5GC fault diagnosis root cause analysis method for realizing intelligent decision and iterative closed loop

Abstract

The invention relates to a 5GC fault diagnosis root cause analysis method for realizing intelligent decision and iterative closed loop, and belongs to the field of 5G mobile communication.The method comprises the steps that experience of various fault analysis steps is extracted and analyzed, and a diagnosis model is designed according to entity nodes and entity relations; empirical data is collected for preprocessing, converted according to a model structure and then imported into a graph database, and a diagnosis model is generated; carrying out dynamic planning on the current decision path; establishing a local knowledge base by adopting historical data and knowledge documents, training a large model, and updating a diagnosis model; traversal analysis of the whole decision path is completed based on the diagnosis model, and a final fault analysis conclusion and a treatment scheme are generated; obtaining the diagnosis result of each node by using the atomic power of the 5GC system; and displaying a return result of each step node and a final fault analysis conclusion and a disposal scheme. According to the method, the current network fault can be quickly identified, positioned and processed, meanwhile, processing experience is accumulated, and the network monitoring and fault processing capacity is continuously improved.

Description

Technical Field

[0001] The present invention belongs to the field of 5G mobile communications, and in particular relates to a 5GC fault diagnosis root cause analysis method for realizing intelligent decision-making and iterative closed loop. Background Art

[0002] In order to realize key features such as network slicing, 5G core network elements are implemented using NFV architecture. Their deployment is based on virtual cloud technology and supports functions such as automated deployment and automated configuration based on containers. The dependencies between network elements and their own functions are more complex, and the cause analysis of faults is more difficult. In order to protect the network, operation and maintenance personnel need to monitor the operation status of network elements in real time and handle abnormal problems of network elements in a timely manner. Therefore, monitoring of network element alarms and timely analysis and troubleshooting are key links in 5GC network operation and maintenance. Improving the response speed of monitoring and troubleshooting plays an important role in ensuring the normal operation of 5GC networks. The existing fault analysis technology for 5GC networks has the following problems:

[0003] (1) The existing technology for handling network faults generally relies on the experience of the operation and maintenance personnel themselves. Different alarms from different equipment manufacturers and different equipment types need to be handled according to the manufacturer's maintenance manual and combined with historical experience, which requires high personnel capabilities.

[0004] (2) Existing technologies are generally based on scattered experience, using relational databases, and using fixed rules for step-by-step analysis, resulting in inflexible processes.

[0005] (3) The existing technology generally analyzes through fixed paths, or traverses all paths regardless of whether they are needed. Scanning unnecessary paths increases the analysis delay, increases the load on the system and equipment, and reduces the response speed of operation and maintenance personnel to faults.

[0006] (4) The existing technology generally establishes the initial model manually. After the model is put into operation, the manual method requires a large workload, and the model update cycle is long or even no longer updated. Summary of the invention

[0007] In view of the above-mentioned deficiencies in the prior art, the purpose of the invention is to provide a 5GC fault diagnosis root cause analysis method that realizes intelligent decision-making and iterative closed loop, which can quickly identify, locate and handle existing network faults, while accumulating handling experience and continuously improving network monitoring and fault handling capabilities.

[0008] The present invention proposes a 5GC fault diagnosis root cause analysis method for realizing intelligent decision-making and iterative closed loop, comprising:

[0009] S1, extracting and analyzing the experience of various fault analysis steps, abstracting the fault analysis process into multiple entity nodes and entity relationships, and designing the diagnosis model in the graph database based on the entity nodes and entity relationships; wherein the diagnosis model includes: multiple types of entity nodes and entity relationships;

[0010] S2, collecting experience data of various fault analyses, preprocessing the experience data to generate standard data; analyzing the preprocessed standard data, converting it according to the model structure and importing it into the graph database to generate a diagnostic model;

[0011] S3, based on the fault handling process, a decision path configuration is generated and imported into the diagnostic model. In the operation phase, the next branch path is automatically determined dynamically and in real time according to the return value of each node, so as to realize dynamic planning of the decision path; wherein the decision path includes multiple branch paths;

[0012] S4, automatically import the initial processing flow data collected through expert research into the diagnostic model, use historical data and knowledge documents to build a local knowledge base and train the big model, use the big model based on the RAG architecture to call the big model interface for fault types that are not in the model, interact with the big model by designing prompts, automatically analyze the formatted results, update the diagnostic model, collect user feedback on the recommended solution, and complete the update and iteration of the model based on user feedback;

[0013] S5, based on the decision path, automatically enters the next node according to the return results of each step node; completes the traversal analysis of the entire decision path based on the diagnostic model, and generates the final fault analysis conclusion and disposal plan;

[0014] S6, when each step node is running, the 5GC system atomic capability is used to obtain the diagnosis results of each node;

[0015] S7, presents the return results of each step node, the final fault analysis conclusion and the disposal plan in the form of a flowchart.

