Network operation and maintenance method, network operation and maintenance assessment processing method, agent, network operation and maintenance assessment processing apparatus, agent cluster system, readable medium, and computer product

Abstract

Provided in the present disclosure is a network operation and maintenance method, which is applied to a first agent. The method comprises: when it is predicted that a target link has a fault within a target period, sending a first fault notification to a second agent, and sending a second fault notification to a third agent, wherein the first fault notification is used for instructing the second agent to reroute, before the target period, a service carried on the target link, and to switch the service back to the target link after the fault of the target link is repaired; and the second fault notification is used for instructing the third agent to locate a root cause of the fault of the target link within the target period and repair the fault. Further provided in the present disclosure are a network operation and maintenance assessment processing method, an agent, a network operation and maintenance assessment processing apparatus, an agent cluster system, a readable medium, and a computer product.

Description

Network operation and maintenance methods, network operation and maintenance assessment and processing methods, intelligent agents, network operation and maintenance assessment and processing devices, intelligent agent cluster systems, readable media and computer products

[0001] Cross-references to related applications

[0002] This application claims priority to Chinese Patent Application No. 202411645238.X, filed with the Chinese Patent Office on November 18, 2024, the entire contents of which are incorporated herein by reference. Technical Field

[0003] This disclosure relates to, but is not limited to, the field of communication technology. Background Technology

[0004] In recent years, AIGC (Artificial Intelligence Generated Content) large-scale model technology has developed rapidly, reshaping the industry landscape, and the global demand for automation and intelligent applications has reached unprecedented levels. Most telecom operators and equipment manufacturers have launched automation and digitalization strategies, introducing various innovative large-scale model application solutions to achieve network automation and intelligence, thereby empowering employees, reducing costs, improving efficiency, and enhancing business experience. Current research and practical results show that compared to traditional AI models, large-scale models exhibit significant advantages in intent understanding and decision-making. In particular, large-scale models built on massive datasets possess excellent generalization capabilities, enabling rapid deployment and application of AI while mitigating site-specific differences. Furthermore, the advantages of large-scale models in intent understanding allow them, equipped with vast amounts of telecom knowledge, to control, use, configure, and manage network equipment through interfaces, accelerating the network's move towards self-intelligence.

[0005] Currently, the industry generally believes that intelligent agents are the basic architecture and mainstream form of commercial application for large-scale models, and also represent the development trend and application direction of large-scale model applications. Intelligent agents based on large-scale models generally refer to intelligent computing entities that use large-scale language models or multimodal models with rich training data as core components, possess highly intelligent understanding and generation capabilities, and are thus able to more accurately perceive the environment, make decisions, and execute actions.

[0006] In summary, leveraging the advantages of large-scale intelligent agents, particularly LM-based multi-agent collaboration technology, to enable OTN (Optical Transport Network) management and control systems to possess closed-loop high-order self-intelligence capabilities in environments with intertwined application scenarios has become a current hot topic in the industry for emerging intelligent optical networks (AN OTN). Currently, the industry's focus on the application of LM-based multi-agent collaboration in network communication mainly includes: network monitoring and analysis, service operation and maintenance, and fault management.

[0007] How to control and schedule LM-based multi-agents to handle faults in a timely manner in network service cluster failure scenarios, and how to quickly restore normal operation of batch services, as well as how to ensure the accuracy and reliability of network operation and maintenance, have become the technical bottlenecks that urgently need to be solved to realize high-level autonomy in OTN networks in the era of large-scale models. Summary of the Invention

[0008] This disclosure provides a network operation and maintenance method, a network operation and maintenance assessment and processing method, an intelligent agent, a network operation and maintenance assessment and processing device, an intelligent agent cluster system, a readable medium, and a computer product.

[0009] In a first aspect, embodiments of this disclosure provide a network operation and maintenance method applied to a first intelligent agent. The method includes: sending a first fault notification to a second intelligent agent and sending a second fault notification to a third intelligent agent when a target link is predicted to fail during a target period; wherein the first fault notification is used to instruct the second intelligent agent to reroute the services carried on the target link before the target period and to switch the services back to the target link after the target link fault is repaired; the second fault notification is used to instruct the third intelligent agent to locate the root cause of the target link fault and repair the fault during the target period.

[0010] Secondly, embodiments of this disclosure provide a network operation and maintenance method applied to a second intelligent agent. The method includes: receiving a first fault notification sent by a first intelligent agent; determining, based on the first fault notification, that a target link has failed during a target period; wherein the first fault notification is sent by the first intelligent agent when it predicts that the target link will fail during the target period; rerouting the services carried on the target link before the target period; and, if the target link fault is repaired, switching the services back to the target link.

[0011] Thirdly, embodiments of this disclosure provide a network operation and maintenance method applied to a third intelligent agent. The method includes: receiving a second fault notification sent by a first intelligent agent; determining, based on the second fault notification, that a target link has failed in a target period; wherein the second fault notification is sent by the first intelligent agent when it predicts that the target link will fail in the target period; locating the root cause of the fault in the target link in the target period and repairing the fault.

[0012] Fourthly, this disclosure provides a network operation and maintenance evaluation processing method, applied to a network operation and maintenance evaluation processing device. The method includes: evaluating each network operation and maintenance task performed by a first intelligent agent within each preset evaluation period to obtain a first correct processing probability for each evaluation of the first intelligent agent; evaluating each network operation and maintenance task performed by a second intelligent agent within the evaluation period to obtain a second correct processing probability for each evaluation of the second intelligent agent; and evaluating each network operation and maintenance task performed by a third intelligent agent within the evaluation period to obtain a third correct processing probability for each evaluation of the third intelligent agent; wherein, the... The first agent uses the operation and maintenance model to execute the network operation and maintenance task using the network operation and maintenance method described above; the second agent uses the operation and maintenance model to execute the network operation and maintenance task using the network operation and maintenance method described above; and the third agent uses the operation and maintenance model to execute the network operation and maintenance task using the network operation and maintenance method described above. A joint probability accuracy is calculated based on the first correct processing probability of the first agent in each evaluation, the second correct processing probability of the second agent in each evaluation, and the third correct processing probability of the third agent in each evaluation. If the joint probability accuracy is less than a preset threshold, the parameters of the operation and maintenance model are adjusted.

[0013] Fifthly, embodiments of this disclosure also provide a first intelligent agent, including a memory and a processor; the memory stores a computer program that can be executed by the processor, and when the computer program is executed by the processor, it implements the network operation and maintenance method.

[0014] In a sixth aspect, embodiments of this disclosure also provide a second intelligent agent, including a memory and a processor; the memory stores a computer program that can be executed by the processor, and when the computer program is executed by the processor, it implements the network operation and maintenance method.

