operation and maintenance operations, and store the execution results in the knowledge base. This completes the operation and maintenance task.

[0210] Step S1005: Invoke the monitoring sentinel agent to execute the current operation and maintenance task.

[0211] Step S1006: Invoke the asset inspection intelligent agent to execute the current operation and maintenance task.

[0212] This application provides a cloud platform operation and maintenance method that corresponds to the aforementioned cloud platform operation and maintenance system, and can achieve the corresponding technical effects.

[0213] This application also provides an electronic device, such as... Figure 11 As shown, it includes a processor 1101, a communication interface 1102, a memory 1103, and a communication bus 1104, wherein the processor 1101, the communication interface 1102, and the memory 1103 communicate with each other through the communication bus 1104.

[0214] Memory 1103 is used to store computer programs;

[0215] When processor 1101 executes the program stored in memory 1103, it performs the following steps:

[0216] The work assignment agent receives a natural language description of the current maintenance task input by the user, a list of target devices corresponding to the current maintenance task selected by the user, and a target agent to execute the current maintenance task. If the target agent is itself, it uses a large language model to generate one or more maintenance operation commands for each target device in the target device list based on the current maintenance task description and the target device list. The one or more maintenance operation commands are then sent to the corresponding devices for execution.

[0217] When the target intelligent agent is the monitoring sentinel intelligent agent, the monitoring sentinel intelligent agent is invoked to execute the operation and maintenance task; the monitoring sentinel intelligent agent, based on the current operation and maintenance task description, obtains one or more monitoring indicator data of the monitored device, uses a large language model to perform data analysis on the monitoring data of each monitored device, and generates analysis results in natural language form; the monitored device is a device in the operation and maintenance device list, or a monitored device specified by the user;

[0218] When the target intelligent agent is an asset inspection intelligent agent, the asset inspection intelligent agent is invoked to execute the maintenance task; the asset inspection intelligent agent is used to obtain the inspection script corresponding to each device to be inspected based on the current maintenance task description and the target device list; and to inspect the devices to be inspected based on each inspection script; the devices to be inspected are the devices in the maintenance device list, or the devices in the inspection requirements input by the user through natural language description.

[0219] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0220] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0221] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0222] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0223] In another embodiment provided in this application, a computer-readable storage medium is also provided, which stores a computer program that, when executed by a processor, implements the steps of any of the cloud platform operation and maintenance methods described above.

[0224] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute any of the cloud platform operation and maintenance methods described in the above embodiments.

[0225] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk (SSD)).

[0226] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0227] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0228] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application are included within the scope of protection of this application.

Claims

1. A cloud platform operation and maintenance system, characterized in that, include: Work assignment agent, monitoring sentinel agent, and asset inspection agent; The work assignment agent is used to receive a natural language description of the current operation and maintenance task input by the user and a list of target devices corresponding to the current operation and maintenance task selected by the user. Obtain the target agent assigned in the current operation and maintenance task description; When the target intelligent agent is itself, the large language model is used to generate one or more operation and maintenance operation commands for each target device in the target device list based on the current operation and maintenance task description and the target device list. Send the one or more operation and maintenance commands to the corresponding devices for execution; If the target agent is the monitoring sentinel agent, the monitoring sentinel agent is invoked to execute the current operation and maintenance task; If the target intelligent agent is an asset inspection intelligent agent, the asset inspection intelligent agent is invoked to execute the current operation and maintenance task. The monitoring sentinel agent is used to acquire one or more monitoring indicator data of the monitored devices, and to perform data analysis on the monitoring data of each monitored device using a large language model to generate analysis results in natural language form; the monitored devices are the devices in the operation and maintenance equipment list, or the monitored devices specified by the user. The asset inspection intelligent agent is used to obtain the inspection script corresponding to each device to be inspected. Based on each inspection script, the equipment to be inspected is inspected. The equipment to be inspected is either the equipment in the maintenance equipment list or the equipment in the inspection requirements input by the user through natural language description.

