Network operation and maintenance system and method

Abstract

The present application provides a network operation and maintenance system comprising a plurality of intelligent agents and a method. Different intelligent agents are used for completing different work, for example, a first intelligent agent actively issues a network operation and maintenance task by means of acquired first operation and maintenance information, and a second intelligent agent determines, on the basis of the network operation and maintenance task, a first scheme used for completing the network operation and maintenance task. The first intelligent agent can autonomously issue a network operation and maintenance task on the basis of acquired operation and maintenance information such as network data, without the need for operation and maintenance personnel to instruct network operation and maintenance work. Moreover, the second intelligent agent is driven by a language model, the second intelligent agent can provide a reasonable and accurate first scheme to complete the network operation and maintenance task, without relying on network operation and maintenance suggestions given by operation and maintenance personnel. Thus, autonomous collaboration among a plurality of intelligent agents in the network operation and maintenance system enables network operation and maintenance tasks to be well completed, without relying on participation of operation and maintenance personnel; the present application shows great application value in network operation and maintenance scenarios.

Description

Network operation and maintenance system and method

[0001] This application claims priority to the Chinese patent applications filed with the State Intellectual Property Office on February 19, 2024, with application number 202410185730.7 and invention name “A method for repairing network failures and related systems” and filed with the State Intellectual Property Office on March 28, 2024, with application number 202410372568.X and invention name “A network operation and maintenance system and method”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present application relates to network operation and maintenance scenarios, and in particular to a network operation and maintenance system and method. Background Art

[0003] In recent years, the rapid development of Large Language Models (LLMs) has driven the application of artificial intelligence (AI) across various industries and professional systems. In network troubleshooting scenarios, when a network failure occurs, human experts are notified through notifications from the network operations management system (NOMS). They then proactively interact with an O&M assistant based on the NMS via text, allowing the assistant to assist the human expert in analyzing the failure. An O&M assistant is an AI-powered assistant capable of operating one or more NMSs and, to a certain extent, completing tasks instructed by humans through text-based dialogue.

[0004] The O&M assistant's operational support capabilities are limited to providing troubleshooting suggestions within a preset scope. Human experts are still required to confirm the successful diagnosis of the fault, the correctness of the fault analysis, and the effectiveness of the fault repair method. Human experts are still required to perform all fault repair operations to verify the successful repair. When encountering unknown faults, the O&M assistant lacks the proactive reasoning capabilities to resolve the faults, and its fault scenario generalization is limited. When the O&M assistant's analysis is incorrect or the repair solution is ineffective, human experts are required to manually reanalyze and provide a repair solution, resulting in limited fault autonomy. The O&M assistant's knowledge and skills in network O&M and troubleshooting are limited, and it cannot fully understand and grasp all network scenarios and fault events.

[0005] Summary of the Invention

[0006] The embodiments of the present application provide a network operation and maintenance system and method, in which multiple intelligent entities in the network operation and maintenance system can complete network operation and maintenance tasks well through autonomous collaboration without relying on the participation of operation and maintenance personnel or technical personnel. The system has a high degree of automation and intelligence and has great application value in network operation and maintenance scenarios.

[0007] In the first aspect, an embodiment of the present application provides a network operation and maintenance system. The network operation and maintenance system includes a first intelligent agent and a second intelligent agent. The first intelligent agent is used to obtain first operation and maintenance information. For example, the first intelligent agent obtains the first operation and maintenance information from the network operation and maintenance system. The operation and maintenance information includes but is not limited to equipment fault information, connection topology between equipment, information in equipment manuals, experience summarized by operation and maintenance personnel, etc. Afterwards, the first intelligent agent is used to publish network operation and maintenance tasks based on the first operation and maintenance information. The network operation and maintenance tasks include but are not limited to network fault repair tasks and network line opening tasks. After the second intelligent agent reads the network operation and maintenance task, it determines a first solution for completing the network operation and maintenance task based on the network operation and maintenance task, and publishes the first solution. The first solution includes an implementation method for completing the network operation and maintenance task, and may also include determining the analysis steps of the first solution and the causal relationship between the steps. The second intelligent agent is driven by a language model, so that the second intelligent agent has strong learning and thinking capabilities, so that it can autonomously provide a reasonable and accurate first solution.

[0008] In this embodiment, a network operation and maintenance system including multiple intelligent agents is provided, and different intelligent agents are used to complete different tasks. For example, the first intelligent agent actively publishes a network operation and maintenance task by obtaining the first operation and maintenance information, and the second intelligent agent determines the first solution for completing the network operation and maintenance task based on the network operation and maintenance task. In the embodiment of the present application, the first intelligent agent can autonomously publish the network operation and maintenance task based on the operation and maintenance information such as the obtained network data, without relying on operation and maintenance personnel or technicians to instruct the network operation and maintenance work. In addition, the second intelligent agent is driven by a language model, which is a deep neural network trained based on knowledge and experience in telecommunications and various network fields. Therefore, the second intelligent agent can give a reasonable and accurate first solution to complete the network operation and maintenance task, and no longer relies on the network operation and maintenance suggestions given by operation and maintenance personnel or technicians. In summary, the network operation and maintenance system provided by the embodiment of the present application can complete the network operation and maintenance task well through autonomous collaboration between multiple intelligent agents, without relying on the participation of operation and maintenance personnel or technicians. The degree of automation and intelligence is high, and it has good application value in network operation and maintenance scenarios.

[0009] In some possible implementations, the network operation and maintenance system also includes a third agent. This third agent is responsible for evaluating the first solution and issuing an evaluation conclusion. The evaluation conclusion specifically includes a conclusion on whether the first solution is feasible. The evaluation conclusion may also include other information, such as the evaluation process and the reasons for the evaluation conclusion. In this implementation, the participation of the third agent allows the first solution provided by the second agent to be evaluated, which helps ensure its feasibility and safety.

[0010] In some possible implementations, the first solution includes analysis steps for determining the first solution. The third agent is specifically configured to evaluate the first solution based on the analysis steps. In other words, the third agent can perform logical reasoning based on the analysis steps used by the second agent to determine the first solution, thereby reaching an evaluation conclusion. The analysis steps herein can specifically be the multiple steps of deliberation and logical analysis performed by the second agent in determining the first solution, each of which has a causal relationship. The third agent can then effectively determine the rationality of the first solution by performing logical reasoning based on the analysis steps provided by the second agent.

[0011] In some possible implementations, the third agent is further configured to publish a first request message based on the first solution. The second agent is further configured to publish analysis steps for determining the first solution based on the first request message. The third agent is specifically configured to evaluate the first solution based on the analysis steps. In other words, if the second agent only publishes the first solution but does not publish the analysis steps for determining the first solution, the third agent may also request the second agent to publish the analysis steps for determining the first solution, thereby enabling the third agent to effectively determine whether the first solution is reasonable by performing logical reasoning based on the analysis steps provided by the second agent.

[0012] In some possible implementations, the third agent is further configured to issue a second request message based on the first solution. The first agent is further configured to obtain second operation and maintenance information based on the second request message and to issue the second operation and maintenance information. The third agent is specifically configured to evaluate the first solution based on the second operation and maintenance information. In other words, if the third agent requires more information to assist in evaluating the first solution, the third agent may request the first agent to provide the second operation and maintenance information, thereby improving the accuracy of the evaluation.

[0013] In some possible implementations, if the evaluation concludes that the first solution is unfeasible, the second agent is further configured to determine a second solution for completing the network operation and maintenance task based on the network operation and maintenance task, and publish the second solution. In other words, upon reading the third agent's evaluation conclusion that the first solution is unfeasible, the second agent will negate its previous analytical approach to determining the first solution. It can then refer to the third agent's evaluation logic and recommendations and determine a second solution based on a new approach, thereby achieving solution optimization.

[0014] In some possible implementations, if the evaluation concludes that the first solution is not feasible, the first agent is further configured to obtain and publish third-party operation and maintenance information. The second agent is specifically configured to determine a second solution based on the third-party operation and maintenance information and the network operation and maintenance tasks. Specifically, to facilitate the second agent's determination of a new second solution, the first agent may obtain the third-party operation and maintenance information from the network operation and maintenance management system and provide the third-party operation and maintenance information to the second agent. The second agent then integrates the third-party operation and maintenance information with the network operation and maintenance tasks to determine a more secure and effective second solution.

[0015] In some possible implementations, the second agent is further configured to issue a third request message based on the network operation and maintenance task. The first agent is configured to obtain fourth operation and maintenance information based on the third request message. The second agent is specifically configured to determine the first solution based on the network operation and maintenance task and the fourth operation and maintenance information. In other words, if the second agent requires more information to assist in analyzing and determining the first solution, the second agent may request the first agent to provide the fourth operation and maintenance information, thereby helping the second agent analyze and determine a more secure and effective first solution.

[0016] In some possible implementations, the second agent is specifically configured to invoke a network operation and maintenance tool based on the network operation and maintenance task to determine the first solution, thereby enriching the methods used by the second agent to determine the first solution. For example, in the case of a network fault repair task, the second agent can send instructions to the network operation and maintenance management system to invoke tools such as a ping command or fault knowledge search within the network operation and maintenance management system to help determine the first solution for resolving the network fault.

[0017] In some possible implementations, the first agent and the second agent operate in a first network operation and management system, or in a system independent of the first network operation and management system, or in different network operation and maintenance management systems. This provides multiple possible operating environments for multiple agents, expanding the application scenarios of this solution.

[0018] In some possible implementations, the interaction between the first agent and the second agent is presented through a conversation in a channel, making it easier to present the conversations between the agents to the operation and maintenance personnel or technicians. This design, which makes the network operation and maintenance task process fully transparent and intervention-accessible, can help the operation and maintenance personnel or technicians increase their trust in the multi-agent network operation and maintenance system. The dual protection of the agents and the operation and maintenance personnel or technicians can improve the reliability of network operation and maintenance. For example, the information and conversation content exchanged between agents is mainly in natural language text and understandable structured expressions (such as data tables and device configuration command lines).