[0015] In a seventh aspect, embodiments of this disclosure also provide a third intelligent agent, including a memory and a processor; the memory stores a computer program that can be executed by the processor, and when the computer program is executed by the processor, it implements the network operation and maintenance method.

[0016] Eighthly, this disclosure also provides a network operation and maintenance assessment processing apparatus, including a memory and a processor; the memory stores a computer program that can be executed by the processor, and when the computer program is executed by the processor, it implements the network operation and maintenance assessment processing method as described above.

[0017] In a ninth aspect, embodiments of this disclosure also provide an intelligent agent cluster, including the first intelligent agent, the second intelligent agent, and the third intelligent agent as described above.

[0018] In a tenth aspect, embodiments of this disclosure also provide a computer-readable medium, wherein when the program is executed, it implements the network operation and maintenance method as described above, or the network operation and maintenance assessment and processing method as described above.

[0019] Eleventhly, this disclosure also provides a computer program product, which includes a computer program that, when executed by a processor, implements the network operation and maintenance method as described above, or the network operation and maintenance assessment and processing method as described above. Attached Figure Description

[0020] In the accompanying drawings of the embodiments disclosed herein:

[0021] Figure 1 is a flowchart illustrating the network operation and maintenance method implemented by the first intelligent agent according to an embodiment of this disclosure;

[0022] Figure 2 is a flowchart illustrating the network operation and maintenance method implemented by the second intelligent agent according to an embodiment of this disclosure;

[0023] Figure 3 is a flowchart illustrating the network operation and maintenance method implemented by a third intelligent agent according to an embodiment of this disclosure.

[0024] Figure 4 is a schematic diagram of how a pipeline mechanism is used to achieve multi-agent collaboration to complete high-order closed-loop self-intelligence in an OTN network under a scenario where multiple services are interrupted due to a fault root cause, as provided in an embodiment of this disclosure.

[0025] Figure 5 is a schematic diagram of how a pipeline mechanism is used to achieve multi-agent collaboration to complete high-order closed-loop self-intelligence in OTN networks in the scenario where multiple services are interrupted due to two root causes of failure provided in the embodiments of this disclosure.

[0026] Figure 6 is a flowchart illustrating the network operation and maintenance assessment method provided in this embodiment of the present disclosure;

[0027] Figure 7 is a schematic diagram of the multi-agent deployment, implementation, and operation and maintenance model provided in the embodiments of this disclosure;

[0028] Figure 8 is a schematic diagram of fine-tuning a network operation and maintenance model for multi-agent reuse using the DQNER algorithm provided in an embodiment of this disclosure;

[0029] Figure 9 is a schematic diagram of cache experience information provided in an embodiment of this disclosure;

[0030] Figure 10 is a schematic diagram of the module composition of the first intelligent agent, the second intelligent agent, the third intelligent agent, and the network operation and maintenance evaluation processing device provided in the embodiments of this disclosure;

[0031] Figure 11 is a schematic diagram of an intelligent agent cluster provided in an embodiment of this disclosure. Detailed Implementation

[0032] To enable those skilled in the art to better understand the technical solutions of this disclosure, the embodiments of this disclosure will be described in detail below with reference to the accompanying drawings.

[0033] The present disclosure will be described more fully below with reference to the accompanying drawings; however, the embodiments shown may be embodied in different forms, and the present disclosure should not be construed as limited to the embodiments set forth below. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will enable those skilled in the art to fully understand the scope of the disclosure.

[0034] The accompanying drawings of the embodiments disclosed herein are provided to further illustrate the embodiments of this disclosure and form part of the specification. They are used together with the detailed embodiments to explain this disclosure and do not constitute a limitation thereof. The above and other features and advantages will become more apparent to those skilled in the art from the description of the detailed embodiments with reference to the accompanying drawings.

[0035] This disclosure may be described with reference to plan and / or cross-sectional views using the ideal schematic diagrams of this disclosure. Therefore, the example illustrations may be modified according to manufacturing techniques and / or tolerances.

[0036] Where there is no conflict, the various embodiments of this disclosure and the features thereof in the embodiments may be combined with each other.

[0037] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to limit the disclosure. The term "and / or" as used in this disclosure includes any and all combinations of one or more of the associated enumerated entries. The singular forms "a" and "the" as used in this disclosure are also intended to include the plural forms, unless the context clearly indicates otherwise. The terms "comprising," "made of," etc., as used in this disclosure specify the presence of the stated feature, integral, step, operation, element, and / or component, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof.

[0038] Unless otherwise specified, all terms used in this disclosure (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and this disclosure, and will not be interpreted as having an idealized or overly formal meaning, unless expressly so defined in this disclosure.

[0039] This disclosure is not limited to the embodiments shown in the accompanying drawings, but includes modifications to the configuration based on the manufacturing process. Therefore, the areas illustrated in the drawings are schematic, and the shapes of the areas shown illustrate specific shapes of areas of an element, but are not intended to be limiting.

[0040] This disclosure provides a network operation and maintenance method, which includes fault prediction, fault handling, and service operation and maintenance. Fault handling may include fault root cause location and fault repair, while service operation and maintenance includes service rerouting and service switching. This network operation and maintenance method is used to achieve high-level network autonomy among multiple agents. The network includes, but is not limited to, OTN, PTN (Packet Transport Network), POTN (Packet Optical Transport Network), SPN (Slicing Packet Network), and IP networks. In this disclosure, OTN is used as an example for illustration.

[0041] The network operation and maintenance method is applied to a first intelligent agent. Figure 1 is a flowchart of the network operation and maintenance method implemented by the first intelligent agent according to an embodiment of the present disclosure. As shown in Figure 1, the network operation and maintenance method may include the following step S11.

[0042] In step S11, if it is predicted that the target link will fail during the target period, a first fault notification is sent to the second agent and a second fault notification is sent to the third agent. The first fault notification is used to instruct the second agent to reroute the services carried on the target link before the target period and to switch the services back to the target link after the target link fault is repaired. The second fault notification is used to instruct the third agent to locate the root cause of the target link fault and repair the fault during the target period.

[0043] In this embodiment of the disclosure, at least three intelligent agents are defined, and the process of achieving multi-agent collaboration to complete the high-order closed-loop self-intelligence of OTN network through a pipeline mechanism is described. The responsibilities of the three intelligent agents are defined as follows.

[0044] The first intelligent agent (Agent A) can be a network analysis intelligent agent used to realize the autonomous application functions of OTN network PHM (Prognostics and Health Management), including hazard identification, fault prediction and OTN network health analysis.

[0045] The second intelligent agent (Agent B) can be a business operation and maintenance intelligent agent used to realize the closed-loop intelligent application function of OTN network business operation and maintenance, including: service intent activation, service intent maintenance, etc.