2. The cloud platform operation and maintenance system according to claim 1, characterized in that, The work assignment agent includes: a task receiving module, a task parsing module, an agent invocation module, a command generation module, and a command issuing module; The task receiving module is used to receive the natural language description of the current operation and maintenance task input by the user and the list of target devices corresponding to the current operation and maintenance task selected by the user. The task parsing module is used to input the natural language description of the current operation and maintenance task, the target device list, and the preset parsing prompts into the large language model; obtain the parsing result returned by the large language model; the parsing result includes at least the target agent, task keywords, device type, and device manufacturer; The agent invocation module is used to send the parsing result to the monitoring sentinel agent and invoke the monitoring sentinel agent to execute the current operation and maintenance task when the target agent is the monitoring sentinel agent; and to send the parsing result to the asset inspection agent and invoke the asset inspection agent to execute the current operation and maintenance task when the target agent is the asset inspection agent; and to receive the completion status returned by the monitoring sentinel agent or the asset inspection agent after completing the current operation and maintenance task. The command generation module is used to match the parsing results with the correspondence between multiple devices and operation and maintenance command templates stored in the knowledge base to obtain a matching operation and maintenance command template; input the parsing results, the knowledge base, and preset search-related prompts into the large language model; enable the large language model to search the knowledge base and return the target prior knowledge corresponding to the current operation and maintenance task; the target prior knowledge includes: historical operation and maintenance tasks and corresponding historical operation and maintenance commands; input the parsing results, target prior knowledge, operation and maintenance command templates, and preset command selection prompts into the large language model; obtain one or more operation and maintenance commands corresponding to the current operation and maintenance task output by the large language model; The command issuing module is used to issue one or more operation and maintenance commands to the corresponding target devices to perform the corresponding operation and maintenance operations.

3. The cloud platform operation and maintenance system according to claim 2, characterized in that, The work assignment agent also includes: a short-term memory module and a long-term memory module; The short-term memory module is used to cache the parsing results of the current operation and maintenance task and one or more generated operation and maintenance commands as the context of the current operation and maintenance task; and to delete the context after the current operation and maintenance task is completed. The long-term memory module is used to store the pre-defined correspondence between multiple devices and operation and maintenance command templates in a knowledge base; and after the current operation and maintenance task is completed, it stores the description of the current operation and maintenance task, the parsing result, and one or more generated operation and maintenance commands as prior knowledge in the knowledge base; the correspondence between the multiple devices and operation and maintenance command templates is generated and stored in advance based on the product manuals, frequently asked questions documents, standard operating procedure documents provided by the equipment manufacturers, and the user's operation and maintenance experience rules.

4. The cloud platform operation and maintenance system according to claim 3, characterized in that, The work assignment intelligent agent also includes: a security memory module and a security judgment module; The safety memory module is used to store a database of keywords for high-risk operations; The security judgment module is used to match the one or more operation and maintenance operation commands with high-risk operation commands in the high-risk operation keyword library using a large language model before the command issuance module issues the one or more operation and maintenance operation commands to the corresponding target device to perform the corresponding operation and maintenance operation. If a high-risk operation command is matched, the module outputs a reminder and recommends safe commands to the user.

5. The cloud platform operation and maintenance system according to claim 2, characterized in that, The work assignment intelligent agent also includes: a user confirmation module; The user confirmation module is used to output one or more operation and maintenance operation commands to the user for confirmation before the command issuing module issues them to the corresponding target device for execution. If the user confirms, the command issuing module issues the one or more operation and maintenance operation commands to the corresponding target device for execution and stores the execution result in the knowledge base. If the user refuses, the current operation and maintenance task description, the refused operation command, and preset summary prompts are input into the large language model. The summary content output by the large language model is obtained and stored in the knowledge base. The refused operation and maintenance operation command is marked as refused in the knowledge base.