[0019] In some possible implementations, the language model is a large language model (LLM). Specifically, the large language model in the embodiments of the present application is a deep neural network trained based on knowledge and experience in telecommunications and various network fields, such as IP network knowledge, optical network knowledge, mobile bearer network knowledge, wireless local area network knowledge, and virtual network knowledge. Thus, driving intelligent agents with a large language model can better enable multiple intelligent agents to collaboratively complete network operation and maintenance tasks.

[0020] In some possible implementations, the first agent and / or the second agent are used to evaluate the first solution and publish the evaluation conclusion. In other words, each agent can have multiple responsibilities and functions, and each agent must perform corresponding operations based on its own responsibilities. If an agent has multiple responsibilities and functions, it can flexibly switch roles and perform different operations based on actual scenario needs. This is equivalent to improving the intelligence of individual agents, which is conducive to improving the ability of multiple agents to collaborate and complete network operation and maintenance tasks.

[0021] In a second aspect, embodiments of the present application provide a network operation and maintenance method, which is applied to a network operation and maintenance system. The network operation and maintenance system includes a first agent and a second agent. The network operation and maintenance method includes the following steps: The first agent obtains first operation and maintenance information and issues a network operation and maintenance task based on the first operation and maintenance information. The second agent determines a first solution for completing the network operation and maintenance task based on the network operation and maintenance task and issues the first solution. The second agent is driven by a language model.

[0022] In some possible implementations, the network operation and maintenance system further includes a third agent. The method further includes: evaluating the first solution by the third agent and publishing an evaluation conclusion.

[0023] In some possible implementations, the first solution includes an analysis step for determining the first solution. Evaluating the first solution by the third agent includes: evaluating the first solution by the third agent according to the analysis step.

[0024] In some possible implementations, the method further includes: issuing, by a third agent, a first request message based on the first solution; and issuing, by a second agent, an analysis step for determining the first solution based on the first request message. Evaluating the first solution by the third agent includes: evaluating the first solution by the third agent based on the analysis step.

[0025] In some possible implementations, the method further includes: issuing a second request message according to the first solution via a third agent; obtaining second operation and maintenance information according to the second request message via the first agent, and issuing the second operation and maintenance information. Evaluating the first solution via the third agent includes: evaluating the first solution according to the second operation and maintenance information via the third agent.

[0026] In some possible implementations, if the evaluation conclusion indicates that the first solution is not feasible, the method further includes: determining, by a second intelligent agent, a second solution for completing the network operation and maintenance task based on the network operation and maintenance task, and publishing the second solution.

[0027] In some possible implementations, if the evaluation conclusion indicates that the first solution is not feasible, the method further includes: obtaining, by the first agent, third operation and maintenance information, and publishing the third operation and maintenance information. Determining, by the second agent, a second solution for completing the network operation and maintenance task based on the network operation and maintenance task includes: determining, by the second agent, the second solution based on the third operation and maintenance information and the network operation and maintenance task.

[0028] In some possible implementations, the method further includes: issuing, by the second agent, a third request message based on the network operation and maintenance task; and obtaining, by the first agent, fourth operation and maintenance information based on the third request message. Determining, by the second agent, a first solution for completing the network operation and maintenance task based on the network operation and maintenance task includes: determining, by the second agent, the first solution based on the network operation and maintenance task and the fourth operation and maintenance information.

[0029] In some possible implementations, determining, by the second agent, a first solution for completing the network operation and maintenance task based on the network operation and maintenance task includes: determining, by the second agent, the first solution by calling a network operation and maintenance tool based on the network operation and maintenance task.

[0030] In some possible implementations, the first intelligent agent and the second intelligent agent run in a first network operation and management system, or the first intelligent agent and the second intelligent agent run in a system independent of the first network operation and management system, or the first intelligent agent and the second intelligent agent run in different network operation and maintenance management systems.

[0031] In some possible implementations, the interaction between the first agent and the second agent is presented through a conversation in a channel.

[0032] In some possible implementations, the language model is a large language model.

[0033] In some possible implementations, the method further includes: evaluating the first solution and publishing an evaluation conclusion through the first agent, and / or evaluating the first solution and publishing an evaluation conclusion through the second agent.

[0034] The embodiment of the present application provides a network operation and maintenance system including multiple intelligent agents, in which different intelligent agents are used to complete different tasks. For example, the first intelligent agent actively publishes a network operation and maintenance task by obtaining the first operation and maintenance information, and the second intelligent agent determines the first solution for completing the network operation and maintenance task based on the network operation and maintenance task. In the embodiment of the present application, the first intelligent agent can autonomously publish the network operation and maintenance task based on the operation and maintenance information such as the obtained network data, without relying on operation and maintenance personnel or technicians to instruct the network operation and maintenance work. In addition, the second intelligent agent is driven by a language model, which is a deep neural network trained based on knowledge and experience in telecommunications and various network fields. Therefore, the second intelligent agent can give a reasonable and accurate first solution to complete the network operation and maintenance task, and no longer relies on the network operation and maintenance suggestions given by operation and maintenance personnel or technicians. In summary, the network operation and maintenance system provided by the embodiment of the present application can complete the network operation and maintenance task well through autonomous collaboration between multiple intelligent agents, without relying on the participation of operation and maintenance personnel or technicians. The degree of automation and intelligence is high, and it has good application value in network operation and maintenance scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] FIG1 is a schematic diagram of a scenario in which multiple agents implement network operation and maintenance in an embodiment of the present application;

[0036] FIG2 is a schematic diagram of an implementation method in which multiple intelligent agents collaborate to complete network operation and maintenance tasks in an embodiment of the present application;

[0037] FIG3 is a schematic diagram of an implementation method of an intelligent agent operating a network operation management system according to an embodiment of the present application;

[0038] FIG4 is a schematic diagram of another implementation method in which multiple agents collaborate to complete network operation and maintenance tasks in an embodiment of the present application;

[0039] FIG5 is a schematic diagram of a first operating environment of multiple agents in an embodiment of the present application;

[0040] FIG6 is a schematic diagram of a second operating environment of multiple agents in an embodiment of the present application;

[0041] FIG7 is a schematic diagram of a third operating environment of multiple agents in an embodiment of the present application;

[0042] FIG8 is a schematic diagram of a fourth operating environment of multiple agents in an embodiment of the present application;

[0043] FIG9 is a schematic diagram of an implementation method in which multiple agents collaborate to resolve network failures in an embodiment of the present application;

[0044] FIG10( a ) is a schematic diagram of a conversation channel in a process in which multiple agents collaborate to resolve a network failure in an embodiment of the present application;

[0045] FIG10( b ) is a schematic diagram of another conversation channel in the process of multiple agents collaborating to solve a network failure in an embodiment of the present application;

[0046] FIG10( c ) is a schematic diagram of another conversation channel in the process of multiple agents collaborating to solve a network failure in an embodiment of the present application;

[0047] FIG11 is a schematic diagram of another implementation method of multiple intelligent agents collaborating to solve network failures in an embodiment of the present application. DETAILED DESCRIPTION

[0048] The embodiments of the present application provide a network operation and maintenance system and method, in which multiple intelligent entities in the network operation and maintenance system can complete network operation and maintenance tasks well through autonomous collaboration without relying on the participation of operation and maintenance personnel or technical personnel. The system has a high degree of automation and intelligence and has great application value in network operation and maintenance scenarios.

[0049] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the descriptions used in this way can be interchangeable where appropriate so that the embodiments can be implemented in a sequence other than that illustrated or described in this application. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or modules is not necessarily limited to those steps or modules clearly listed, but may include other steps or modules that are not clearly listed or that are inherent to these processes, methods, products or devices. The naming or numbering of the steps in the embodiments of the present application does not mean that the steps in the method flow must be executed in the time / logical sequence indicated by the naming or numbering. The process steps that have been named or numbered can change the execution order according to the technical purpose to be achieved, as long as the same or similar technical effects can be achieved. The division of units that appears in the embodiments of the present application is a logical division. In actual applications, there may be other division methods. For example, multiple units can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between each other shown or discussed can be through some interfaces, and the indirect coupling or communication connection between units can be electrical or other similar forms, which are not limited in the embodiments of the present application. Moreover, the units or sub-units described as separate components may or may not be physically separated, may or may not be physical units, or may be distributed into multiple circuit units. Some or all of the units can be selected according to actual needs to achieve the purpose of the present application scheme.

[0050] To facilitate understanding, some technical terms involved in the embodiments of this application are first introduced below.

[0051] (1) Network Operation Management System

[0052] A network operations management system refers to computer software that has network operations and maintenance-related functions, such as network management, device control, and operations maintenance. A network operations management system is used to manage and control network equipment and related facilities. It can collect network-related data, modify the configuration of network equipment or network functions, and has the ability to diagnose and repair network faults. One or more network operations management systems may exist in a network environment. Some of the intelligent agents in the embodiments of this application can communicate with and operate each network operations management system through the instruction executor as described above.

[0053] (2) Large language model (LLM)

[0054] A large language model refers to a deep learning model trained using a large amount of text data, which can generate natural language text or understand the meaning of language text. Large language models can handle a variety of natural language tasks, such as text classification, question-answering, dialogue, etc., and are an important path to artificial intelligence. Specifically, the large language model in the embodiment of the present application refers to a deep neural network whose input and output are both natural language text or structured expressions, which can simulate human understanding, thinking, behavior and expression, and is based on knowledge and experience in telecommunications and various network fields. For example, this type of knowledge includes IP network knowledge, optical network knowledge, mobile bearer network knowledge, wireless local area network knowledge and virtual network knowledge.