[0046] The third intelligent agent (Agent C) can be a fault management intelligent agent used to realize the closed-loop intelligent application function of OTN network fault management, including: fault root cause identification and location, fault repair scheme design and implementation, fault repair simulation and verification, etc.

[0047] If the first agent predicts that the target link will fail in the target period, it sends a first fault notification to the second agent and a second fault notification to the third agent. Upon receiving the first fault notification, the second agent reroutes the services carried on the target link to other links before the target period. The third agent identifies the target link fault, locates the root cause, and repairs the fault within the target period. After the third agent completes the fault repair, in a period following the target period, it sends a fault repair completion notification to the second agent. Upon receiving this notification, the second agent switches the services rerouted to other links back to the target link.

[0048] This disclosure proposes a pipelined mechanism to achieve high-order closed-loop self-intelligent functionality in OTN networks through multi-agent collaboration, demonstrating significant advantages in resolving group faults and rapidly restoring batch services. The pipelined mechanism refers to decomposing a repetitive sequential process into several sub-processes, each of which can be effectively executed simultaneously with other sub-processes on its dedicated functional segment. By leveraging the advantages of the pipelined mechanism, the efficient and orderly completion of multi-agent collaborative high-order self-intelligent functionality in OTN networks can be ensured, preventing implementation conflicts between different intelligent application function loops.

[0049] The network operation and maintenance method in this embodiment is applied to a first intelligent agent. The method includes: sending a first fault notification to a second intelligent agent and a second fault notification to a third intelligent agent when a fault is predicted to occur in a target link during a target period; the first fault notification is used to instruct the second intelligent agent to reroute the services carried on the target link before the target period and to switch the services back to the target link after the target link fault is repaired; the second fault notification is used to instruct the third intelligent agent to locate the root cause of the target link fault and repair the fault during the target period; this embodiment adopts a pipeline mechanism, in which multiple intelligent agents cooperate to achieve a closed-loop self-intelligence in the form of interwoven multi-application galaxy orbits, which can perform fault prediction, fault root cause location and fault repair in service cluster fault scenarios, and realize rapid emergency response for batch services.

[0050] In some embodiments, after sending a first fault notification to the second agent when it is predicted that the target link will fail during the target period (i.e., step S11), the network operation and maintenance method may further include the following steps: monitoring the network operating status; and sending a first adjustment instruction to the second agent according to the network operating status when the service is switched back to the target link, wherein the first adjustment instruction is used to instruct the second agent to adjust the service switching duration.

[0051] During network operation and maintenance, the first intelligent agent monitors the network's operating status. During service switching, network link degradation and fluctuations occur frequently. In this case, the first intelligent agent sends a first adjustment command to the second intelligent agent, instructing the second intelligent agent to increase the service switching duration to ensure the stability of the repaired target link.

[0052] This disclosure also provides a network operation and maintenance method, which is applied to a second intelligent agent. Figure 2 is a flowchart illustrating the network operation and maintenance method implemented by the second intelligent agent according to this disclosure. As shown in Figure 2, the network operation and maintenance method may include the following steps S21 to S23.

[0053] In step S21, a first fault notification sent by the first intelligent agent is received, and the target link is determined to have failed during the target period based on the first fault notification; wherein, the first fault notification is sent by the first intelligent agent when it predicts that the target link will fail during the target period.

[0054] The second agent receives the first fault notification sent by the first agent, obtains the target link information and target period information carried in the first fault notification, and determines that the target link will fail during the target period.

[0055] In step S22, the services carried on the target link are rerouted before the target period.

[0056] If there is at least one cycle between the target period and the period in which the first agent predicts the target link will fail (hereinafter referred to as the prediction period), the second agent will reroute the services carried on the target link during that interval. If the target period is the next period after the prediction period, i.e., there is no interval between the target period and the prediction period, the services carried on the target link will be rerouted during the target period while the third agent locates the root cause of the target link failure and repairs the failure.

[0057] In step S23, if the target link failure is repaired, the service is switched back to the target link.

[0058] To reduce the impact on existing network services, services can be rerouted and / or switched over during their off-peak hours.

[0059] In some embodiments, rerouting services carried on the target link before the target period (i.e., step S22) includes the following steps: when at least two services are carried on the target link, determining the rerouting order of each service based on the service level and / or service downtime; and rerouting each service before the target period according to the rerouting order. A target link can carry multiple services. If the target link fails, all services carried on it need to be rerouted. Each service is rerouted sequentially, and the rerouting order can be determined based on the service level and / or service downtime.

[0060] The corresponding step of switching services back to the target link (i.e., step S23) includes the following steps: when there are at least two services carried on the target link, each service is switched back to the target link according to the rerouting order.

[0061] In some embodiments, the network operation and maintenance method may further include the following steps: upon receiving a first adjustment instruction sent by a first intelligent agent, adjusting the service switchover duration according to the first adjustment instruction. During the service switchover process, network link degradation and fluctuations frequently occur. In this case, the first intelligent agent sends a first adjustment instruction to a second intelligent agent, and the second intelligent agent increases the service switchover duration according to the first adjustment instruction to ensure the stability of the repaired target link.

[0062] This disclosure also provides a network operation and maintenance method, which is applied to a third intelligent agent. Figure 3 is a flowchart illustrating the network operation and maintenance method implemented by the third intelligent agent according to this disclosure. As shown in Figure 3, the network operation and maintenance method may include the following steps S31 and S32.

[0063] In step S31, a second fault notification sent by the first agent is received, and the target link is determined to have failed during the target period based on the second fault notification; wherein, the second fault notification is sent by the first agent when it predicts that the target link will fail during the target period.

[0064] The third agent receives the second fault notification sent by the first agent, obtains the target link information and target period information carried in the second fault notification, and determines that the target link is about to fail during the target period.

[0065] In step S32, the root cause of the fault in the target link is located and the fault is repaired during the target cycle.

[0066] During the target cycle, the third agent senses the occurrence of a target link failure and performs root cause localization and fault repair verification.

[0067] In some embodiments, a fourth agent (Agent M) is also defined. This fourth agent can be a system scheduling agent, used to orchestrate and design a pipeline implementation scheme for multiple agents to collaboratively complete multiple autonomous application functions of the corresponding OTN. Based on the performance of each agent during the pipeline implementation process, it records, reflects on, and summarizes the results, iteratively improving the current pipeline implementation scheme. In some embodiments, the fourth agent can also monitor the fault repair process, scheduling the third agent after it completes fault repair and adjusting the fault repair duration.

[0068] The network operation and maintenance method may further include the following steps: receiving a second adjustment instruction sent by a fourth intelligent agent, and adjusting the fault repair duration according to the second adjustment instruction; wherein, the second adjustment instruction is sent by the fourth intelligent agent after the second intelligent agent switches the service back to the target link, based on the fault repair monitoring results, and the service is the service rerouted by the second intelligent agent.