[0016] Furthermore, in S1, the various types of fault analysis step experience include: fault handling experience, fault location experience and fault analysis experience.

[0017] Further, in S1, the multiple types of entity nodes include:

[0018] Scenario nodes are used to record different trigger scenario information of faults;

[0019] Step node, used to record the fault handling step information in the current scenario;

[0020] The step conclusion node is used to record the return result of the current step, including the return result enumeration, the description of the return result and the next step node identifier;

[0021] Fault root cause node, used to record the root cause information of fault analysis;

[0022] The solution node is used to record the solution corresponding to the root cause of the fault.

[0023] Furthermore, in S1, the multiple types of entity relationships include:

[0024] The relationship between scenarios and steps is used to record the one-to-many relationship of scenarios including multiple processing steps;

[0025] The relationship between steps and conclusions is used to record a one-to-many relationship where a step includes multiple possible test results;

[0026] The relationship between the conclusion and the root cause is used to record the one-to-many relationship between the returned result and multiple root causes;

[0027] The relationship between the root cause and the solution is used to record the one-to-one relationship between the root cause and the solution.

[0028] Furthermore, in S3,

[0029] A plurality of said branch paths are executed concurrently at the same time;

[0030] Multiple branch paths converge at a node: wait for all branch paths to complete before entering the next node.

[0031] Furthermore, in S4, user feedback data on the results returned by the model is collected, the feedback data is statistically corrected and sorted in real time, the sorting of the results to be selected in the graph database nodes is adjusted, the recommended updated processes and conclusions are screened out, and the automatic update of the model is completed.

[0032] Furthermore, in S4, a fault analysis request triggered by a user is received, the fault analysis request is sent to the diagnostic model for analysis and processing, and the analysis conclusion is sent to the user. If the user determines that the analysis conclusion does not need to be revised, a historical log is recorded; if the user determines that the analysis conclusion needs to be revised, the correction frequency is counted. When the correction frequency reaches the threshold, the node information weight is adjusted according to the rules.

[0033] Furthermore, in S6, the 5GC system atomic capability package includes: alarm query, status query and configuration query.

[0034] Furthermore, in S6, the final result is called back using the 5GC function according to the node relationship between the results of each step in the diagnosis model and the root cause.

[0035] Furthermore, in S7, the detailed interaction process and result data of each step are displayed, and the current analysis progress is marked with different colors.

[0036] The beneficial effects of the present invention are as follows: The method provided by the present invention realizes innovative optimization of the traditional 5GC fault diagnosis root cause analysis process, introduces new technical solutions to realize intelligent decision-making and model iterative update closed loop of root cause analysis, specifically including the following effects:

[0037] (1) Graph Database Model Design

[0038] The method of the present invention is based on the tree structure of graph database entity-relationship, designs the analysis steps and conclusions of each fault as a separate entity, designs the contextual relationship of the process as an entity relationship, and the model structure is relatively flexible, which is convenient for subsequent expansion. Aiming at the needs of fault analysis, the present invention creatively utilizes the structure of graph database entity-relationship to store the complete fault processing process, and designs a configurable dynamic path identification mechanism to dynamically determine the analysis path for the return results of different nodes, thereby improving the flexibility and intelligence level of fault analysis.

[0039] (2) Dynamic planning of decision paths

[0040] Based on the graph database model, the present invention determines the next scanning node according to the real-time results, realizes the dynamic planning of the decision path, reduces the traversal of useless paths, and greatly improves the analysis automation level and efficiency.

[0041] (3) Model iterative closed-loop design

[0042] The present invention combines two methods to automatically update the model:

[0043] First, based on the collaboration of large models, data cleaning and extraction of professional fault handling knowledge are continuously performed, a local knowledge base is established based on incremental data and a large model is trained. The RAG architecture is used to utilize the large model's powerful intent recognition, text processing, knowledge integration, format output and other capabilities. By designing prompts to interact with the large model, the updated version of the processing flow is obtained, and after structuring, the model is imported into the graph database to complete the model update. The present invention innovatively designs an automatic update mechanism for the diagnostic model, which can automatically complete the model update based on the RAG architecture large model collaboration and user feedback, reduce the workload of manual experience collection, improve the practicality and accuracy of the model, and improve the production efficiency of front-line operation and maintenance. The present invention is creatively based on large model technology, and designs a diagnostic model update mechanism based on large model collaboration based on the RAG architecture, making full use of the rich professional knowledge base of the large model, obtaining the processing steps of new faults from the large model based on a dedicated prompt, and combining the statistics of user feedback to automatically complete the model update, reduce the workload of manual experience collection, and improve the efficiency of model update.

[0044] Secondly, based on the user's feedback data on the results returned by the model, after statistics and sorting of the feedback data, the recommended updated processes and conclusions are screened out to complete the automatic update of the model, thereby realizing the automatic iteration closed loop of the model, reducing the workload of manual maintenance of the model, and ensuring the long-term stability and availability of the model.