[0055] (3) Agent or AI Agent

[0056] An intelligent agent is an intelligent entity capable of simulating the network operation and maintenance tasks of an operator or technician. Multiple agents can interact and collaborate to complete network operation and maintenance tasks. Some agents can use the network operation and management system to perceive and understand network status or access auxiliary tools to perform troubleshooting operations by calling the network operation and management system's application programming interface (API). When processing and calculating relevant data, developing and proposing troubleshooting solutions, and other steps, an agent's autonomous thinking process and conclusions can be shared in a conversational channel, making them accessible to other agents. This allows them to understand the logic behind their thinking and maintains transparency to operators or technicians. Agents can learn new operation and maintenance knowledge through their large-scale language models and use this knowledge to reason and decompose new or complex network operation and maintenance tasks they have never encountered before. Agents can be divided into multiple types based on their responsibilities. Certain types of agents possess enhanced thinking and operational capabilities in specific domains and tasks, enabling them to complete tasks that other agents cannot.

[0057] (4) Autoregressive Large Language Model

[0058] The autoregressive large language model enables intelligent agents to think while simultaneously utilizing other tools to assist in their thinking. By combining toolformer technology with the autoregressive large language model and a command executor, the intelligent agent can invoke APIs of relevant tools in the network operations management system through output commands during the thinking and output process. This allows the agent to run related tools (such as scientific calculators and ping commands) to ensure the accuracy of key outputs.

[0059] (5) Non-autoregressive large language models

[0060] Non-autoregressive large language models have stronger context-sensitive capabilities and can be combined with the feature representation and output of downstream task models and large language models to expand the capabilities of specific agents in specific tasks. For example, agents based on this type of large language model can more accurately identify regulatory opinions and determine whether the current step complies with regulatory opinions by combining with sentiment analysis task models.

[0061] The present invention provides a network operation and maintenance system comprising multiple agents. These agents are used to complete operation and maintenance tasks based on network data. Each agent's responsibilities include, but are not limited to, acquiring network data, analyzing network status, generating fault repair methods, implementing authorized and capable repair operations, and verifying the repair status. Taking the network operation and maintenance scenario of fault repair as an example, multiple agents, through division of labor and collaboration, can determine whether the fault analysis conclusions are correct and whether the fault repair methods are effective and safe when attempting to repair a network fault. They can also autonomously verify whether the fault has been repaired after executing authorized and capable repair operations. The collaboration of multiple agents can solve problems that a single agent cannot. For example, a single agent's subjective understanding of a fault repair solution when repairing a network fault may lead to a single agent's subjective understanding of an incorrect fault repair solution, but due to this subjectivity, it cannot detect the error in the solution. Another example is that a single agent's professional capabilities are biased, and its intelligence is limited, making it unable to solve problems beyond its capabilities. Another example is that a single agent can only operate a single network operation and management system, while diagnosing a particular network fault requires information from multiple network operation and management systems.

[0062] Figure 1 is a schematic diagram of a scenario in which multiple agents implement network operation and maintenance in an embodiment of the present application. As shown in Figure 1, at least one of the multiple agents can interact with the network operation and management system, and multiple agents can collaborate to complete network operation and maintenance tasks through interaction. Specifically, multiple agents communicate, discuss and communicate through information transmission and dialogue, and complete network operation and maintenance tasks, such as repairing network faults, without the intervention of operation and maintenance personnel or technical personnel through collaboration and relay. Among them, the information and dialogue content are expressed in natural language text or understandable structured expressions (such as data tables, device configuration command lines). In actual applications, the number of agents can be flexibly adjusted according to application scenarios and needs. More agents can enhance the overall capabilities of the agent cluster, and fewer agents can reduce unnecessary computing overhead.

[0063] Interactions between multiple agents can be achieved through dialogue. For example, multiple agents interact within a single dialogue channel, each sending text or structured expressions within the channel. For example, each agent can communicate its own thoughts and conclusions regarding fault analysis to other agents through the dialogue channel. Each agent uses the information within the dialogue channel to fulfill its own network operation and maintenance responsibilities. For example, each agent can diagnose data provided by other agents and verify the repair solutions proposed by other agents. The order in which agents speak is determined by each agent and is not subject to rules. For example, each agent can determine whether to perform its own responsibilities and respond to the content of the dialogue channel based on information such as text or expressions in the dialogue channel. Messages sent by any agent can be replied to by one or more agents, meaning that the interaction between multiple agents in the dialogue channel resembles a "group chat" model. The dialogue channel can also be referred to as a chat window, where information such as natural language text, data symbols, and visual images posted by each agent can be presented. Agents will interact repeatedly until a consensus is reached, thus completing the operation and maintenance task. When multiple agents deviate from a task or a repair plan fails, they interact again until the task is successfully completed. Alternatively, for impossible network maintenance tasks, such as equipment requiring physical repair, multiple agents will present information such as the task, process, and conclusions to maintenance personnel or technicians through a conversation channel to quickly resolve network maintenance issues. Multiple agents can reach consensus and advance network maintenance tasks based on a one-vote veto rule. Disagreements can be resolved and their approval achieved through spontaneous communication and discussion. For example, multiple agents can cross-verify to ensure the correctness of each step in the troubleshooting process, avoiding unsafe, unreliable, or ineffective troubleshooting operations.

[0064] This dialogue channel serves as a means of information transmission between agents and can also be supervised by operations personnel or technicians. Taking the network fault repair scenario as an example, operations personnel or technicians with corresponding permissions can pause, terminate, or modify ongoing fault analysis or fault repair operations at any stage in the fault diagnosis or repair process. Operations personnel or technicians can also provide processing suggestions or instructions for the current stage at any stage, such as analyzing data, analyzing faults, providing repair plans, confirming repair plans, or verifying repair status. The operations or processing of operations personnel or technicians are not necessary steps for multiple agents to repair faults, but rather a mechanism to ensure that multiple agents complete operation and maintenance tasks. This design that makes the fault repair process fully transparent and intervention-friendly can help operations personnel or technicians increase their trust in the multi-agent network operation and maintenance system. The dual protection of agents and operations personnel or technicians can improve the reliability of network operation and maintenance.

[0065] Some or all of the multiple agents are driven by a language model, which can specifically be the large language model shown in Figure 1. Among them, agents that are not driven by a language model can also participate in the interaction between multiple agents. For example, such agents can be driven by non-intelligent software or tools. The embodiment of the present application does not limit the number of large language models. The method of associating large language models with agents can be flexibly selected according to the type of large language models. The types of large language models include but are not limited to autoregressive large language models and non-autoregressive large language models.

[0066] As an example, a large language model is associated one-to-one with an agent. Each agent is based on a dedicated large language model, which drives only that single agent. This large language model can be trained specifically for the responsibilities of its associated agent. For example, the large language model associated with the operation agent can be fine-tuned for the task of generating network operation management system operation instructions. This improves the accuracy of the large language model's output of network operation management system operation instructions and avoids grammatical errors.

[0067] As another example, a large language model is associated with agents in a one-to-many relationship. Each agent is based on a large language model, and multiple agents share the same large language model, that is, the same large language model drives multiple agents. In scenarios where computing resources are limited, multiple agents can share the same large language model and use methods such as prompt engineering (Prompt Engineering, Prompting) to assist the large language model in fulfilling different responsibilities and completing different tasks. When sharing a large language model, multiple agents usually have certain common responsibilities and capabilities. For example, the analysis agent and the review agent that can share a large language model both have a lot of network fault repair knowledge.

[0068] The embodiments of the present application do not limit the number of network operation and management systems, and the method of associating the network operation and management systems with the intelligent agents can be flexibly selected according to the actual application scenario.

[0069] As an example, a network operation management system is associated one-to-one with an agent. A single agent capable of operating a network operation management system only operates that one network operation management system, and each network operation management system is operated by only one agent. This type of agent can operate the corresponding software by calling the network operation management system's API or other software tool instructions.

[0070] As another example, a network operations management system (NOMS) has a many-to-one association with an agent. A single agent capable of operating a NMS can operate multiple NMSs, while the same NMS is operated by only one agent. This type of agent can specifically call upon the APIs and software tools of multiple NMSs to operate the corresponding software.

[0071] As another example, a network operations management system (NOMS) is associated with multiple agents in a one-to-many relationship. Multiple agents capable of operating the NMS operate the same NMS, with multiple agents operating a single NMS. These multiple agents fulfill different roles and collectively operate the single NMS. Agents with different responsibilities will possess stronger and more specialized capabilities in their respective areas.

[0072] The following describes a specific implementation method for multiple agents to collaboratively complete network operation and maintenance tasks. The embodiments of this application should include at least a first agent and a second agent, and may further include a third agent and other agents. The number of agents depends on actual needs and is not limited here. Agents with the same responsibilities can be added to enhance the credibility of the agent cluster, agents with different responsibilities can be split or merged, and other types of agents with capabilities that can assist in network operation and maintenance can be added. In practical applications, agents are typically differentiated based on their responsibilities and functions, meaning that different agents typically have different responsibilities and functions. For example, the first agent is an operation agent, the second agent is an analysis agent, and the third agent is a review agent. Of course, in some possible scenarios, each agent can also have multiple responsibilities and functions, for example, each agent can have the responsibilities and functions of an operation agent, an analysis agent, and a review agent. In other words, each agent must perform corresponding operations within its own scope of responsibility. If an agent has multiple responsibilities and functions, it can flexibly switch roles and perform different operations based on the actual needs of the scenario.

[0073] Here we first introduce the respective functions of the operation agent, analysis agent and review agent in the network operation and maintenance scenario.

[0074] 1. Operation Agents. Compared to other agents, operation agents can operate one or more network operations management systems by outputting operational commands and communicate with the network operations management system, operating system, and other software programs using a command executor. Operation agents can understand natural language text sent by other agents and convert it into API commands for invoking and operating the network operations management system. Operation agents are also used to receive new fault notifications from the network operations management system and query and receive network-related data provided by the network operations management system. Operation agents can read and understand text messages from other agents and operations and maintenance personnel or technicians in the conversation channel. They can independently determine whether these text messages contain tasks requiring their own action. They can then convey the understood content and tasks to be completed through feature representation and output to their own command executor, which then completes the relevant tasks, such as communicating with the network operations management system through API calls. Operation agents can also send data and information transmitted by the network operations management system into the conversation channel using natural language text or structured representations such as lists for other agents to access and understand. The responsibilities of the operating agent are not limited to data query and troubleshooting operations related to network failures. It can understand and process text information and instructions for various network operation and maintenance tasks, and perform various queries and other operations.