6. The cloud platform operation and maintenance system according to claim 1, characterized in that, The monitoring sentinel intelligent agent includes: a data acquisition module, a time series data preprocessing module, a time series analysis module, an anomaly analysis module, and a result output module; The data acquisition module is used to acquire time-series data of one or more monitoring indicators of the monitored device; The time series data preprocessing module is used to input time series data of different device and indicator types into the corresponding pre-trained machine learning models respectively; each machine learning model is used to classify a set of time series data of each indicator under the device type based on the fluctuation characteristics of the time series data segments in each preset sliding window for each preset sliding window, and obtain the fluctuation type of each time series data segment; and determine whether the time series data segment is abnormal based on the fluctuation type. The time series analysis module is used to input the abnormal time series data segment, the fluctuation type of the time series data segment, the relevant equipment tags, and the preset diagnostic prompt words into the big language model for abnormal time series data segments; and obtain the diagnostic results and time series statistical results of the abnormal equipment output by the big language model. The anomaly analysis module is used to input abnormal time-series data segments, diagnostic results and time-series statistical results of abnormal equipment, a preset anomaly pattern library and anomaly analysis prompt words into the large language model; enabling the large language model to perform end-to-end anomaly discrimination and semantic attribution, and output anomaly analysis results. The result output module is used to output the anomaly analysis results output by the large language model to the user in natural language form, and to retrieve the knowledge base and output the historical data and corresponding solutions or relevant content from the standard operating procedure document to the user.

7. The cloud platform operation and maintenance system according to claim 6, characterized in that, The time series analysis module includes: a time series analysis process selection submodule, a first time series analysis submodule, and a second time series analysis submodule; The time series analysis process selection submodule is used to select the first time series analysis process and call the first time series analysis submodule when the fluctuation type classification results, related equipment and indicator labels indicate that only a single equipment and single indicator analysis is involved for abnormal time series data segments; and select the second time series analysis process and call the second time series analysis submodule when the fluctuation type classification results, related equipment and indicator labels indicate that multiple indicators of a single equipment are involved. The first time series analysis submodule inputs the abnormal time series data segment, the fluctuation type and individual indicators of the time series data segment, and the preset diagnostic prompt words into the big language model; and obtains the diagnostic results and time series statistics of the abnormal devices for individual indicators output by the big language model. The second time series analysis submodule, for each indicator, inputs the abnormal time series data segment, the fluctuation type of the time series data segment, the indicator and the preset diagnostic prompt words into multiple large language models in parallel; and obtains the diagnostic results and time series statistical results of multiple single-indicator abnormal devices output by each large language model. The anomaly analysis module includes a first anomaly analysis submodule and a second anomaly analysis submodule; The first anomaly analysis submodule is used to input the abnormal time-series data segment, the diagnostic results and time-series statistical results of the single-index abnormal device output by the first time-series analysis submodule, the preset anomaly pattern library and anomaly analysis prompt words into the large language model; so that the large language model can perform end-to-end anomaly discrimination and semantic attribution, and output the single-item anomaly analysis result; The second anomaly analysis submodule is used to input the abnormal time-series data segment, the diagnostic results and time-series statistical results of the multiple indicators of the abnormal device output by the second time-series analysis submodule, the preset anomaly pattern library and anomaly analysis prompt words into multiple large language models; obtain multiple individual anomaly analysis results generated by each large language model after anomaly identification; input all individual anomaly analysis results into the global analysis large language model to generate a comprehensive anomaly diagnosis report covering the overall system status.