[0045] (4) Overall system architecture design

[0046] Based on the characteristics of the above-mentioned processes, the present invention comprehensively utilizes IT technologies such as Neo4j, large language model, and Restful to connect the processes in series, realizes an efficient and stable system architecture, and ensures the stable operation of the system in the existing network.

[0047] (5) Process innovation of fault diagnosis

[0048] The present invention provides a fault diagnosis process innovation, and can quickly add new fault handling processes through dynamic configuration, thereby enhancing the flexibility of fault identification and analysis. At the same time, compared with the existing fixed analysis process solution, unnecessary analysis steps can be skipped, further improving the efficiency of analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings are only used to illustrate specific embodiments and are not considered to limit the present invention. In the entire drawings, the same reference symbols represent the same components. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0050] Figure 1 This is a flow chart of a 5GC fault diagnosis root cause analysis method for realizing intelligent decision-making and iterative closed loop according to an embodiment of the present invention;

[0051] Figure 2 This is an architecture diagram of a 5GC fault diagnosis root cause analysis method for realizing intelligent decision-making and iterative closed loop according to an embodiment of the present invention;

[0052] Figure 3 A schematic diagram of dynamic planning of decision paths according to an embodiment of the present invention;

[0053] Figure 4 A schematic diagram of a fault analysis process according to an embodiment of the present invention;

[0054] Figure 5 A schematic diagram of an entity-relationship model of an embodiment of the present invention;

[0055] Figure 6 A schematic diagram of a large model collaborative iteration according to an embodiment of the present invention;

[0056] Figure 7A schematic diagram of user feedback iteration according to an embodiment of the present invention. DETAILED DESCRIPTION

[0057] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making creative work should fall within the scope of protection of the present invention.

[0058] Furthermore, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts disclosed in the present invention.

[0059] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the orientation or position relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. The terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0060] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of methods and systems consistent with some aspects of the present invention as detailed in the appended claims.

[0061] The technical terms involved in the present invention are explained below.

[0062] 5GC: 5G Core, which is the core network part of the 5G mobile communication network.

[0063] NFV: Network Functions Virtualization. In 5GC, core network elements are virtualized using virtualization technology to divide network node functions into several functional blocks, which are implemented separately in software and are no longer limited to dedicated hardware architecture.

[0064] Graph Database: A database that stores and queries data in a "graph" data structure. Its data model is mainly based on nodes and relationships (edges).

[0065] Large model: The present invention mainly uses the large language model LLM (Large Language Model) in the large model, which is an AI system that specializes in understanding and generating human language by processing large amounts of text data, thereby performing various natural language processing tasks such as text classification, question and answer, dialogue, content summarization, etc.

[0066] Prompt: In a large AI model, the main function of Prompt is to give the AI ​​model prompts of the context of input information and parameter information of the input model, which can help the model better understand the intent of the input and make corresponding responses in the required format.

[0067] RAG: Retrieval-augmented Generation, one of the cutting-edge technologies of large models, the retrieval-augmented generation model combines language models and information retrieval technology. When the model needs to generate text or answer questions, it will first retrieve relevant information from a large document collection, and then use the retrieved information to guide text generation, thereby improving the quality and accuracy of predictions.

[0068] The present invention proposes a 5GC fault diagnosis root cause analysis method that realizes intelligent decision-making and iterative closed loop, which belongs to the technical field related to core network element management in 5G mobile communications, and involves technologies such as 5G core network, graph database, and large model. The overall process of the method includes two parts: model generation iteration and analysis decision and display.

[0069] Model generation iteration: model design-data collection and preprocessing-model generation-model iteration;

[0070] Analysis, decision-making and display: Analysis and decision-making - atomic capability call - result display.

[0071] like Figure 1 and Figure 2 As shown, the 5GC fault diagnosis root cause analysis method for realizing intelligent decision-making and iterative closed loop in an embodiment of the present invention includes the following steps:

[0072] S1, extract and analyze the experience of various fault analysis steps, abstract the fault analysis process into multiple entity nodes and entity relationships, and design the diagnosis model in the graph database based on the entity nodes and entity relationships. In this step, the various fault analysis step experiences include: fault handling experience, fault location experience, and fault analysis experience.

[0073] Specifically, this step implements the graph database model design. Based on the extraction and analysis of the experience of various fault analysis steps, the fault analysis process is abstracted into several entities and relationships, mainly including several common objects such as fault scenarios, processing steps, conclusions of each step, comprehensive root causes, and disposal plans. Based on these objects and their sequential relationships, the diagnostic model in the graph database is designed.

[0074] The diagnosis model includes: multiple types of entity nodes and entity relationships. Several important stages in the fault analysis steps are abstracted into entity nodes, and the intermediate results of each step are stored in the nodes. Figure 5 An entity node-entity relationship model is shown.