[0075] 2. Analytical Agents. Fine-tuned and trained using a large number of fault operation and maintenance cases and historical experience, these agents possess enhanced fault analysis and diagnosis capabilities. They can identify possible root causes and reasoning logic, and propose remediation methods and steps. The analytical agent can utilize tools within the network operations management system (e.g., ping commands, path planning, or fault knowledge search) in conjunction with a command executor and toolformer technology to assist in fault analysis. For example, the analytical agent can analyze and diagnose network faults based on the network status and fault information provided by the operational agent, and then provide remediation solutions based on the network faults. The analytical agent can read and understand text messages from other agents and operations personnel or technicians in the conversation channel. It can independently determine whether these messages contain tasks requiring its own action and can send reasoning processes, thinking results, and remediation solutions to the conversation channel in natural language text and symbols. When additional information is needed for judgment and analysis, the analytical agent will request additional information from the operational agent. If the remediation plan involves physical repairs, the analytical agent will seek assistance from the physical equipment operation and maintenance team; if the fault cannot be repaired, the analytical agent will seek assistance from the operations personnel or technicians. The responsibilities of the analysis agent extend beyond analyzing network fault information. It can analyze the steps and operations required to complete any network operation and maintenance task, and transmit the analysis process, including thought processes and conclusions, to the conversation channel. Toolformer technology can be used to determine the current network status or other tools to assist in completing operation and maintenance tasks.

[0076] 3. Review Agents: These review agents possess a stronger understanding of network operation and maintenance security and extensive knowledge of fault causal relationships. They can evaluate the root causes and remediation solutions proposed by other agents, audit the rigor and accuracy of analysis and reasoning steps, and assess the safety and impact of remediation operations. The review agent can combine command executors and toolformer technologies to test the effectiveness of remediation solutions within the network operation and management system's simulation environment, or through other tools and methods. The review agent can combine white-box logical decomposition and black-box simulation to evaluate and provide feedback on the troubleshooting conclusions proposed by other agents, ensuring safe and effective remediation. For example, the review agent can assess the accuracy, safety, and effectiveness of the fault diagnosis information and remediation solutions provided by the analysis agent. The review agent can read and understand text messages from other agents and operations personnel or technicians in the conversation channel, autonomously determine whether these messages contain tasks requiring its own action, and send its review process, results, and reasons to the analysis agent via natural language text to inform the analysis agent of the feasibility of the diagnosis and solution. The review agent can leverage toolformer technology and its built-in command executor to invoke tools such as path planning to aid logical reasoning and evidence verification. When additional information is needed for judgment and analysis, the review agent will request more information from the operation agent. The review agent's responsibilities extend beyond evaluating network fault repair solutions; it can also assess the correctness, security, and effectiveness of operational procedures in other network operation and maintenance tasks.

[0077] For the sake of ease of introduction, the following first introduces an implementation method of multiple agents collaboratively completing network operation and maintenance tasks, taking the first agent as an operating agent, the second agent as an analyzing agent, and the third agent as a reviewing agent as an example. In some possible scenarios, the agent first completes the corresponding operation based on the current role. After the current link of the network operation and maintenance scenario ends and enters the next link, the agent can also switch to other roles to complete the corresponding operation. The specific method can be flexibly adjusted according to the actual network operation and maintenance scenario. For example, the first agent can also be an analyzing agent and / or a reviewing agent, the second agent can also be a reviewing agent, and so on, that is, the first agent is a collection of an operating agent and an analyzing agent, or the first agent is a collection of an operating agent and a reviewing agent, and the second agent is a collection of an analyzing agent and a reviewing agent.

[0078] In scenarios where multiple agents collaborate to complete network operation and maintenance tasks, for agents driven by a large language model, the large language model is like the "brain" of the agent. The agent implements specific functions based on its own responsibilities and through the large language model. Specifically, the agent can build in some corresponding statements based on its own responsibilities. These statements are used to inform the large language model of the agent's responsibilities and the functions to be implemented. The built-in statements of agents with different responsibilities are also different. For example, the built-in statements of the operation agent include but are not limited to "You are the role responsible for operation", the built-in statements of the analysis agent include but are not limited to "You are the role responsible for analysis", and the built-in statements of the review agent include but are not limited to "You are the role responsible for evaluation". An agent can copy the information posted by other agents in the conversation channel, and splice the copied information with the built-in statements to send to the large language model. The large language model can implement corresponding functions based on the information transmitted by the intelligent agent. Taking the review intelligent agent as an example, the large language model can know that its current responsibility is analysis based on the sentence "You are the role responsible for analysis" transmitted by the review intelligent agent, and know the specific content to be analyzed based on the information of the dialogue channel transmitted by the review intelligent agent. The solution obtained by the large language model after analysis will be fed back to the analysis intelligent agent, and the analysis intelligent agent will publish it on the dialogue channel to realize interaction with other intelligent agents.

[0079] The association method between the intelligent agent and the large language model introduced above can be that the intelligent agent sends a request to the large language model based on its own responsibilities, and the large language model implements the corresponding function according to the request. Among them, the content of the request includes but is not limited to the sentences built into the intelligent agent and the sentences spliced ​​together by the information in the dialogue channel as described in the example above. The information transmission between the intelligent agent and the large language model can be implemented based on the HTTPS protocol or the JSON data model, etc., and the specifics are not limited here. Whether the large language model is associated with the intelligent agent one-to-one, or the large language model is associated with the intelligent agent one-to-many, each intelligent agent can be associated with the corresponding large language model using the method introduced here. In other words, for the scenario where the large language model is associated with the intelligent agent one-to-many, the large language model is not limited to a certain responsibility and function. The large language model can flexibly switch responsibilities and functions to be implemented according to the requests sent by different intelligent agents. The specific association method of each intelligent agent and the large language model will be introduced below in combination with the responsibilities and functions of each intelligent agent in the network operation and maintenance scenario in the embodiment of the present application.

[0080] Figure 2 is a schematic diagram of an implementation method of multiple agents collaborating to complete network operation and maintenance tasks in an embodiment of the present application. As shown in Figure 2, this implementation method includes the following steps.

[0081] 101. The first intelligent agent obtains operation and maintenance information 1 from the network operation and management system.

[0082] In the embodiments of the present application, all information or data related to network operation and maintenance fall within the scope of operation and maintenance information. For example, operation and maintenance information includes but is not limited to equipment fault information, connection topology between equipment, information in equipment manuals, and experience summarized by operation and maintenance personnel. The embodiments of the present application do not limit the specific way for the first intelligent agent to obtain operation and maintenance information. One example is that the first intelligent agent obtains operation and maintenance information from the network operation and maintenance management system. The first intelligent agent can also directly obtain operation and maintenance information from outside the system through the input and output interface. In one possible scenario, the network operation and maintenance management system can actively send operation and maintenance information 1 to the first intelligent agent. For example, if the network operation and maintenance management system finds that a device has failed, it will actively send the fault information to the first intelligent agent. In another possible scenario, the first intelligent agent can also regularly query the network operation and maintenance management system for operation and maintenance information, and the network operation and maintenance management system will feed back the operation and maintenance information to the first intelligent agent.

[0083] 102. The first intelligent agent publishes a network operation and maintenance task.

[0084] The first agent determines a network operation and maintenance task based on the acquired operation and maintenance information 1 and publishes the network operation and maintenance task in the dialogue channel. The presentation format of the information published by the first agent includes, but is not limited to, natural language text, data symbols, and visual images, and network operation and maintenance tasks include, but are not limited to, network fault repair tasks and network line commissioning tasks. As an example, the first agent is driven by a large language model. The first agent sends the operation and maintenance information 1 to the large language model and informs the large language model of the functions to be implemented. For example, the first agent sends the following request to the large language model: "Please determine the network operation and maintenance task based on the operation and maintenance information 1." As another example, the first agent is not driven by a large language model. The first agent can generate a network operation and maintenance task based on a pre-set program and in combination with the operation and maintenance information 1. For example, the first agent generates a network operation and maintenance task by extracting keywords from the operation and maintenance information 1.

[0085] In some possible scenarios, the first agent may also publish the acquired operation and maintenance information 1 to the dialogue channel. Alternatively, the first agent may also pre-analyze and filter the acquired operation and maintenance information 1, and publish the filtered information to the dialogue channel. The specific filtering conditions are associated with the network operation and maintenance tasks. For example, the operation and maintenance task published by the first agent is to solve network congestion, but the operation and maintenance information 1 does not reflect that the network equipment has failed. In this case, the first agent may exclude information related to the network equipment failure, and filter out information related to traffic and publish it to the dialogue channel. As an example, the first agent may send a relevant request to the large language model to request the large language model to pre-analyze and filter the operation and maintenance information 1, and the large language model will send the filtered information to the first agent.

[0086] 103. The second intelligent agent determines the first plan based on the network operation and maintenance task and publishes the first plan.

[0087] The second agent can read the network operation and maintenance task published by the first agent from the dialogue channel, and analyze the network operation and maintenance task to obtain a first solution for completing the network operation and maintenance task. The second agent publishes the first solution in the dialogue channel. The first solution should specifically include the implementation method for completing the network operation and maintenance task. In addition, the first solution may also include determining the analysis steps of the first solution and the causal relationship between the steps, etc., which are not limited here. As an example, the second agent is driven by a large language model. The second agent sends a relevant request to the large language model to request the large language model to analyze the network operation and maintenance task to obtain a solution for completing the network operation and maintenance task. The large language model sends the first solution obtained by analysis to the second agent.