[0069] The fourth agent monitors the fault repair process and obtains the fault repair monitoring results by statistically analyzing the fault repair time. If the actual fault repair time of the third agent is shorter than the reserved fault repair time, the fourth agent uses the second adjustment command to instruct the third agent to reduce the fault repair time, that is, to compress the fault repair stage in order to reserve sufficient time for fault repair verification.

[0070] To clearly illustrate the technical solutions of the embodiments of this disclosure, the following detailed description of the solutions of the embodiments of this disclosure is provided in conjunction with FIG4 and FIG5 through two exemplary embodiments.

[0071] Figure 4 is a schematic diagram illustrating the implementation of a pipeline mechanism to achieve high-order closed-loop self-intelligence in an OTN network under a scenario where a fault root cause leads to the interruption of multiple services. As shown in Figure 4, the first agent is network analysis agent A, the second agent is service operation and maintenance agent B, the third agent is fault management agent C, and the fourth agent is system scheduling agent M. The target link is Link1, which carries OTN service 1 and OTN service 2. In this embodiment, a fault root cause leads to the interruption of multiple services, i.e., a fault in Link1 causes the interruption of OTN service 1 and OTN service 2. Multiple agents collaborate to complete multiple closed-loop self-intelligence application functions such as OTN service operation and maintenance and fault management. The network operation and maintenance process includes the following steps S401-S406.

[0072] Step S401: During period T1, Agent A predicts that Link1 will experience signal degradation during period T3. Agent A sends a first fault notification and a second fault notification to Agent B and Agent C, respectively.

[0073] For example, Agent A sends a first fault notification to Agent B: Link1 will experience a degradation fault in period T3. Please prepare for service repair. Agent A then sends a second fault notification to Agent C: Link1 will experience a degradation fault in period T3. Please prepare for diagnosis and repair. Agent B and Agent C reply to Agent A: Received.

[0074] Step S402: In cycle T2, Agent B reroutes the OTN service 1 carried by Link1.

[0075] The links marked with dashed lines in period T2 are the rerouting links. Agent B performs rerouting processing during the fall time of OTN service 1 in period T2.

[0076] Step S403: During the T3 period, Link1 experiences a degradation fault. Agent B reroutes OTN service 2 during the T3 period's sink time. Simultaneously, Agent C identifies and locates the fault during the T3 period and repairs and verifies the root cause of the Link1 fault.

[0077] It should be noted that OTN service 1 and OTN service 2 can also be rerouted within the same cycle. The cycle in which OTN service 1 and OTN service 2 are rerouted depends on the tide and fall times of the two services.

[0078] Step S404: During cycle T4, Agent B switches OTN service 1 back to Link 1 during the tide time of OTN service 1.

[0079] Step S405: During the T5 cycle, Agent B switches OTN Service 2 back to Link 1 during the Tidefall time of OTN Service 2.

[0080] Step S406: In cycle T6, Agent M reflects on and summarizes the entire pipeline implementation process through the multi-Agent interaction window and makes corresponding functional adjustments, such as instructing Agent C to compress the repair time; Agent A instructs Agent B to extend the business failover time from 10 seconds to 20 seconds. For example, in cycle T6, each Agent (M, A, B, C) conducts a review, iterates the business self-intelligence and fault management closed-loop knowledge flywheel, thereby achieving knowledge growth and evolution.

[0081] Figure 5 illustrates a pipelined mechanism for achieving high-order closed-loop self-intelligence in an OTN network under a scenario where two root causes of failure lead to multiple service interruptions. As shown in Figure 5, the first agent is network analysis agent A, the second agent is service operation and maintenance agent B, the third agent is fault management agent C, and the fourth agent is system scheduling agent M. The target links are Link1 and Link2, with Link1 carrying OTN service 1 and Link2 carrying OTN service 2. In this embodiment, two or more root causes of failure lead to multiple service interruptions, i.e., a failure of Link1 causes the interruption of OTN service 1, and a failure of Link2 causes the interruption of OTN service 2. The multiple agents collaborate to complete OTN service operation and maintenance, fault management, and other multi-closed-loop self-intelligence application functions. The network operation and maintenance process includes the following steps S501-S506.

[0082] Step S501: During period T1, Agent A predicts that Link1 will experience signal degradation during period T3. Agent A sends a first fault notification and a second fault notification to Agent B and Agent C, respectively.

[0083] Step S502: During period T2, Agent A predicts that Link2 will experience signal degradation during period T4. Agent A sends a first fault notification and a second fault notification to Agent B and Agent C respectively. At the same time, Agent B reroutes OTN service 1 carried by Link1 during the T2 period.

[0084] Step S503: During the T3 period, Link1 experiences a degradation fault. Agent B reroutes OTN service 2 during the T3 period's sinking time. Simultaneously, Agent C identifies and locates the fault during the T3 period and repairs and verifies the root cause of the Link1 fault.

[0085] Step S504: During the T4 cycle, Link2 experiences a degradation fault. Agent B switches OTN service 1 back to Link1 during the OTN service 1's fall time. At the same time, Agent C performs fault identification and location during the T4 cycle and repairs and verifies the root cause of the Link2 fault.

[0086] Step S505: During the T5 cycle, Agent B switches OTN Service 2 back to Link 2 during the Tidefall time of OTN Service 2.

[0087] Step S506: In cycle T6, Agent M reflects on and summarizes the entire pipeline implementation process through the multi-Agent interaction window and makes corresponding functional adjustments, such as instructing Agent C to compress the repair time; Agent A instructs Agent B to extend the business failover time from 10 seconds to 20 seconds. For example, in cycle T6, each Agent (M, A, B, C) conducts a review, iterates the business self-intelligence and fault management closed-loop knowledge flywheel, thereby achieving knowledge growth and evolution.

[0088] This disclosure also provides a network operation and maintenance assessment processing method, which is applied to a network operation and maintenance assessment processing device. Figure 6 is a schematic flowchart of the network operation and maintenance assessment processing method provided in this disclosure. As shown in Figure 6, the network operation and maintenance assessment processing method may include the following steps S41 to S43.

[0089] In step S41, for each preset evaluation period, the network operation and maintenance tasks performed by the first agent within the evaluation period are evaluated to obtain the first correct processing probability of the first agent in each evaluation; and the network operation and maintenance tasks performed by the second agent within the evaluation period are evaluated to obtain the second correct processing probability of the second agent in each evaluation; and the network operation and maintenance tasks performed by the third agent within the evaluation period are evaluated to obtain the third correct processing probability of the third agent in each evaluation; wherein, the first agent uses the operation and maintenance model to perform network operation and maintenance tasks using the network operation and maintenance method described above, the second agent uses the operation and maintenance model to perform network operation and maintenance tasks using the network operation and maintenance method described above, and the third agent uses the operation and maintenance model to perform network operation and maintenance tasks using the network operation and maintenance method described above.