[0075] In the embodiment of the present invention, the diagnosis model includes 5 types of entity nodes and 4 types of entity relationships. It should be noted that the types of entity nodes and entity relationships can be expanded as needed.

[0076] As shown in Table 1, the five types of entity nodes include: scenario node, step node, step conclusion node, fault root cause node and disposal plan node.

[0077] Table 1

[0078] Node Name illustrate TroScene Scene Node TroMethod Step Node TroResult Step Conclusion Node TroCause Fault root cause node TroSolution Disposal plan node

[0079] 1) Scene node (TroScene)

[0080] Scenario nodes are used to record different triggering scenario information of faults, such as equipment alarms, performance indicator fluctuations, abnormal operations, etc. Scenario nodes are the starting nodes of fault analysis steps and the root nodes of the fault analysis tree. Under each scenario node, the processing steps and all possible path branches of the scenario are recorded.

[0081] 2) Step Node (TroMethod)

[0082] The step node is used to record the fault handling step information in the current scenario. The step node generally includes the step number, step name, step operation (a list of atomic capabilities to be called), etc. All possible steps for analyzing the fault are recorded here. In actual operation, different paths are selected according to the results of each step, and some step nodes may be skipped. After all step nodes are traversed, a fault analysis is completed and the corresponding conclusion is output.

[0083] 3) Step conclusion node (TroResult)

[0084] The step conclusion node is used to record the return result of the current step, including the return result enumeration, the description of the return result, and the next step node identifier. The step conclusion node includes the return result enumeration, the description of the return result, and the next step node identifier. Each step node is associated with at least one conclusion node. During dynamic path planning, the corresponding step conclusion node will be called according to different return results, and then enter the next step node to complete the path traversal.

[0085] 4) Fault root cause node (TroCause)

[0086] The fault root cause node is used to record the root cause information of the fault analysis. Possible root cause information will be associated with each step conclusion node. When the path traversal is completed, the root cause nodes associated with each step node will be sorted according to the number of root cause nodes, and the root cause node with the largest correlation will be taken as the final root cause of the fault.

[0087] 5) Treatment solution node (TroSolution)

[0088] The solution node is used to record the solution corresponding to the root cause of the fault, including process, operation and other information. Each root cause node is associated with a corresponding solution node. When the root cause of the fault is determined, the fault solution will also be determined accordingly.

[0089] The entity relationship associates entity nodes with entity nodes to form a complete fault analysis tree, guiding the fault analysis module to traverse the complete analysis process.

[0090] As shown in Table 2, the four types of entity relationships include: scenarios contain steps, steps contain conclusions, conclusions correspond to root causes, and root causes correspond to solutions.

[0091] Table 2

[0092] Relationship Name illustrate SceneToMethod The scenario includes the following steps: MethodToResult Steps include conclusion ResultToCause Conclusion Corresponding Root Cause CauseToSolution Root cause response plan

[0093] 1) Relationship between scenes and steps (SceneToMethod)

[0094] The relationship between scenarios and steps is used to record the one-to-many relationship in which a scenario includes multiple processing steps.

[0095] 2) Relationship between steps and results (MethodToResult)

[0096] The relationship between steps and conclusions is used to record a one-to-many relationship where a step includes multiple possible test results.

[0097] 3) Relationship between conclusion and root cause (ResultToCause)

[0098] The relationship between the conclusion and the root cause is used to record the one-to-many relationship between the returned result and multiple root causes.

[0099] 4) Cause to Solution

[0100] The relationship between the root cause and the solution is used to record the one-to-one relationship between the root cause and the solution.

[0101] S2, collects empirical data for various fault analyses, preprocesses the empirical data, and generates standard data; analyzes the preprocessed standard data, converts it according to the model structure, and imports it into the graph database to generate a diagnostic model.

[0102] Specifically, after collecting expert experience from experts in various fields, the experience data is preprocessed and processed into a standard step format before the model generation module completes the model import. The preprocessed standard data is analyzed, converted according to the model structure, and imported into the graph database to generate a diagnostic model for subsequent fault diagnosis and troubleshooting.

[0103] S3, based on the fault handling process, generates the decision path configuration and imports the diagnostic model. In the operation phase, it automatically judges based on the return value of each node, dynamically determines the next branch path in real time, and realizes dynamic planning of the decision path. The decision path includes multiple branch paths.

[0104] Specifically, the decision path is a model with multiple branches, and all paths need to traverse different paths according to different results in actual operation. The decision path dynamic programming includes the following steps:

[0105] 1. In the preparation stage, the decision path configuration is generated based on the fault handling process and the diagnosis model is imported;

[0106] 2. During the operation phase, the system automatically determines the next branch path based on the return value of each node, and dynamically plans different paths for different situations.

[0107] 3. During the operation phase, the returned results of each step are integrated and the graph database is called to obtain the final root cause and disposal plan;

[0108] 4. In terms of multi-branch process support, it supports the concurrent execution of multiple branch paths;

[0109] 5. In terms of multi-branch process support, it supports multiple branch path convergence nodes, that is, waiting for all branches to complete operation before entering the next node.