[0088] In one possible scenario, the second agent needs more information to assist in analysis and determine the first solution. The second agent publishes a request message 1 on the dialogue channel to request the first agent to provide more operation and maintenance information. For example, the large language model associated with the second agent discovers through analysis that more operation and maintenance information is needed, and then notifies the second agent to publish the request message 1. The first agent reads the request message 1 from the dialogue channel and obtains operation and maintenance information 2 from the network operation and maintenance management system. Operation and maintenance information 2 may include information not included in operation and maintenance information 1. The first agent publishes operation and maintenance information 2 on the dialogue channel, and the second agent reads operation and maintenance information 2 from the dialogue channel and determines the first solution based on operation and maintenance information 2 and the network operation and maintenance task.

[0089] In another possible scenario, the second agent can also invoke a network operation and maintenance tool based on the network operation and maintenance task to help determine the first solution. For example, a large language model associated with the second agent may analyze and discover the need to invoke a network operation and maintenance tool to assist in analysis, thereby instructing the second agent to issue an invocation of the tool. For example, if the network operation and maintenance task is a network fault repair task, the second agent can send a command to the network operation and maintenance management system to invoke tools such as the ping command or fault knowledge search within the network operation and maintenance management system to help determine the first solution for resolving the network fault.

[0090] 104. The third agent reads the network operation and maintenance task and the first plan, evaluates the first plan and publishes evaluation conclusion 1.

[0091] The third agent reads the network operation and maintenance task published by the first agent and the first solution published by the second agent from the dialogue channel, and evaluates the first solution based on the network operation and maintenance task. Furthermore, the third agent publishes the evaluation conclusion 1 in the dialogue channel, wherein the evaluation conclusion 1 specifically includes the conclusion on whether the first solution is feasible. In addition, the evaluation conclusion 1 may also include other information such as the evaluation ideas and the reasons for determining the evaluation conclusion, which are not limited here. As an example, the third agent is driven by a large language model, and the third agent sends a relevant request to the large language model to request the large language model to evaluate the first solution based on the network operation and maintenance task. The large language model sends the evaluation conclusion 1 obtained by the evaluation to the third agent.

[0092] In one possible scenario, the third agent performs logical reasoning based on the analytical steps used by the second agent to determine the first solution, thereby arriving at Evaluation Conclusion 1. Specifically, the analytical steps can be the multiple steps of thought and logical analysis performed by the second agent in determining the first solution, each of which has a causal relationship. By performing logical reasoning based on the analytical steps provided by the second agent, the third agent can effectively determine whether the first solution is reasonable. For example, if the first solution published by the second agent already includes the analytical steps used to determine the first solution, the third agent can directly perform logical reasoning based on the analytical steps to arrive at Evaluation Conclusion 1. For another example, if the first solution published by the second agent does not include the analytical steps used to determine the first solution, the third agent publishes Request Message 2 on the conversation channel, requesting the second agent to provide the analytical steps used to determine the first solution. The second agent reads Request Message 2 from the conversation channel and publishes the analytical steps used to determine the first solution on the conversation channel. Furthermore, the third agent reads the analytical steps used to determine the first solution from the conversation channel and performs logical reasoning based on the analytical steps to arrive at Evaluation Conclusion 1. For example, the third agent sends the analysis steps of the first solution to the associated large language model, and the large language model performs logical reasoning based on the analysis steps to obtain evaluation conclusion 1 and sends it to the third agent.

[0093] In another possible scenario, a third agent simulates the first solution and determines Evaluation Conclusion 1 based on the simulation results. For example, the third agent sends instructions to the network operations management system to test the first solution in the simulation environment within the network operations management system. If the simulation results indicate that the first solution is feasible, Evaluation Conclusion 1 is issued. Otherwise, Evaluation Conclusion 1 is issued indicating that the first solution is not feasible. For example, a large language model associated with the third agent instructs the third agent to simulate the first solution. The third agent sends the simulation results to the large language model, which then draws Evaluation Conclusion 1 based on the simulation results and sends it to the third agent.

[0094] In another possible scenario, the third agent requires more information to help evaluate the first solution. The third agent publishes a request message 3 on the conversation channel, requesting the first agent to provide more operational information. For example, the large language model associated with the third agent discovers that more operational information is needed to help evaluate the first solution, and then notifies the second agent to publish the request message 3. The first agent reads the request message 3 from the conversation channel and obtains the operational information 3 from the network operation and management system. The first agent publishes the operational information 3 on the conversation channel, and the third agent reads the operational information 3 from the conversation channel and evaluates the first solution based on the operational information 3.

[0095] 105. The first intelligent agent reads the first solution, evaluates the first solution and publishes evaluation conclusion 2.

[0096] Optionally, in some possible scenarios, the first agent can also perform the functions of the third agent. Specifically, the first agent reads the first solution from the conversation channel and evaluates it based on the network operation and maintenance task. Furthermore, the first agent publishes evaluation conclusion 2 on the conversation channel. The specific method for the first agent to evaluate the first solution is similar to that implemented by the third agent. For details, please refer to the description of step 104 and will not be repeated here.

[0097] 106. The second agent evaluates the first solution and publishes evaluation conclusion 3.

[0098] Optionally, in some possible scenarios, the second agent can also perform the functions of the third agent. Specifically, the second agent evaluates the first solution based on the network operation and maintenance task. Furthermore, the second agent publishes evaluation conclusion 3 in the conversation channel. The specific method for the second agent to evaluate the first solution is similar to that of the third agent. For details, please refer to the description of step 104 and will not be repeated here.

[0099] 107. Multiple intelligent agents determine whether the first solution is feasible based on the evaluation conclusion.

[0100] Specifically, if the first solution is determined to be feasible, the first agent executes steps 108 and 109. For example, if only the third agent publishes Evaluation Conclusion 1 in the conversation channel, and Evaluation Conclusion 1 indicates that the first solution is feasible, the first agent executes steps 108 and 109. For another example, if, in addition to Evaluation Conclusion 1 published by the third agent in the conversation channel, the first agent also publishes Evaluation Conclusion 2, and / or the second agent also publishes Evaluation Conclusion 3, in this case, if Evaluation Conclusions 1, 2, and 3 all indicate that the first solution is feasible, the first agent executes steps 108 and 109. Conversely, if even one evaluation conclusion indicates that the first solution is not feasible, multiple agents must re-discuss and determine a new feasible solution. The evaluation conclusions published by each agent in the conversation channel are accessible to other agents, and depending on the different conclusions regarding the feasibility or infeasibility of the first solution, each agent will perform different subsequent actions based on the conclusions.

[0101] 108. The first agent reads the first solution from the dialogue channel.

[0102] Each agent can obtain information from the conversation channel by actively consulting the channel or by having the channel forward the received information to each agent. For example, the first agent may actively consult the conversation channel, conclude that the first solution is feasible, and then read the first solution. Alternatively, the conversation channel may send the conclusion that the first solution is feasible and the first solution to the first agent.

[0103] 109. The first intelligent entity operates the network operation management system to execute the first plan.

[0104] Specifically, the first agent operates the network operation management system according to the first solution to perform network operation and maintenance tasks. For example, the natural language text conversion of the first solution output by the first agent can call the instructions of the network operation management system API or other tools, and the executor outputs instructions to the network operation management system to operate the network operation management system to perform network operation and maintenance tasks according to the first solution. As an example, the first agent is driven by a large language model. The first agent can send a relevant request to the large language model to request the large language model to convert the natural language text of the first solution into instructions, and the large language model sends the converted instructions to the executor. Among them, if the first solution includes multiple execution steps, the large language model can also convert each step into a corresponding instruction. As another example, the first agent is not driven by a large language model, and the executor of the first agent can convert the natural language text of the first solution into instructions based on a set program. For example, the executor of the first agent generates instructions by extracting keywords from the first solution.

[0105] When communicating via an API, the first agent's operations are processed by the network operations management system's security authentication module and instruction distribution module, and then implemented by the corresponding functional modules within the network operations management system. For example, the security authentication module first authenticates the instructions output by the actuator to ensure their security. Passing security authentication, the instruction distribution module then distributes them to different functional modules (such as the management module, control module, and analysis module) for execution.

[0106] FIG3 is a schematic diagram of an embodiment of an intelligent agent operating a network operation and management system in an embodiment of the present application. As shown in FIG3 , an executor is a software program that can convert natural language text, data symbols, and other content into computer instructions. An executor can also be referred to as an interpreter. An executor can be considered a part of an intelligent agent, and each intelligent agent can configure its own executor. For example, a first intelligent agent operates a network operation and management system through an executor to perform network operation and maintenance tasks. For another example, a second intelligent agent uses an executor to call a network operation and maintenance tool in the network operation and management system to help determine a first solution. For another example, a third intelligent agent uses an executor to call a simulation environment in the network operation and management system to evaluate the first solution.

[0107] In some possible scenarios, after the first agent executes the first solution, multiple agents can collaborate to evaluate the effectiveness of the first solution. In other possible scenarios, if the first solution is deemed unfeasible, multiple agents will need to re-discuss and negotiate a new solution to complete the network operation and maintenance task. These possible scenarios are further described below with reference to specific embodiments.

[0108] Figure 4 is a schematic diagram of another implementation method of multiple agents collaborating to complete network operation and maintenance tasks in an embodiment of the present application. As shown in Figure 4, this implementation method includes the following steps.

[0109] 110. The first intelligent agent obtains operation and maintenance information 4 from the network operation and management system.

[0110] Specifically, after the first agent operates the network operation and management system to execute the first solution, the first agent obtains operation and maintenance information 4 from the network operation and management system. Because the network operation and management system has already executed the first solution, the newly obtained operation and maintenance information 4 may differ from the previously obtained operation and maintenance information. For example, the network operation and management system has resolved some network faults by executing the first solution, and the operation and maintenance information 4 may contain fewer alarms than the previous operation and maintenance information.

[0111] 111. The first intelligent agent issues a task to check the effectiveness of network operation and maintenance implementation.

[0112] The first intelligent agent determines the inspection task of the network operation and maintenance implementation effect based on the acquired operation and maintenance information 4, and publishes the inspection task of the network operation and maintenance implementation effect in the dialogue channel.

[0113] 112. The second intelligent agent determines the inspection result of the network operation and maintenance implementation effect based on the inspection task of the network operation and maintenance implementation effect.