[0090] In step S42, the joint probability accuracy is calculated based on the first correct processing probability of the first agent in each evaluation, the second correct processing probability of the second agent in each evaluation, and the third correct processing probability of the third agent in each evaluation.

[0091] In step S43, if the joint probability accuracy is less than a preset threshold, the parameters of the operation and maintenance model are adjusted.

[0092] It should be noted that if the joint probability accuracy is greater than or equal to the preset threshold, it indicates that the parameters of the operation and maintenance model are reasonable, and this process ends.

[0093] The network operation and maintenance evaluation method of this disclosure is based on the effect of multi-agent collaboration in handling network faults. It periodically evaluates the accuracy of the large model in the multi-agent network operation and maintenance, and adjusts the large model whose accuracy does not meet the requirements, so as to ensure the accuracy and reliability of network operation and maintenance.

[0094] In some embodiments, the joint probability accuracy can be calculated according to the following formula (1):

[0095] Among them, P i (A) represents the probability of the first correct action of the first agent in the i-th evaluation, P i (B) represents the probability of the second correct action of the second agent in the i-th evaluation, P i (C) represents the third correct processing probability of the third agent in the i-th evaluation, i = (1, 2, ..., N), and N is the total number of evaluations in each preset evaluation period.

[0096] In some embodiments, the operation and maintenance model is obtained by mirroring the original operation and maintenance model. Before evaluating the network operation and maintenance tasks performed by the first agent within each preset evaluation period to obtain the first correct processing probability of each evaluation; and before evaluating the network operation and maintenance tasks performed by the second agent within each evaluation period to obtain the second correct processing probability of each evaluation; and before evaluating the network operation and maintenance tasks performed by the third agent within each evaluation period to obtain the third correct processing probability of each evaluation (i.e., step S41), the network operation and maintenance evaluation processing method may further include the following steps: mirroring the original operation and maintenance model three times to obtain three operation and maintenance models; distributing one operation and maintenance model to the first agent, the second agent, and the third agent respectively for deployment and configuration by the first agent, the second agent, and the third agent.

[0097] In some embodiments, after adjusting the parameters of the operation and maintenance model (i.e., step S43), the network operation and maintenance assessment processing method may further include the following steps S44 to S46.

[0098] In step S44, the original operation and maintenance model is adjusted according to the adjusted parameters to obtain the adjusted original operation and maintenance model.

[0099] In step S45, the adjusted original operation and maintenance model is mirrored and copied three times to obtain three adjusted operation and maintenance models.

[0100] In step S46, an adjusted operation and maintenance model is distributed to the first intelligent agent, the second intelligent agent, and the third intelligent agent respectively, for deployment and configuration by the first intelligent agent, the second intelligent agent, and the third intelligent agent.

[0101] After the parameters of the operation and maintenance model are adjusted, the adjusted operation and maintenance model is mirrored and distributed to each agent so that each agent can perform network operation and maintenance according to the adjusted operation and maintenance model, thereby improving reliability and accuracy.

[0102] Figure 7 is a schematic diagram of the multi-agent deployment and implementation operation and maintenance model provided in this embodiment. The following, in conjunction with Figure 7, using the ChatZ large model as an example, details the process of deploying and implementing the multi-agent operation and maintenance model. As shown in Figure 7, the network operation and maintenance assessment processing method includes the following steps 1-3.

[0103] Step 1: During distributed reasoning, the network operation and maintenance assessment processing device replicates and distributes the unified ChatZ large model image to each agent, acting as the brain base for each agent. For example, ChatZ large model image instance ZA is distributed to network analysis agent A, ChatZ large model image instance ZB is distributed to business operation and maintenance agent B, and ChatZ large model image instance ZC is distributed to fault management agent C.

[0104] ChatZ large model mirroring ensures that agents work in parallel, preventing usage conflicts caused by agents sharing the same model instance.

[0105] Step 2: Implement multi-agent collaborative reasoning using a pipeline mechanism to complete each agent's closed-loop self-intelligence task, i.e., each agent's network operation and maintenance task.

[0106] The specific implementation process of this step is the same as the implementation process of the aforementioned network operation and maintenance method, and will not be repeated here.

[0107] Step 3: Periodically evaluate the large model ChatZ and calculate the joint probability accuracy of the first agent, the second agent, and the third agent. If the joint probability accuracy is too low, i.e., it cannot meet the threshold requirements, it indicates that the inference accuracy of the mirror instance within each agent of the large model ChatZ is too low. In this case, the large model ChatZ will be fine-tuned in a centralized and unified manner.

[0108] Under the multi-agent collaborative architecture, the deployment and implementation scheme of the multi-agent large model base, and the mechanism for periodically evaluating the accuracy of the large model and fine-tuning the multi-agent large model base whose accuracy fails to meet the threshold requirements are as follows:

[0109] 1. Multiple agents reuse the same large model ChatZ, that is, the operation and maintenance model. This large model is mirrored and replicated as the brain base of each agent, and distributedly completes the CoT (Chain of Thought) reasoning implementation of the closed-loop task SOP (Standard Operating Procedure) that each agent needs to complete for its own agent.

[0110] 2. When the inference accuracy of the large model ChatZ in each agent's mirror instance (i.e., the joint probability of multiple agents working together to correctly complete the self-intelligent task in the specified application scenario) is too low and cannot meet the preset threshold requirements, the large model ChatZ will be fine-tuned in a centralized and unified manner.

[0111] 3. The finely tuned large model ChatZ is then mirrored and distributed to each agent to refresh the brain base of each agent.

[0112] In some embodiments, the network operation and maintenance evaluation processing method may further include the following steps: collecting experience information of the first agent, the second agent, and the third agent performing each network operation and maintenance task within a preset evaluation period, and caching the experience information.

[0113] In some embodiments, adjusting the parameters of the operation and maintenance model (i.e., step S43) includes the following steps: using the DQN ER (Deep Q-network Experience Replay) algorithm to adjust the parameters of the operation and maintenance model based on experience information within the evaluation period. That is, the joint probability accuracy is calculated according to the evaluation period; if the joint probability accuracy of the current evaluation period is less than a preset threshold, the operation and maintenance model is fine-tuned based on experience information within the current evaluation period.