[0110] by Figure 3 Take the dynamic programming of decision path as an example to illustrate.

[0111] First, identify the faulty AMF. Specifically, extract the values ​​corresponding to [Server IP address / Local IP], [Client IP address / Peer IP], and [VRFID / Local VRF] in the alarm. Depending on the completeness of the alarm information, the following two branches are set:

[0112] (1) If the alarm information is complete, then [Server IP address / Local IP]: xxx; [Client IP address / Peer IP]: xxx; [VRFID / Local VRF]: xxx are identified.

[0113] NF acquisition and control can reach:

[0114] 1. Execute SHOW COMMU DNS SERVER on the alarm network element; accurately match the [Server IP Address] of the alarm with the [DNS Server IP Address] in the result returned by the acquisition and control. If the match is successful, the alarm network element (AMF) acquisition and control is reachable.

[0115] 2. Match all other AMF network elements in the province according to the alarm network element, and execute: SHOW COMMU DNSSERVER; if each line can normally obtain [DNS server IP address] [DNS client IP address] [VRFID] and there is no missing, it is considered that the same network elements in the province are reachable for procurement and control.

[0116] (2) If xxx information is missing, exit the diagnosis. Give the diagnosis conclusion: The alarm information is incomplete and the diagnosis cannot be performed. Give the disposal suggestion: Please check the network element configuration information on the AMF! Exit the diagnosis process.

[0117] S4 automatically imports the initial processing flow data collected through expert research into the diagnostic model, uses historical data and knowledge documents to establish a local knowledge base and train the big model, uses the big model based on the RAG architecture to call the big model interface for fault types that are not in the model, interacts with the big model by designing prompts, automatically analyzes the formatted results, updates the diagnostic model, collects user feedback on recommended solutions, and completes model updates and iterations based on user feedback.

[0118] This step realizes the closed loop of model iteration. By cleaning and extracting data of professional fault handling knowledge, training a big model based on data, using the RAG architecture to use the text synthesis and format adjustment capabilities of the big model, and updating the graph database nodes and relationships based on the big model in collaboration with user feedback data, the dynamic iteration update of the diagnostic model throughout its life cycle is completed.

[0119] 1. Initial Model

[0120] After expert investigation, initial processing flow data is collected and automatically imported into the diagnostic model.

[0121] 2. Collaborative large model iteration

[0122] Use historical data and knowledge documents to build a local knowledge base and train a large model. Based on the RAG architecture, use the large model's powerful intent recognition, text processing, knowledge integration, format output and other capabilities to call the large model interface for fault types that are not in the model, interact with the large model by designing prompts, automatically analyze the formatted results, and update the diagnosis model.

[0123] Figure 6 Schematic diagram of large model collaborative iteration in an embodiment of the present invention. Figure 6 The various features in are described below.

[0124] 1. Semi-structured expert experience: This refers to data in a relatively structured table with rows and columns. It refers to data in a JSON format, as follows:

[0125] {

[0126] "Step 1": "Detect network connection",

[0127] "Step Content": [

[0128] "Check whether the network devices connected to the UPF, such as routers and switches, are working properly."

[0129] "Use the ping command to test connectivity to the UPF."

[0130] "Check whether the network configuration, such as IP address, subnet mask, gateway, etc., is correct."

[0131] "Confirm that firewalls or security policies are not blocking communications to the UPF."

[0132] }

[0133] "Return result": "If the network equipment is not working properly or the ping command fails, or the network is misconfigured, or a firewall is blocking communication, there is a problem with the network connection."

[0134] "Failure reason": "Network device failure, network configuration error, improper firewall or security policy configuration."

[0135] “Disposal experience”: “Check the status of network equipment and repair network equipment failures; check network configuration to ensure that the IP address, subnet mask, gateway, etc. are configured correctly; check the firewall or security policy to ensure that the UPF communication is not blocked.”

[0136] }

[0137] 2. Semi-structured output prompt: refers to the prompt written to output expert experience in JSON format, which provides prompt information for large model input.

[0138] 3. Big model: refers to the language big model, similar to the big model platform like Wenxin Yiyan.

[0139] 4. Semi-structured text parsing: refers to parsing the empirical data in JSON format output by the large model into tabular data in standard row and column formats that can be stored in a relational database.

[0140] 5. Structured expert experience: refers to the expert experience data that has been parsed and stored in a relational database.

[0141] 6. Manual review: Whether the automatically generated experience is actually put into the live network needs to be manually reviewed to avoid the inaccuracy of the large model at this stage.

[0142] 7. Diagnostic model: refers to the diagnostic model that can eventually be updated.

[0143] The following is an explanation of the collaborative iteration process of the large model.

[0144] (1) Answer the user's questions based on the questions entered by the user, and answer in Chinese.