[0114] The second agent can read the network operation and maintenance implementation effectiveness inspection task posted by the first agent from the conversation channel and analyze the inspection task to obtain the inspection results. The second agent then publishes the inspection results on the conversation channel. The inspection results may include parameters used to measure the network operation and maintenance implementation effectiveness and an evaluation of the network operation and maintenance implementation effectiveness. In addition, the inspection results may also include the analysis steps used to determine the inspection results and the causal relationships between the steps, etc., which are not limited here.

[0115] 113. The third agent reads the inspection task and inspection results, evaluates the inspection results and publishes the evaluation conclusion 4.

[0116] The third agent reads the inspection task posted by the first agent and the inspection result posted by the second agent from the conversation channel and evaluates the inspection result based on the inspection task. Furthermore, the third agent publishes Evaluation Conclusion 4 on the conversation channel. Evaluation Conclusion 4 specifically includes a conclusion on whether the inspection result is accurate. Furthermore, Evaluation Conclusion 4 may include other information, such as the evaluation process and the reasons for determining the evaluation conclusion, though these details are not limited here.

[0117] The method of multiple agents collaboratively evaluating the execution effect of the first solution described in steps 110-113 is similar to the method of multiple agents collaboratively determining the first solution for completing the network operation and maintenance task described in steps 101-104. In both cases, the first agent issues the task, the second agent provides a solution for completing the task, and the third agent evaluates the solution. Therefore, the specific implementation of steps 110-113 can refer to the relevant description of steps 101-104 above and will not be repeated here. Optionally, the first agent and / or the second agent can also evaluate the above-mentioned inspection results and publish the evaluation conclusion.

[0118] 114. The second intelligent agent determines a second plan based on the network operation and maintenance task and publishes the second plan.

[0119] If the first solution is determined to be infeasible through the above step 107, the second agent will re-determine the second solution based on the network operation and maintenance task, and publish the second solution in the dialogue channel. The infeasibility of the first solution means that the first solution cannot be actually executed. For example, the network operation and maintenance task cannot be completed according to the first solution, or the effect of executing the first solution is not ideal, or executing the first solution will bring other problems. For example, the evaluation conclusion 1 published by the third agent indicates that the third agent has checked certain configurations and found that the first solution will introduce some problems. It recommends rethinking and providing a new solution to complete the network operation and maintenance task. After reading the evaluation conclusion 1 published by the third agent, the second agent will deny the analysis idea that determined the first solution before, and then refer to the evaluation logic and suggestions of the third agent and determine the second solution based on a new idea.

[0120] In some possible scenarios, in order to facilitate the second intelligent agent to determine a new second plan, the first intelligent agent can obtain new operation and maintenance information 5 from the network operation and maintenance management system, and provide the operation and maintenance information 5 to the second intelligent agent, so that the second intelligent agent can determine the second plan by integrating the operation and maintenance information 5 and the network operation and maintenance tasks.

[0121] For example, after reading the evaluation conclusion 1 published by the third agent in the conversation channel, the first agent realizes that the first solution currently provided by the second agent is not feasible. The first agent can then proactively obtain new operation and maintenance information 5 from the network operation and management system and publish this information in the conversation channel, allowing the second agent to determine a new second solution based on this information.

[0122] As another example, during its analysis and evaluation of the first solution, the third agent discovers that the second agent's proposed first solution is unfeasible due to limited information. The third agent can then proactively publish a request message 4 on the conversation channel, requesting the first agent to obtain new operation and maintenance information 5 from the network operation and management system. The third agent then publishes this information on the conversation channel, allowing the second agent to determine a new second solution based on this information.

[0123] As another example, after reading the evaluation conclusion 1 published by the third agent on the conversation channel, the second agent realizes that the currently proposed first solution is not feasible, and speculates that this is due to limited information. The second agent can then proactively publish a request message 5 on the conversation channel, requesting the first agent to obtain new operation and maintenance information 5 from the network operation and management system. The second agent then publishes this information on the conversation channel, allowing the second agent to determine a new second solution based on this information.

[0124] After the second agent posts the second solution in the conversation channel, all agents, including the third agent, can evaluate the second solution and negotiate to determine its feasibility. The specific process is similar to steps 104-107 above and will not be repeated here. If multiple agents fail to reach a consensus after rejecting multiple solutions in the above process, they can consider escalating the issue to operations personnel or technical personnel in the conversation channel and wait for their response before taking further action.

[0125] In the above-described embodiments, while the agents collaborate to complete network operation and maintenance tasks, each agent can also perform corresponding operations based on externally input instructions. For example, an operator or technician, while not necessarily involved in the network operation and maintenance task, can, if authorized, view the conversation channel where the agents are interacting at any time to track the progress of the network operation and maintenance task. If authorized, they can also join the conversation channel at any time and provide guidance or instructions on the network operation and maintenance task via text messages. The agent will regard the instructions of the operator or technician as the highest priority.

[0126] The operating environment of multiple intelligent agents can be designed in a variety of different ways, which is conducive to adapting to various application scenarios. They are introduced below.

[0127] Figure 5 is a schematic diagram of the first operating environment of multiple intelligent agents in an embodiment of the present application. As shown in Figure 5, multiple intelligent agents run in the same network operation and management system, that is, the network operation and maintenance system including multiple intelligent agents and the traditional network operation and management system constitute the network operation and management system provided in an embodiment of the present application. The traditional network operation and management system includes different functional modules such as an authorization module, a management module, a control module and an analysis module. Optionally, the large language model can be built into the network operation and management system provided in an embodiment of the present application, or the large language model can be deployed using cloud resources, which is not specifically limited here. In the scenario shown in Figure 5, the communication method within the system is adopted between multiple intelligent agents and between intelligent agents and traditional network operation and management systems. In other words, the intelligent agent and its associated network operation and management system are in the software system of the network operation and management system. Each intelligent agent needs to be started based on its corresponding network operation and management system. As one of the components of the network operation and management system, it cannot run alone and communicates directly with other functional components in the network operation and management system through internal interfaces.

[0128] Figure 6 is a schematic diagram of the second operating environment of multiple agents in an embodiment of the present application. Different from the scenario shown in Figure 5, as shown in Figure 6, multiple agents run in a system independent of the network operation and management system. In the scenario shown in Figure 6, multiple agents use an intra-system communication method, and cross-system API communication is used between the agents and the network operation and management system. Cross-system API communication can be implemented through protocols such as the Border Gateway Protocol (BGP) or Hypertext Transfer Protocol Secure (HTTPS), or through data models such as JS Object Notation (JSON). In other words, the collection of multiple agents is in an independent software system, can run independently, does not rely on the environment of the network operation and management system, and communicates with the network operation and management system through an external interface. The independent software system can be in the same operating system of a physical device as the network operation and management system, or it can be on a separate physical device.

[0129] Figure 7 is a schematic diagram of the third operating environment of multiple intelligent agents in an embodiment of the present application. Similar to the scenario shown in Figure 6, as shown in Figure 7, multiple intelligent agents run in a system independent of the network operation and management system, and cross-system API communication is adopted between the intelligent agents and the network operation and management system. Different from the scenario shown in Figure 6, as shown in Figure 7, there are multiple network operation and management systems in the network, such as a first network operation and management system, a second network operation and management system, a third network operation and management system, etc. Multiple network operation and management systems can manage various types of networks, for example, cross-domain or cross-vendor network operation and management systems, and for example, network operation and management systems across element management systems (EMS), operation support systems (OSS) or network management systems (NMS). Some network operation and maintenance tasks may require multiple network operation and management systems to cooperate with intelligent agents to jointly solve. For example, based on the network type, there may be IP operation and management systems, transmission operation and management systems, wireless operation and management systems, etc.

[0130] Figure 8 is a schematic diagram of the fourth operating environment of multiple agents in an embodiment of the present application. As shown in Figure 8, there are multiple network operation and management systems in the network, and multiple network operation and management systems can manage various types of networks, for example, cross-domain or cross-vendor network operation and management systems, and for example, cross-EMS, OSS or NMS network operation and management systems. There are a certain number of agents with various responsibilities embedded in each network operation and management system. In other words, different from the scenarios shown in Figures 5 to 7 above, different agents can also run in different independent systems. Taking Figure 8 as an example, the first agent runs in the first network operation and management system, and the second agent runs in the second network operation and management system. Some network operation and maintenance tasks may require multiple network operation and management systems to cooperate with agents in multiple network operation and management systems to jointly solve them. Usually, an agent in one of the network operation and management systems initiates or shuts down network operation and maintenance tasks, and interacts with agents in other network operation and management systems through cross-system API communication. Cross-system API communication can be implemented through protocols such as BGP or HTTPS, or through data models such as JSON. In the scenario shown in Figure 8, each network operation and management system includes its own conversation channel. For example, information published by a first agent can be presented in the conversation channel of the first network operation and management system, and can also be received by a second agent and then presented in the conversation channel of the first network operation and management system. In other words, the content presented in the conversation channels of different network operation and management systems remains synchronized, and the interactive content between multiple agents can be synchronously presented in each conversation channel.

[0131] In the scenario shown in Figure 8, as an example, the network contains network devices from Manufacturer A, which are centrally managed by Manufacturer A's operations management system. The network also contains network devices from Manufacturer B, which are centrally managed by Manufacturer B's operations management system. Manufacturer A's operations management system cannot manage Manufacturer B's network devices, and vice versa. When some network operations and maintenance tasks involve equipment from both manufacturers, such as a connection problem between a switch from Manufacturer A and a base station from Manufacturer B, the network operations management systems of Manufacturers A and B need to work together to resolve the issue.

[0132] The following describes several possible network operation and maintenance scenarios in combination with specific network operation and maintenance tasks.