8. The cloud platform operation and maintenance system according to claim 1, characterized in that, The asset inspection intelligent agent includes: an inspection strategy matching module, an inspection script acquisition module, an inspection impact judgment module, an inspection script execution module, an inspection risk assessment module, and an inspection result archiving module. The inspection strategy matching module is used to obtain the device function type of the device to be inspected corresponding to the current inspection task, and obtain the inspection strategy corresponding to the device to be inspected based on the preset correspondence between device function type and inspection strategy. The inspection script acquisition module is used to acquire inspection scripts for the equipment to be inspected, including: acquiring a first inspection script for the equipment to be inspected; or acquiring a second inspection script uploaded by the user based on the on-site situation; or, inputting the inspection requirements described in natural language by the user, the knowledge base, and the preset inspection script generation prompts into the large language model to acquire a third inspection script generated by the large language model; and binding the acquired inspection scripts with the equipment to be inspected to form one or more inspection task units. The inspection impact judgment module is used to input the equipment to be inspected in each inspection unit, the preset asset topology knowledge graph, and the preset asset impact assessment prompts into the big language model; so that the big language model queries the upstream and downstream equipment of the equipment to be inspected in the asset topology knowledge graph, and evaluates the impact of executing the inspection script in the current inspection unit on business services based on the upstream and downstream equipment of the equipment to be inspected, and outputs recommended adjustment information for the inspection strategy when the impact reaches the preset level; The inspection script execution module is used to inspect the equipment to be inspected based on the inspection strategy corresponding to the current operation and maintenance task or the adjusted inspection strategy, and to obtain the inspection results by using the inspection scripts in each inspection task unit. The inspection risk assessment module is used to input the current inspection task, inspection results, knowledge base and preset evaluation prompts into the big language model to obtain an inspection report output by the big language model, which includes inspection conclusions, risk assessment and actionable suggestions. The inspection result archiving module is used to store the current inspection task and inspection report in the knowledge base.

9. The cloud platform operation and maintenance system according to claim 8, characterized in that, The inspection script acquisition module further verifies and performs security scans on the second and third inspection scripts; the inspection scripts that pass the verification and security scans are bound to the target asset object to form an inspection task unit.

10. The cloud platform operation and maintenance system according to claim 1, characterized in that, It also includes human-computer interaction interfaces; The task assignment agent receives a natural language description of the maintenance task input by the user and a list of target devices corresponding to the maintenance task selected by the user through the human-computer interaction interface; the monitoring sentinel agent receives monitoring instructions triggered by the user and the specified monitored devices through the human-computer interaction interface. The asset inspection intelligent agent receives inspection requests input by the user in natural language through the human-computer interaction interface.

11. The cloud platform operation and maintenance system according to claim 1, characterized in that, Also includes: Alarm diagnostic agent and resource demand prediction agent; The alarm diagnosis intelligent agent is used to parse and diagnose the received raw alarm data based on the alarm diagnosis knowledge base and asset database, and to obtain alarm diagnosis results using a large language model; wherein, the alarm diagnosis knowledge base includes at least: a historical fault case database and a troubleshooting work mind chain template; the alarm diagnosis results include at least: alarm root cause analysis results; The resource demand prediction agent is used to process resource allocation-related indicators of the target business system using a resource prediction model to obtain the amount of resources that need to be allocated to the target business system for a preset resource requirement; wherein, the resource prediction model is a multiple regression model constructed using historical resource allocation data of the business system.

12. A cloud platform operation and maintenance method, characterized in that, Applied to cloud platform operation and maintenance system, the cloud platform operation and maintenance system includes: work assignment intelligent agent, monitoring sentinel intelligent agent and asset inspection intelligent agent; The method includes: The task assignment agent receives a natural language description of the current maintenance task input by the user and a list of target devices corresponding to the current maintenance task selected by the user; obtains the target agent assigned in the current maintenance task description; if the target agent is itself, it uses a large language model to generate one or more maintenance operation commands for each target device in the target device list based on the current maintenance task description and the target device list; and sends the one or more maintenance operation commands to the corresponding devices for execution. When the target intelligent agent is the monitoring sentinel intelligent agent, the monitoring sentinel intelligent agent is invoked to execute the operation and maintenance task; the monitoring sentinel intelligent agent obtains one or more monitoring indicator data of the monitored device, uses a large language model to perform data analysis on the monitoring data of each monitored device, and generates analysis results in natural language form; the monitored device is the device in the operation and maintenance device list, or the monitored device specified by the user. When the target intelligent agent is an asset inspection intelligent agent, the asset inspection intelligent agent is invoked to execute the operation and maintenance task; so that the asset inspection intelligent agent obtains the inspection script corresponding to each device to be inspected; based on each inspection script, the device to be inspected is inspected; the device to be inspected is the device in the operation and maintenance equipment list, or the device in the inspection requirements input by the user through natural language description.

13. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the steps of the method of claim 12.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method described in claim 12.

Images