[0145] (2) The problem input by the user is divided into multiple steps for solution. There needs to be continuity and sequence between the steps. The next step can only be executed after the previous step is completed.

[0146] (3) Each step should have a corresponding summary title, followed by the content of the step. The content of each step needs to be answered point by point. The specific content of each step needs to be executable, not summary, and well-organized.

[0147] (4) After each step is executed, the corresponding return structure is required to return the result description and summary content. There are two points: whether to jump to the next step or to directly find the problem point after the step is executed and summarize the problem point found.

[0148] (5) After each step is completed, if the problem point is found directly, there is no need to proceed to the next step, so as to obtain the corresponding fault cause and treatment suggestions. The fault cause and treatment experience need to be output separately after the return result of each step.

[0149] (6) The answer must be in a specific format, and each step must be returned separately in JSON format. The JSON format reply template is: [

[0151] {

[0152] "Step 1": xxxx

[0153] "Step Content": xxxx

[0154] "Return result": xxxx

[0155] "Failure reason": xxxx

[0156] "Handling experience": xxxx

[0157] }

[0158] {

[0159] "Step 2": xxxx

[0160] "Step Content": xxxx

[0161] "Return result": xxxx

[0162] "Failure reason": xxxx

[0163] "Handling experience": xxxx

[0164] } ... ]

[0167] Among them, the user's question is: How to solve the problem of UPF unreachability?

[0168] 3. Iteration based on user feedback

[0169] Based on user feedback on recommended solutions, the revision frequency is counted in real time, and the order of the selected results in the graph database nodes is adjusted according to the rules to complete the update and iteration of the model. The revision frequency refers to the number of times the user makes revisions.

[0170] Specifically, collect user feedback data on the results returned by the model, perform real-time statistics and sorting of the feedback data based on the frequency of correction, adjust the sorting of the results to be selected in the graph database nodes, screen out the recommended updated processes and conclusions, and complete the automatic update of the model.

[0171] like Figure 7As shown in the figure, a fault analysis request triggered by a user is received, and the fault analysis request is sent to the diagnosis model for analysis and processing, and the analysis conclusion is sent to the user. If the user determines that the analysis conclusion does not need to be revised, the historical log is recorded; if the user determines that the analysis conclusion needs to be revised, the revision frequency is counted, and when the revision frequency reaches the threshold, the node information weight is adjusted according to the rule. The above rule means that when the threshold is reached, the revised node weight is increased (the weight here is similar to the meaning of priority) so that its priority exceeds the original old node, so that the new node can be selected in the next analysis, thus realizing the update of the node.

[0172] S5, based on the decision path, automatically enters the next node according to the return results of each step node; completes the traversal analysis of the entire decision path based on the diagnostic model, and generates the final fault analysis conclusion and disposal plan.

[0173] S6, when each step node is running, the 5GC system atomic capability is used to obtain the diagnosis results of each node.

[0174] Specifically, when each step node is running, the external system capabilities are called according to the solution design, mainly the 5GC system atomic capabilities, to obtain the diagnostic results of each node and provide a basis for the judgment of the next node.

[0175] In the embodiment of the present invention, the 5GC system atomic capability package includes: alarm query, status query, configuration query, etc. According to the node relationship between the results of each step and the root cause in the diagnosis model, the final result is called back using the 5GC function.

[0176] like Figure 4 As shown, the fault analysis process is as follows:

[0177] (1) The user triggers a fault analysis request;

[0178] (2) Reading the process configuration from the diagnosis model summary to determine the call configuration; wherein the diagnosis model includes: entity nodes and entity relationships, and a decision model;

[0179] (3) Through 5GC atomic capabilities, the process is called and the results are returned;

[0180] (4) Through the 5GC atomic capability, according to the returned result, the subsequent calls are determined until the step is completed;

[0181] (5) Return the optimal result based on the node relationship between the results of each step in the diagnosis model and the root cause;

[0182] (6) The final fault diagnosis result is returned through the 5GC function callback.

[0183] S7, presents the return results of each step node, the final fault analysis conclusion and the disposal plan in the form of a flowchart.

[0184] In this step, the detailed interaction process and fault diagnosis result data of each step are displayed, and the current analysis progress is marked with different colors. For example, the steps with successful operation are marked with a green "√", and the steps with failed operation are marked with a red "×"; for the steps with abnormalities, the abnormal content of the fault diagnosis is marked with red font.

[0185] The 5GC fault diagnosis root cause analysis method for realizing intelligent decision-making and iterative closed loop of the present invention is based on graph database technology and large model professional capabilities, summarizes expert experience, establishes an automatic fault analysis path, forms an automated processing flow, quickly classifies and judges the equipment alarms that occur, and provides reasonable analysis conclusions and disposal suggestions, helping front-line operation and maintenance personnel to quickly identify, locate, and handle existing network faults, while accumulating disposal experience and continuously improving network monitoring and fault handling capabilities.