[0133] Figure 9 is a schematic diagram of an implementation method of multiple intelligent agents collaborating to solve a network failure in an embodiment of the present application. Figure 10(a) is a schematic diagram of a conversation channel in the process of multiple intelligent agents collaborating to solve a network failure in an embodiment of the present application. Figure 10(b) is a schematic diagram of another conversation channel in the process of multiple intelligent agents collaborating to solve a network failure in an embodiment of the present application. Figure 10(c) is a schematic diagram of another conversation channel in the process of multiple intelligent agents collaborating to solve a network failure in an embodiment of the present application. As shown in Figure 9, the implementation method includes the following steps. The implementation method shown in Figure 9 can be applied to the several scenarios shown in Figures 5 to 7 above.

[0134] 201. The first intelligent agent obtains fault information from the network operation and management system.

[0135] In the event of a network failure, the network operation and management system can proactively send failure information to the first agent, or the first agent can periodically query the network operation and management system for failure information, which the network operation and management system then feeds back to the first agent. The failure information can include fault alarm information from network devices, etc.

[0136] 202. The first agent issues a fault repair task.

[0137] The first agent determines a fault repair task based on the acquired fault information and publishes it on the conversation channel. This fault repair task involves the agents negotiating a solution to the network fault. As shown in Figure 10(a), the first agent, responsible for software operations, can publish the following content on the conversation channel: "New fault detected: Service interruption on the XX dedicated line, please diagnose and correct it," "List of devices and interfaces along the XX dedicated line: [List]," and "List of alarms reported by managed components along the XX dedicated line: [List]."

[0138] 203. The second agent determines a fault repair plan based on the fault repair task and publishes the fault repair plan.

[0139] The second agent can read the fault repair task posted by the first agent from the conversation channel and analyze the fault repair task to obtain a fault repair plan for completing the task. The second agent then posts the fault repair plan on the conversation channel. The fault repair plan should specifically include the implementation method for completing the fault repair plan. Furthermore, the fault repair plan may also include analysis steps to determine the causal relationships between the steps, etc., although these details are not limited here.

[0140] In one possible scenario, the second agent needs more information to analyze and determine a fault repair solution. The second agent posts a request message on the conversation channel, requesting more fault-related data from the first agent. The first agent reads the request message from the conversation channel and obtains the fault-related data from the network operations management system. The first agent posts the fault-related data on the conversation channel, and the second agent reads the data from the conversation channel and determines a fault repair solution based on the data and the fault repair task.

[0141] In another possible scenario, the second agent can also call network operation and maintenance tools to help determine the fault repair plan based on the fault repair task. For example, it can call tools such as Ping commands or fault knowledge search in the network operation and management system to help determine the fault repair plan for resolving the network fault.

[0142] As shown in Figure 10(a), the second agent used for fault analysis can publish the following content in the dialogue channel: "A link interruption alarm is found on the dedicated line path, and the diagnosis is that the dedicated line service interruption fault scenario is caused by the link interruption", "Please notify the line operation and maintenance team to perform physical troubleshooting for the link interruption fault", "To restore the XX dedicated line service, please perform automatic protection switching (APS) on the dedicated line and enable the protection path of the dedicated line service."

[0143] 204. The third agent reads the fault repair task and the fault repair plan, evaluates the fault repair plan and publishes the evaluation conclusion.

[0144] The third agent reads the fault repair task posted by the first agent and the fault repair solution posted by the second agent from the conversation channel and evaluates the fault repair solution based on the fault repair task. The third agent then publishes its evaluation conclusion on the conversation channel. The evaluation conclusion specifically includes a conclusion on whether the fault repair solution is feasible. The evaluation conclusion may also include other information, such as the evaluation process and the reasons for the evaluation conclusion, though these details are not limited here. Based on the evaluation conclusions, the multiple agents then negotiate a fault repair solution that all agents agree on.

[0145] As shown in Figure 10(a), the third agent used for solution review can publish the following content in the conversation channel: "After verification, we agree with the repair plan for the link interruption" and "For the APS switching operation, more information is required to make a judgment. Please provide the main device path topology and surrounding network device information of the XX dedicated line." In other words, the third agent publishes a request in the conversation channel, requesting the first agent to provide more information.

[0146] As shown in Figure 10(b), the first agent used for software operation can publish the following content in the dialogue channel: "The main device path and surrounding information of the XX dedicated line have been collected, and the alarms of the surrounding network devices have been queried: [List], and the visual topology for operation and maintenance personnel or technical personnel to review: [Topology Map]". The second agent used for fault analysis can publish the following content in the dialogue channel: "It is found that the protection paths are different, and the repair plan of APS switching is rejected." The third agent used for solution review can publish the following content in the dialogue channel: "Sorry, the protection path of the XX dedicated line is different due to another alarm. Both the primary and backup paths cannot be enabled, and the fault scenario is upgraded to a network element full blocking scenario", "It is now necessary to rescue the service of the XX dedicated line fault. Please check whether it is feasible to connect to the CD network device."

[0147] As shown in Figure 10(c), the first agent used for software operation can publish the following content in the dialogue channel: "It has been confirmed that the CD network device can establish a data link." The second agent used for fault analysis can publish the following content in the dialogue channel: "New repair plan: connect the CD network device, change the protection path from C→B to C→D→A, and enable the protection path." The third agent used for solution review can publish the following content in the dialogue channel: "After judgment, the new protection path repair plan is safe and effective, and the repair plan is agreed." The first agent used for software operation can publish the following content in the dialogue channel: "Confirm the connection of the CD network device, modify and switch the protection path of the XX dedicated line, and provide a visual topology for operation and maintenance personnel or technical personnel to review: [Topology Map]."

[0148] 205. The first agent reads the fault repair solution from the dialogue channel.

[0149] 206. The first intelligent agent operates the network operation management system to execute the fault repair plan.

[0150] In some possible scenarios, if the fault repair solution can be implemented through the network operation and management system, then steps 205 and 206 are executed. In other possible scenarios, if the fault repair solution exceeds the capabilities of the network operation and management system, for example, physical operations are required to repair the fault, then steps 207 and 209 are executed.

[0151] 207. The first agent reads the request for help from the dialogue channel.

[0152] 208. The first intelligent agent requests the network operation and management system to notify the operation and maintenance personnel or technical personnel.

[0153] 209. The network operation management system notifies the operation and maintenance personnel or technical personnel.

[0154] Figure 11 is a schematic diagram of another implementation method for multiple agents to collaboratively solve network failures in an embodiment of the present application. As shown in Figure 11, the implementation method includes the following steps. The implementation method shown in Figure 11 can be applied to the scenario shown in Figure 8 above. The types of the first network and the second network in the implementation method shown in Figure 11 are not limited here. For example, the first network is a wireless network and the second network is an IP network. The first agent associated with the first network operation and management system has the responsibilities and functions of the above-mentioned operation agent, analysis agent and review agent, and the second agent associated with the second network operation and management system also has the responsibilities and functions of the above-mentioned operation agent, analysis agent and review agent.

[0155] 301. The first agent obtains fault information from the first network operation and management system.

[0156] In the scenario where a fault occurs in the first network, the first agent first switches to the role of the operating agent. The first network operation and management system can proactively send fault information to the first agent, or the first agent can periodically query the first network operation and management system for fault information, which is then fed back to the first agent.

[0157] 302. The first agent analyzes the problem of the first network.

[0158] After the first agent obtains the fault information from the first network operation and management system, it will switch to the role of the analysis agent. The first agent can first analyze the problem of the first network. In one possible scenario, if the fault is caused by a problem in the first network, the first agent determines the fault repair plan through analysis. Then, the first agent switches to the role of the review agent to evaluate the fault repair plan. If the fault repair plan is evaluated to be feasible, the first agent switches to the role of the operation agent to operate the first network operation and management system to execute the fault repair plan. In another possible scenario, if the cause of the fault does not lie in the first network or the cause of the fault is not only in the first network, the second agent and the second network operation and management system are also required to participate in the fault repair task.

[0159] 303. The first agent publishes a cross-network failure in the dialogue channel.

[0160] If the first agent, through analyzing the problem on the first network, determines that fault repair requires the involvement of a second agent and the second network operations management system, the first agent will publish a cross-network fault in the conversation channel. Furthermore, the first agent may also publish the results and analysis process of the first network problem in the conversation channel. For example, if the first network is a wireless network and the second network is an IP network, and the first agent discovers a connection problem between a base station in the wireless network and a switch in the IP network, the first agent will publish information in the conversation channel indicating that cooperation between the relevant devices in both networks is required to complete the fault repair.

[0161] 304. The second agent reads the cross-network fault from the conversation channel.

[0162] 305. The second agent obtains fault-related data from the second network operation and management system.

[0163] Specifically, after the second agent reads the cross-network fault from the voice channel, it will first switch to the role of the operating agent and obtain fault-related data from the second network operation and management system.

[0164] 306. The second agent analyzes the problem of the second network.

[0165] After obtaining fault-related data from the second network operation and management system, the second agent switches to the role of the analysis agent and analyzes the problem of the second network. Furthermore, the second agent can publish the results and analysis process of the analysis of the problem of the second network in the dialogue channel.

[0166] 307. The first agent and the second agent determine a fault repair plan through negotiation.

[0167] The first agent posts the results and process of its analysis of the first network's problem in the conversation channel. The second agent posts the results and process of its analysis of the second network's problem in the conversation channel. This is equivalent to the first and second agents exchanging their analysis processes and conclusions. Combining this shared information, the first and second agents can each propose a troubleshooting solution. Furthermore, the first and second agents can both switch to the role of review agents, evaluating the troubleshooting solutions currently provided by both parties and exchanging their conclusions. If the first and second agents reach a consensus on their evaluation conclusions, the specific execution phase of the troubleshooting can begin. If they do not reach a consensus on their evaluation conclusions, they will continue to analyze their respective networks and discuss and exchange opinions until they reach a consensus on their evaluation conclusions.

[0168] 308. The first intelligent agent operates the first network operation and management system to execute a fault repair plan.

[0169] 309. The second intelligent agent operates the second network operation and management system to execute the fault repair plan.

[0170] If the first intelligent agent and the second intelligent agent reach a consensus on the evaluation conclusion of the fault repair plan, the first intelligent agent operates the first network operation and management system to execute the fault repair plan, and the second intelligent agent operates the second network operation and management system to execute the fault repair plan, thereby realizing cross-network fault repair.