[0114] To clearly illustrate the solutions of this disclosure, the fine-tuning process of the network operation and maintenance model is described in detail below through an embodiment. Figure 8 is a schematic diagram of fine-tuning a multi-agent network operation and maintenance model using the DQN ER algorithm provided in this disclosure embodiment. As shown in Figure 8, this embodiment displays an OTN self-intelligent application scenario completed collaboratively by three agents: network analysis agent A, service operation and maintenance agent B, and fault management agent C. The three agents collaboratively complete functions such as network fault / potential prediction / early warning, service quality poor perception / service rerouting / switching, fault root cause identification / location / repair / verification, etc. In this application scenario, the experience samples required for fine-tuning the DQN algorithm are extracted, and some parameters or adapters of the large model base ChatZ (CZ) are fine-tuned in batches using ER technology. When implementing fine-tuning using the DQN ER technology in this example, the definitions of each concept and parameter are explained as follows:

[0115] In the ER technique of DQN, an experience is presented as a transition, which can be represented as: s i ,a i ,r i ,s i+1 ,in:

[0116] s i ,s i+1 —This represents two adjacent states in an algorithmic trajectory in DQN. In this embodiment, seven states are defined: s0, s1, s2; s'1, s'2, s'3, s'4, where...

[0117] s0, s1, s2 — are three adjacent states in the same trajectory;

[0118] s0 represents the OTN network state at a certain moment, during which the network is operating stably without any abnormalities;

[0119] s1 — This represents the OTN network state when Agent A predicts that Link1 will fail in cycle T3;

[0120] s2 — This represents the OTN network status after Agent B reroutes the services carried by Link1, and the resource and performance parameters of each service, link, and node are refreshed.

[0121] s'1, s'2, s'3, s'4 — these are four adjacent states in another trajectory;

[0122] s'1 — This indicates the OTN network status when a Link1 failure actually occurs during the T3 cycle, accompanied by related alarms, anomaly logs, and other information;

[0123] s'2 — This indicates that Agent C's analysis shows the root cause of the Link 1 failure is a damaged A-side receiving optical module, and marks the root cause and location of the failure.

[0124] s'3 — Indicates the OTN network status after Agent C completes the fault repair verification and resumes stable operation;

[0125] s'4 — Indicates the OTN network state after Agent B switches the services originally carried by Link1.

[0126] a i — This represents the s in the trajectory of an algorithm implementation in DQN. i Actions performed under a given state.

[0127] In this embodiment of the disclosure, in order to use the DQN ER algorithm to complete the fine-tuning of the large model base, the following actions are defined corresponding to the above states:

[0128] a0 — This indicates that Agent A predicts in state s0 that Link1 will fail in period T3;

[0129] a1 — indicates that Agent B reroutes the service carried by Link1 in state s1;

[0130] a'1 — This indicates that Agent C, in state s'1, analyzes and concludes that the root cause of the Link1 failure is the damage to the A-side receiving optical module of Link1;

[0131] a'3 — This indicates that Agent B, in state s'3, performs a service switchover on Link1, which was originally carried by the link.

[0132] r i —This represents the trajectory of an algorithm implementation in DQN, located in s. i Action a performed in the state i The timely reward received.

[0133] Because in the DQN algorithm, Q(s) i ,a i ,θ)≈r i +γ·Q(s i+1 ,a i+1 ,θ);

[0134] Q(s i,a i ,θ)——represents the state at s i In the state, the agent performs action a i Action state value Q-value;

[0135] θ — represents the parameters or adapter parameters that are fine-tuned within the large model base;

[0136] γ — represents the value in s of an algorithmic trajectory in DQN. i Action a performed in the state i The expected return discount factor.

[0137] Therefore, r i The value of can be obtained through Q(s) i ,a i ,θ) are reflected. In this embodiment of the disclosure, in order to complete the fine-tuning of the large model base through the DQN ER algorithm, the following action state value Q values ​​are defined corresponding to the above states and actions:

[0138] Q(s0,a0) represents the action state value Q of action a0 performed in state s0. It is the output of the large model base that is fine-tuned in the DQN fine-tuning algorithm after the action a0 is performed in state s0.

[0139] Q(s1,a1) represents the action state value Q of action a1 performed in state s1. It is the output of the large model base that is fine-tuned in the DQN fine-tuning algorithm after the action a1 is performed in state s1.

[0140] Q(s'1,a'1) represents the action state value Q of action a'1 performed in state s'1. It is the output of the large model base that is fine-tuned in the DQN fine-tuning algorithm after performing action a'1 in state s'1.

[0141] Q(s'3,a'3) represents the action state value Q of action a'3 performed in state s'3. It is the output of the large model base that is fine-tuned in the DQN fine-tuning algorithm after performing action a'3 in state s'3.

[0142] In summary, this example defines the following four empirical transitions:

[0143] Transition1: s0, a0, r0, s1, is implemented only within Agent A;

[0144] Transition 2: s1, a1, r1, s2, implemented through collaboration between Agent A and Agent B;

[0145] Transition3: s'1,a'1,r'1,s'2, is implemented only within Agent C;

[0146] Transition 4: s'3,a'3,r'3,s'4, is implemented through collaboration between Agent C and Agent B.

[0147] Transition 1 and Transition 3 involve action implementation within a single agent; Transition 2 and Transition 4 involve action implementation across two agents.

[0148] It should be understood that in Figure 8, x1…x q This represents basic information for each business function, l1…l m Represents the performance information of each link, n1…n k This represents the attribute information of each node, l1 represents degraded link information (e.g., in state S1) or repaired link information (e.g., in state S′3), alm1…alm t This indicates alarm information. Wtrtime represents the rerouting service response time, and NotifyMsg represents the fault repair notification. In state S2, Agent B outputs updated parameters for basic service information, link performance information, and node attribute information after Link1 service rerouting, along with the output evaluation value (Q). In state S′4, Agent B outputs updated parameters for basic service information, link performance information, and node attribute information after Link1 service response, along with the output evaluation value (Q).

[0149] Figure 9 is a schematic diagram of cached experience information provided in an embodiment of this disclosure. As shown in Figure 9, the experience information defined above is stored in the experience replay buffer. The experience information is used for subsequent centralized and unified fine-tuning of the large model base for multi-agent reuse using the DQN ER algorithm.

[0150] In some embodiments, the step of using the DQN ER algorithm to adjust the parameters of the operation and maintenance model based on experience information within the evaluation period includes the following steps S431 to S433.

[0151] In step S431, obtain the experience information within the evaluation period, and calculate the value of the loss function of the parameters of the operation and maintenance model based on the experience information.

[0152] In some embodiments, the loss function of the operation and maintenance model parameters is represented by the following formula (2):

[0153] Where θ represents the parameters of the operation and maintenance model; δ i For time difference error, δ i =q i -y i , q i =Q(s) i ,a i ;θ);T represents the total amount of empirical information within an evaluation period.

[0154] In step S432, the gradient value of the operation and maintenance model corresponding to the loss function value is calculated.