[0186] The method of the present invention innovates the 5GC fault diagnosis root cause analysis process and adopts a closed-loop optimization solution based on a graph database. Through this solution, the professional knowledge of fault handling is imported into the big model, and the processing flow is automatically added based on the interaction between the graph database and the big model. At the same time, combined with user feedback, the analysis path, analysis conclusion and recommended disposal plan in the graph database are continuously updated through automatic statistics and sorting, thereby realizing the self-optimization and self-iteration of the graph database troubleshooting model, thereby supporting the long-term, stable and effective use of the model, reducing the workload of front-line operation and maintenance personnel, and realizing the intelligent decision-making and iterative closed-loop capabilities of the process.

[0187] The present invention innovates and optimizes the fault diagnosis process and supports a closed-loop solution for iterative model optimization based on a graph database. Through the method of the present invention, a knowledge base based on a large model RAG architecture is established, and a new processing flow is automatically added based on the interaction between the graph database and the large model. At the same time, combined with user feedback, the analysis path, analysis conclusion, and recommended disposal plan in the graph database are continuously updated through automatic statistics and sorting, thereby realizing the self-optimization and self-iteration of the graph database troubleshooting model, thereby supporting the long-term, stable, and effective use of the model and reducing the workload of front-line operation and maintenance personnel.

[0188] The 5GC fault diagnosis root cause analysis method for realizing intelligent decision-making and iterative closed loop according to the embodiment of the present invention has the following application scenarios:

[0189] In order to realize key features such as network slicing, 5G core network elements are implemented using NFV architecture. Their deployment is based on virtual cloud technology, supporting functions such as automated deployment and automated configuration based on containers. The dependencies between network elements and their own functions are more complex, and the cause analysis of failures is more difficult. To protect the network, operation and maintenance personnel need to monitor the operation status of network elements in real time and handle abnormal problems of network elements in a timely manner. Therefore, monitoring of network element alarms and timely analysis and troubleshooting are key links in 5GC network operation and maintenance. Improving the response speed of monitoring and troubleshooting plays an important role in ensuring the normal operation of 5GC networks.

[0190] The method of the present invention can be applied in the 5GC intelligent driving management and scheduling subsystem of the new generation cloud network operation system in the telecommunications industry to realize automatic analysis of existing 5GC network equipment failures, improve the automation and intelligence level of fault monitoring, reduce the complex workload of front-line operation and maintenance personnel, and improve production efficiency. It is being more widely used in production environments.

[0191] It should be noted that the method of the present invention can be used not only in the telecommunications industry, but also in other management systems with similar network monitoring requirements. Scenarios with similar requirements can collect professional troubleshooting experience, connect professional atomic capabilities and professional large models for adaptation based on the solution of the present invention, thereby achieving efficient, automatic and intelligent network monitoring functions and empowering various industries.

[0192] The 5GC fault diagnosis root cause analysis method that realizes intelligent decision-making and iterative closed loop of the present invention is mainly used to monitor 5G core network equipment failures; the graph database in the present invention is mainly used for the association between the fault and its analysis steps, conclusions, and solutions, and a technical solution for collaborative updating of large models is proposed.

[0193] The present invention uses a graph database to store fault processing, and can determine the next node based on the results of each node in real time, which has high flexibility. In addition, the present invention also designs a relatively complete model update mechanism, that is, a combination of large model collaboration and user feedback.

[0194] Monitoring of network element alarms and timely analysis and troubleshooting are key links in 5GC network operation and maintenance. Improving the response speed of monitoring and troubleshooting plays an important role in ensuring the normal operation of the 5GC network.

[0195] In the field of network monitoring, troubleshooting generally relies on two types of knowledge. The first is basic knowledge and principles, that is, the fault analysis of equipment needs to be based on the professional knowledge in authoritative manuals such as the equipment operation manual, fault handling manual, and operation and maintenance manual. This is the basis for fault analysis and processing. The second is the experience accumulation of operation and maintenance personnel, that is, after long-term work practice, a set of accurate and efficient processing procedures are formed for common faults. This processing procedure is an optimization of the standard procedure and is based on the refinement of multiple actual processing work, so it is more practical and effective.

[0196] In the field of professional knowledge, the present invention innovatively introduces the knowledge comprehensive processing capability of a large model, and designs and utilizes a large model with core network operation and maintenance knowledge to update and iterate the operation and maintenance model.

[0197] In terms of experience accumulation, based on the structural characteristics of professional knowledge, the present invention innovatively introduces Neo4j graph database software into the process, imports experts' fault handling experience and processes into the database, and forms a processing chain of fault scenario-processing steps-cause analysis-disposal plan. At the same time, it supports updating the processing chain based on user feedback to realize the accumulation and updating of processing experience.

[0198] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the above embodiments, or replace some of the technical features therein by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or replacements that can be easily thought of by those skilled in the art within the technical scope disclosed in the present invention should be covered within the protection scope of the present invention.