[0171] In addition to the network operation and maintenance tasks in the fault repair scenarios introduced in Figures 9 and 11 above, the specific implementation methods of multiple intelligent agents collaborating to complete other types of network operation and maintenance tasks are also similar. The following is a brief explanation using the line opening operation and maintenance task as an example. The specific implementation methods can refer to the relevant introductions of the above embodiments and will not be elaborated here.

[0172] In one possible scenario, the network operations management system sends information to a first agent, providing the source and destination network devices. The first agent then forwards this information to a conversation channel. A second agent then requests the first agent's direct network link topology between the source and destination network devices. The first agent obtains the network link topology from the network operations management system and publishes it to the conversation channel. The second agent analyzes the line provisioning process based on the network link topology. This analysis process includes, but is not limited to, using toolformer technology to invoke graph algorithm tools to calculate the shortest path from the source to the destination network device, confirming with the first agent whether all links along the path can establish a new line, whether the remaining bandwidth meets the line requirements, and adjusting the path if necessary. After completing the analysis, the second agent publishes the analysis steps and conclusions (e.g., the path of the new line) to the conversation channel for review by a third agent. The third agent's review process includes, but is not limited to, determining whether the analysis process considers the necessary network conditions, whether the operation steps are continuous and whether there are any skipped steps, and requesting the first agent to use a sandbox environment to verify the feasibility of the new line. After the third agent passes the review, it will publish the conclusion that the assessment is feasible in the dialogue channel. The first agent will read the operating steps required for the line opening and operation and maintenance tasks, and perform corresponding operations on the network operation and management system.

[0173] The manner in which agents interact within a conversation channel in the embodiments of this application is only one possible implementation. In some scenarios, the conversation channel may not be displayed externally, and the collaborative progress of multiple agents may be displayed through a log stream, console, or progress panel, or no communication information may be displayed at all. To enable oversight by operations personnel or technical personnel, methods such as command lines, command buttons, and operations assistants may be combined with the aforementioned log stream, console, or progress panel without reflecting a "conversational" nature. By packaging the conversation process, for example, only a progress announcer can be exposed in the conversation channel, without displaying the agent's thought and communication processes. Assuming that the agents maintain the same level of intelligence, in some operations and maintenance scenarios, a single agent has the potential to replace the work of multiple agents. A single agent, "talking to itself" or combined with other software tools or methods, can approximate the network operations and maintenance capabilities of multiple agents. When agents possess ultra-high intelligence, a single, highly intelligent agent can replace multiple agents with less intelligence.

[0174] As can be seen from the above description, multiple agents, through division of responsibilities and dialogue and collaboration, can solve network operation and maintenance tasks that a single agent cannot. This avoids or resolves the problems of a single agent's limited intelligence and overly subjective thinking, ensuring that the solution to network operation and maintenance tasks is correct, safe, and effective. If the current solution fails, new solutions can be explored, without the intervention of operations personnel or technical personnel. When using multiple specialized agents, the large model parameters and computing power available to a single agent can be biased towards a specific area or capability, achieving the best results for that type of task, without emphasizing the comprehensive intelligence of a single agent. This is because multiple agents can complement each other's capabilities when working together.

[0175] Finally, it should be noted that the above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present application should be included within the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A network operation and maintenance system, characterized in that: The network operation and maintenance system includes a first intelligent agent and a second intelligent agent; The first agent is used to obtain first operation and maintenance information and issue a network operation and maintenance task according to the first operation and maintenance information; The second intelligent agent is used to determine a first solution for completing the network operation and maintenance task based on the network operation and maintenance task, and publish the first solution, wherein the second intelligent agent is driven by a language model.

2. The network operation and maintenance system according to claim 1, characterized in that: The network operation and maintenance system further includes a third agent; The third agent is used to evaluate the first solution and publish the evaluation conclusion.

3. The network operation and maintenance system according to claim 2, characterized in that: The first protocol includes an analysis step for determining the first protocol; The third agent is specifically used to evaluate the first solution according to the analysis step.

4. The network operation and maintenance system according to claim 2, characterized in that: The third agent is further configured to publish a first request message according to the first solution; The second agent is further configured to publish analysis steps for determining the first solution according to the first request message; The third agent is specifically used to evaluate the first solution according to the analysis step.

5. The network operation and maintenance system according to any one of claims 2 to 4, characterized in that: The third agent is further configured to publish a second request message according to the first solution; The first agent is further configured to obtain second operation and maintenance information according to the second request message, and publish the second operation and maintenance information; The third agent is specifically used to evaluate the first solution based on the second operation and maintenance information.

6. The network operation and maintenance system according to any one of claims 2 to 5, characterized in that: If the evaluation conclusion indicates that the first solution is not feasible, the second intelligent agent is further used to determine a second solution for completing the network operation and maintenance task based on the network operation and maintenance task, and publish the second solution.

7. The network operation and maintenance system according to claim 6, characterized in that: If the evaluation conclusion indicates that the first solution is not feasible, the first agent is further configured to obtain third operation and maintenance information and publish the third operation and maintenance information; The second intelligent agent is specifically used to determine the second solution based on the third operation and maintenance information and the network operation and maintenance task.

8. The network operation and maintenance system according to any one of claims 1 to 7, characterized in that: The second agent is further configured to issue a third request message according to the network operation and maintenance task; The first agent is configured to obtain fourth operation and maintenance information according to the third request message; The second intelligent agent is specifically used to determine the first solution based on the network operation and maintenance task and the fourth operation and maintenance information.

9. The network operation and maintenance system according to any one of claims 1 to 8, characterized in that: The second intelligent agent is specifically used to call the network operation and maintenance tool according to the network operation and maintenance task to determine the first solution.

10. The network operation and maintenance system according to any one of claims 1 to 9, characterized in that: The first intelligent agent and the second intelligent agent run in the first network operation and management system, or the first intelligent agent and the second intelligent agent run in a system independent of the first network operation and management system, or the first intelligent agent and the second intelligent agent run in different network operation and maintenance management systems.

11. The network operation and maintenance system according to any one of claims 1 to 10, characterized in that: The interaction between the first agent and the second agent is presented through a conversation in a channel.

12. The network operation and maintenance system according to any one of claims 1 to 11, characterized in that: The language model is a large language model LLM.

13. The network operation and maintenance system according to claim 1, wherein: The first agent and / or the second agent is used to evaluate the first solution and publish an evaluation conclusion.

14. A network operation and maintenance method, characterized in that: The method is applied to a network operation and maintenance system, wherein the network operation and maintenance system includes a first intelligent agent and a second intelligent agent; the method includes: Obtaining first operation and maintenance information through the first agent, and issuing a network operation and maintenance task according to the first operation and maintenance information; The second intelligent agent determines a first solution for completing the network operation and maintenance task based on the network operation and maintenance task, and publishes the first solution, wherein the second intelligent agent is driven by a language model.

15. The method according to claim 14, characterized in that The network operation and maintenance system further includes a third agent, and the method further includes: The first solution is evaluated by the third agent, and an evaluation conclusion is published.

16. The method according to claim 15, characterized in that The first solution includes an analysis step for determining the first solution, and the evaluation of the first solution by the third agent includes: The first solution is evaluated by the third agent according to the analyzing step.

17. The method according to claim 15, characterized in that The method further comprises: Publishing a first request message according to the first solution through the third agent; issuing, by the second agent, an analysis step for determining the first solution according to the first request message; The evaluation of the first solution by the third agent includes: The first solution is evaluated by the third agent according to the analyzing step.

18. The method according to any one of claims 15 to 17, characterized in that The method further comprises: Publishing a second request message according to the first solution through the third agent; Obtaining, by the first agent, second operation and maintenance information according to the second request message, and publishing the second operation and maintenance information; The evaluation of the first solution by the third agent includes: The first solution is evaluated by the third agent based on the second operation and maintenance information.

19. The method according to any one of claims 15 to 18, characterized in that If the evaluation conclusion indicates that the first solution is not feasible, the method further includes: The second intelligent agent determines a second solution for completing the network operation and maintenance task based on the network operation and maintenance task, and publishes the second solution.

20. The method according to claim 19, characterized in that If the evaluation conclusion indicates that the first solution is not feasible, the method further includes: Obtaining third operation and maintenance information through the first agent, and publishing the third operation and maintenance information; Determining, by the second agent according to the network operation and maintenance task, a second solution for completing the network operation and maintenance task includes: The second solution is determined by the second intelligent agent based on the third operation and maintenance information and the network operation and maintenance task.

21. The method according to any one of claims 14 to 20, characterized in that The method further comprises: Publishing a third request message according to the network operation and maintenance task through the second intelligent agent; Obtaining fourth operation and maintenance information through the first agent according to the third request message; Determining, by the second agent, a first solution for completing the network operation and maintenance task according to the network operation and maintenance task includes: The first solution is determined by the second intelligent agent based on the network operation and maintenance task and the fourth operation and maintenance information.

22. The method according to any one of claims 14 to 21, characterized in that Determining, by the second agent, a first solution for completing the network operation and maintenance task according to the network operation and maintenance task includes: The first solution is determined by the second intelligent agent calling a network operation and maintenance tool according to the network operation and maintenance task.

23. The method according to any one of claims 14 to 22, characterized in that The first intelligent agent and the second intelligent agent run in the first network operation and management system, or the first intelligent agent and the second intelligent agent run in a system independent of the first network operation and management system, or the first intelligent agent and the second intelligent agent run in different network operation and maintenance management systems.

24. The method according to any one of claims 14 to 23, characterized in that The interaction between the first agent and the second agent is presented through a conversation in a channel.

25. The method according to any one of claims 14 to 24, characterized in that The language model is a large language model LLM.

26. The method according to claim 14, wherein The method further comprises: The first solution is evaluated and an evaluation conclusion is published through the first agent, and / or the first solution is evaluated and an evaluation conclusion is published through the second agent.

Images