[0155] In some embodiments, the gradient value of the operation and maintenance model corresponding to the loss function value is calculated using the following formula (3):

[0156] Among them, g i Let be the gradient value corresponding to the time difference error of the i-th empirical rule.

[0157] In step S433, the parameters of the operation and maintenance model are iteratively adjusted until the gradient value converges and the iteration stops. The adjusted parameters of the operation and maintenance model are determined based on the parameters of the operation and maintenance model at the time of stopping the iteration.

[0158] Iteratively adjust the parameters θ of the operation and maintenance model, and calculate the target gradient value corresponding to the parameters θ of the operation and maintenance model. Until the target gradient value The iteration stops when the value approaches 0, and the adjusted parameters of the operation and maintenance model are calculated based on the parameter θ corresponding to the time when the iteration stops and the preset learning rate.

[0159] In some embodiments, the target gradient value can be calculated according to the following formula (4), and the parameters of the operation and maintenance model after adjustment can be calculated according to the following formula (5):

[0160] Where α is the preset learning rate.

[0161] This disclosure proposes a method for centralized and unified fine-tuning of a large-scale multi-agent reuse model base using an experience replay (ER) technique employing DQN. It collects experience information related to multi-agent collaboration and intra-agent experience, storing it in an experience replay buffer. Specifically, it stores the quadruple data (state, action, reward, next state) sampled from each application scenario in the replay buffer. When fine-tuning the parameters of the operation and maintenance model, the model is configured using the DQN action-state-value (Q-network) fine-tuning algorithm. Within each evaluation period, several experience information entries are randomly sampled from the replay buffer to form several batches for fine-tuning.

[0162] This disclosure also provides a first intelligent agent, as shown in FIG10, including a memory and a processor; the memory stores a computer program that can be executed by the processor, and when the computer program is executed by the processor, it implements any of the network operation and maintenance methods of this disclosure.

[0163] Among them, the processor is a device with data processing capabilities, including but not limited to the central processing unit (CPU); the memory is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), and flash memory (FLASH); the I / O interface (read-write interface) is connected between the processor and the memory, enabling information exchange between the memory and the processor, including but not limited to the data bus (Bus).

[0164] This disclosure also provides a second intelligent agent, as shown in FIG10, including a memory and a processor; the memory stores a computer program that can be executed by the processor, and when the computer program is executed by the processor, it implements any of the network operation and maintenance methods of this disclosure.

[0165] Among them, the processor is a device with data processing capabilities, including but not limited to the central processing unit (CPU); the memory is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), and flash memory (FLASH); the I / O interface (read-write interface) is connected between the processor and the memory, enabling information exchange between the memory and the processor, including but not limited to the data bus (Bus).

[0166] This disclosure also provides a third intelligent agent, as shown in FIG10, including a memory and a processor; the memory stores a computer program that can be executed by the processor, and when the computer program is executed by the processor, it implements any of the network operation and maintenance methods of this disclosure.

[0167] Among them, the processor is a device with data processing capabilities, including but not limited to the central processing unit (CPU); the memory is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), and flash memory (FLASH); the I / O interface (read-write interface) is connected between the processor and the memory, enabling information exchange between the memory and the processor, including but not limited to the data bus (Bus).

[0168] This disclosure also provides a network operation and maintenance assessment processing device, as shown in FIG10, which includes a memory and a processor; the memory stores a computer program that can be executed by the processor, and when the computer program is executed by the processor, it implements any one of the network operation and maintenance assessment processing methods of this disclosure.

[0169] Among them, the processor is a device with data processing capabilities, including but not limited to the central processing unit (CPU); the memory is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), and flash memory (FLASH); the I / O interface (read-write interface) is connected between the processor and the memory, enabling information exchange between the memory and the processor, including but not limited to the data bus (Bus).

[0170] This disclosure also provides an intelligent agent cluster, as shown in FIG11, including the first intelligent agent, the second intelligent agent and the third intelligent agent described above.

[0171] In some embodiments, the intelligent agent cluster may further include a fourth intelligent agent, which is used to send a second adjustment instruction to a third intelligent agent based on the fault repair monitoring results after the second intelligent agent switches the service back to the target link; the third intelligent agent is used to adjust the fault repair duration according to the second adjustment instruction.

[0172] In some embodiments, the intelligent agent cluster may further include the network operation and maintenance assessment processing device as described above.

[0173] This disclosure also provides a computer-readable medium having a computer program stored thereon, wherein when the program is executed, it implements the network operation and maintenance method as described above, or the network operation and maintenance assessment and processing method as described above.

[0174] This disclosure also provides a computer program product, which includes a computer program that, when executed by a processor, implements the network operation and maintenance method as described above, or the network operation and maintenance assessment and processing method as described above.

[0175] Those skilled in the art will understand that all or some of the steps, systems, and devices disclosed above, as functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0176] In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be executed by several physical components working together.

[0177] Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit (CPU), digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technique for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory (FLASH) or other disk storage; read-only optical disc (CD-ROM), digital versatile disc (DVD) or other optical disc storage; magnetic cartridges, magnetic tapes, disk storage or other magnetic storage; and any other media that can be used to store desired information and can be accessed by a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0178] This disclosure has disclosed exemplary embodiments, and although specific terminology has been used, it is for general illustrative purposes only and should not be construed as limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in conjunction with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in conjunction with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of this disclosure as set forth by the appended claims.

Claims

1. A network operation and maintenance method, applied to a first intelligent agent, the method comprising: If a failure is predicted in the target link during the target period, a first failure notification is sent to the second agent and a second failure notification is sent to the third agent. Wherein, the first fault notification is used to instruct the second agent to reroute the service carried on the target link before the target period, and to switch the service back to the target link after the target link fault is repaired; the second fault notification is used to instruct the third agent to locate the root cause of the fault in the target link and repair the fault during the target period.

2. The method of claim 1, wherein, After sending a first fault notification to the second agent when a fault is predicted to occur in the target link during the target period, the method further includes: Monitor network operating status; When the service is switched back to the target link, a first adjustment instruction is sent to the second agent according to the network operating status. The first adjustment instruction is used to instruct the second agent to adjust the service switching duration.

3. A network operation and maintenance method, applied to a second intelligent agent, the method comprising: The system receives a first fault notification sent by a first intelligent agent and determines, based on the first fault notification, that the target link has failed during the target period; wherein, the first fault notification is sent by the first intelligent agent when it predicts that the target link will fail during the target period. Rerouting the services carried on the target link before the target period; If the target link is repaired, the service will be switched back to the target link.

4. The method of claim 3, wherein, The rerouting of services carried on the target link before the target period includes: When there are at least two services carried on the target link, the rerouting order of each service is determined according to the service level and / or service downtime of each service; According to the rerouting order, each of the services is rerouted before the target period; The step of switching the service back to the target link includes: If at least two services are carried on the target link, each service is switched back to the target link according to the rerouting order.