Claims

1. A 5GC fault diagnosis root cause analysis method for realizing intelligent decision-making and iterative closed loop, characterized in that: include: S1, extracting and analyzing the experience of various fault analysis steps, abstracting the fault analysis process into multiple entity nodes and entity relationships, and designing the diagnosis model in the graph database based on the entity nodes and entity relationships; wherein the diagnosis model includes: multiple types of entity nodes and entity relationships; S2, collecting experience data of various fault analyses, preprocessing the experience data to generate standard data; analyzing the preprocessed standard data, converting it according to the model structure and importing it into the graph database to generate a diagnostic model; S3, based on the fault handling process, a decision path configuration is generated and imported into the diagnostic model. In the operation phase, the next branch path is automatically determined dynamically and in real time according to the return value of each node, so as to realize dynamic planning of the decision path; wherein the decision path includes multiple branch paths; S4, automatically import the initial processing flow data collected through expert research into the diagnostic model, use historical data and knowledge documents to build a local knowledge base and train the big model, use the big model based on the RAG architecture to call the big model interface for fault types that are not in the model, interact with the big model by designing prompts, automatically analyze the formatted results, update the diagnostic model, collect user feedback on the recommended solution, and complete the update and iteration of the model based on user feedback; S5, based on the decision path, automatically enters the next node according to the return results of each step node; completes the traversal analysis of the entire decision path based on the diagnostic model, and generates the final fault analysis conclusion and disposal plan; S6, when each step node is running, the 5GC system atomic capability is used to obtain the diagnosis results of each node; S7, presents the return results of each step node, the final fault analysis conclusion and the disposal plan in the form of a flowchart.

2. According to claim 1, a 5GC fault diagnosis root cause analysis method for realizing intelligent decision-making and iterative closed loop is characterized in that: In S1, the various types of fault analysis step experience include: fault handling experience, fault location experience and fault analysis experience.

3. According to claim 1, a 5GC fault diagnosis root cause analysis method for realizing intelligent decision-making and iterative closed loop is characterized in that: In S1, the multiple types of entity nodes include: Scenario nodes are used to record different trigger scenario information of faults; Step node, used to record the fault handling step information in the current scenario; The step conclusion node is used to record the return result of the current step, including the return result enumeration, the description of the return result and the next step node identifier; Fault root cause node, used to record the root cause information of fault analysis; The solution node is used to record the solution corresponding to the root cause of the fault.

4. According to claim 1, a 5GC fault diagnosis root cause analysis method for realizing intelligent decision-making and iterative closed loop is characterized in that: In S1, the multiple entity relationships include: The relationship between scenarios and steps is used to record the one-to-many relationship of scenarios including multiple processing steps; The relationship between steps and conclusions is used to record a one-to-many relationship where a step includes multiple possible test results; The relationship between the conclusion and the root cause is used to record the one-to-many relationship between the returned result and multiple root causes; The relationship between the root cause and the solution is used to record the one-to-one relationship between the root cause and the solution.

5. According to the 5GC fault diagnosis root cause analysis method realizing intelligent decision-making and iterative closed loop according to claim 1, it is characterized in that: In S3, A plurality of said branch paths are executed concurrently at the same time; Multiple branch paths converge at a node: wait for all branch paths to complete before entering the next node.

6. According to a 5GC fault diagnosis root cause analysis method for realizing intelligent decision-making and iterative closed loop according to claim 1, it is characterized in that: In S4, the feedback data of users on the results returned by the model is collected, the statistics and sorting of the feedback data are corrected in real time, the sorting of the results to be selected in the graph database nodes is adjusted, the recommended updated processes and conclusions are screened out, and the automatic update of the model is completed.

7. A 5GC fault diagnosis root cause analysis method for realizing intelligent decision-making and iterative closed loop according to claim 6, characterized in that: In S4, a fault analysis request triggered by a user is received, the fault analysis request is sent to the diagnostic model for analysis and processing, and the analysis conclusion is sent to the user. If the user determines that the analysis conclusion does not need to be revised, a historical log is recorded; if the user determines that the analysis conclusion needs to be revised, the correction frequency is counted. When the correction frequency reaches the threshold, the node information weight is adjusted according to the rules.

8. The 5GC fault diagnosis root cause analysis method for realizing intelligent decision-making and iterative closed loop according to claim 1, characterized in that: In S6, the 5GC system atomic capability package includes: alarm query, status query and configuration query.

9. The 5GC fault diagnosis root cause analysis method for realizing intelligent decision-making and iterative closed loop according to claim 1, characterized in that: In S6, the final result is called back using the 5GC function based on the node relationship between the results of each step in the diagnosis model and the root cause.

10. The 5GC fault diagnosis root cause analysis method for realizing intelligent decision-making and iterative closed loop according to claim 1, characterized in that: In S7, the detailed interaction process and result data of each step are displayed, and the current analysis progress is marked with different colors.

Images