5. The method of claim 3, wherein, Also includes: Upon receiving the first adjustment instruction sent by the first intelligent agent, the service switchover duration is adjusted according to the first adjustment instruction.

6. A network operation and maintenance method, applied to a third intelligent agent, the method comprising: The system receives a second fault notification sent by a first intelligent agent and determines, based on the second fault notification, that the target link has failed during the target period; wherein, the second fault notification is sent by the first intelligent agent when it predicts that the target link will fail during the target period. The root cause of the fault in the target link is located and the fault is repaired within the target period.

7. The method of claim 6, wherein, Also includes: The system receives a second adjustment instruction from a fourth intelligent agent and adjusts the fault repair duration according to the second adjustment instruction. The second adjustment instruction is sent by the fourth intelligent agent after the second intelligent agent switches the service back to the target link, based on the fault repair monitoring results.

8. A network operation and maintenance assessment processing method, applied to a network operation and maintenance assessment processing device, the method comprising: For each preset evaluation period, the network operation and maintenance tasks performed by the first agent within the evaluation period are evaluated to obtain a first correct processing probability for each evaluation of the first agent; and the network operation and maintenance tasks performed by the second agent within the evaluation period are evaluated to obtain a second correct processing probability for each evaluation of the second agent; and the network operation and maintenance tasks performed by the third agent within the evaluation period are evaluated to obtain a third correct processing probability for each evaluation of the third agent; wherein, the first agent uses the operation and maintenance model to perform the network operation and maintenance tasks using the network operation and maintenance method as described in claim 1 or 2, the second agent uses the operation and maintenance model to perform the network operation and maintenance tasks using the network operation and maintenance method as described in any one of claims 3-5, and the third agent uses the operation and maintenance model to perform the network operation and maintenance tasks using the network operation and maintenance method as described in claim 6 or 7. The joint probability accuracy is calculated based on the first correct processing probability of the first agent in each evaluation, the second correct processing probability of the second agent in each evaluation, and the third correct processing probability of the third agent in each evaluation. If the joint probability accuracy is less than a preset threshold, the parameters of the operation and maintenance model are adjusted.

9. The method of claim 8, wherein, The operation and maintenance model is obtained by mirroring the original operation and maintenance model; before evaluating the network operation and maintenance tasks performed by the first agent within each preset evaluation period to obtain the first correct processing probability of the first agent in each evaluation; and before evaluating the network operation and maintenance tasks performed by the second agent within the evaluation period to obtain the second correct processing probability of the second agent in each evaluation; and before evaluating the network operation and maintenance tasks performed by the third agent within the evaluation period to obtain the third correct processing probability of the third agent in each evaluation, the method further includes: The original operation and maintenance model is mirrored and copied three times to obtain three operation and maintenance models; The operation and maintenance model is distributed to the first intelligent agent, the second intelligent agent, and the third intelligent agent respectively, for deployment and configuration by the first intelligent agent, the second intelligent agent, and the third intelligent agent.

10. The method of claim 9, wherein, After adjusting the parameters of the operation and maintenance model, the following is also included: Based on the adjusted parameters, the original operation and maintenance model is adjusted to obtain the adjusted original operation and maintenance model; The adjusted original operation and maintenance model is mirrored and copied three times to obtain three adjusted operation and maintenance models; The adjusted operation and maintenance model is distributed to the first intelligent agent, the second intelligent agent, and the third intelligent agent respectively, for deployment and configuration by the first intelligent agent, the second intelligent agent, and the third intelligent agent.

11. The method of claim 8, wherein, Also includes: Collect experience information on the first, second, and third agents performing each network operation and maintenance task within the preset evaluation period; The experience information is cached.

12. The method of claim 11, wherein, The adjustment of the parameters of the operation and maintenance model includes: The parameters of the operation and maintenance model are adjusted based on the experience information during the evaluation period using the Deep Q-Network Experience Replay (DQN ER) algorithm.

13. The method of claim 12, wherein, The step of using the Deep Q-Network Experience Replay (DQN ER) algorithm to adjust the parameters of the operation and maintenance model based on the experience information within the evaluation period includes: Obtain empirical information within the evaluation period, and calculate the value of the loss function of the parameters of the operation and maintenance model based on the empirical information; Calculate the gradient value of the operation and maintenance model corresponding to the loss function value; The parameters of the operation and maintenance model are iteratively adjusted until the gradient value converges, at which point the iteration stops, and the adjusted parameters of the operation and maintenance model are determined based on the parameters of the operation and maintenance model at the point where the iteration stops.

14. A first intelligent agent, comprising a memory and a processor; the memory storing a computer program executable by the processor, wherein the computer program, when executed by the processor, implements the network operation and maintenance method of claim 1 or 2.

15. A second intelligent agent, comprising a memory and a processor; the memory storing a computer program executable by the processor, wherein the computer program, when executed by the processor, implements the network operation and maintenance method according to any one of claims 3-5.

16. A third intelligent agent, comprising a memory and a processor; the memory storing a computer program executable by the processor, wherein the computer program, when executed by the processor, implements the network operation and maintenance method of claim 6 or 7.

17. A network operation and maintenance assessment processing apparatus, comprising a memory and a processor; the memory stores a computer program executable by the processor, wherein the computer program, when executed by the processor, implements the network operation and maintenance assessment processing method according to any one of claims 8-13.

18. A swarm of intelligent agents comprising: The first intelligent agent as described in claim 14, the second intelligent agent as described in claim 15, and the third intelligent agent as described in claim 16.

19. The swarm of intelligent beings of claim 18, wherein, It also includes a fourth intelligent agent, which is used to send a second adjustment instruction to the third intelligent agent based on the fault repair monitoring result after the second intelligent agent switches the service back to the target link; The third intelligent agent is used to adjust the fault repair time according to the second adjustment instruction.

20. The intelligent agent cluster according to claim 18, further comprising the network operation and maintenance assessment processing device according to claim 17.

21. A computer readable medium having stored thereon a computer program, wherein, When the program is executed, it implements the network operation and maintenance method according to claim 1 or 2, or the network operation and maintenance method according to any one of claims 3-5, or the network operation and maintenance method according to claim 6 or 7, or the network operation and maintenance assessment and processing method according to any one of claims 8-13.

22. A computer program product comprising a computer program that, when executed by a processor, implements the network operation and maintenance method as described in claim 1 or 2, or the network operation and maintenance method as described in any one of claims 3-5, or the network operation and maintenance method as described in claim 6 or 7, or the network operation and maintenance assessment and processing method as described in any one of claims 8-13.

Images