Network fault resolution solution generation method and device, computer device, storage medium and program product

Abstract

The application relates to a network fault solution generation method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: acquiring alarm information of a network system, inputting the alarm information into a first large language model, determining a fault network segment in each network segment by the first large language model; performing fault analysis on the alarm information by using each fault analysis model corresponding to the fault network segment to obtain network fault analysis information output by each fault analysis model; inputting the network fault analysis information output by each fault analysis model into a second large language model to generate a network fault analysis result by the second large language model; and generating a network fault solution based on the network fault analysis result by the second large language model in the case that the network fault analysis result represents that the alarm information is real alarm information. The method can improve the fault processing efficiency of the network system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a network fault solution generation method and device, computer equipment, storage medium and computer program product. BACKGROUND

[0002] In the intelligent network operation scene, a traditional AI model is often used to build a network optimization system for identifying network faults and providing solutions, to provide network optimization solutions for the network system. However, since such a network optimization system only uses a traditional AI model, it can only provide a general network optimization direction according to the network detection result provided by the AI model, and cannot provide detailed solutions, thereby failing to efficiently respond to fault problems that are about to occur or have occurred in the network system, and failing to maintain the network system in a timely manner.

[0003] Therefore, the conventional technology has the problem of low efficiency in handling faults of the network system. SUMMARY

[0004] Therefore, it is necessary to provide a network fault solution generation method, device, computer equipment, computer readable storage medium and computer program product capable of improving the efficiency of handling faults of the network system.

[0005] A network fault solution generation method applied to a network system, the network system comprising network segments, the method comprising:

[0006] obtaining alarm information of the network system, and inputting the alarm information into a first large language model to determine a fault network segment in each network segment by the first large language model; the first large language model being obtained by pre-training a general large language model using professional domain knowledge corpus data of each network segment;

[0007] performing fault analysis on the alarm information using each fault analysis model corresponding to the fault network segment to obtain network fault analysis information output by each fault analysis model;

[0008] inputting the network fault analysis information output by each fault analysis model into a second large language model to generate a network fault analysis result by the second large language model; the second large language model being obtained by fine-tuning the first large language model using business scenario knowledge corpus data of each network segment;

[0009] in a case where the network fault analysis result represents that the alarm information is real alarm information, generating a network fault solution based on the network fault analysis result by the second large language model; the network fault solution being used to assist in generating a fault handling work order for the alarm information.

[0010] In one of the embodiments, the alarm information of the network system is acquired, including:

[0011] The alarm index description information sent by each network segment in the network system is received;

[0012] The alarm index description information sent by each network segment is spliced and fused to obtain the alarm information.

[0013] In one of the embodiments, the network fault solution is generated based on the network fault analysis result by the second large language model, including:

[0014] The knowledge graph and knowledge base of the network system are called by the second large language model to query the structured data matched with the network fault analysis result;

[0015] The structured data is semantically understood by the second large language model to generate the network fault solution.

[0016] In one of the embodiments, the method further includes:

[0017] The full amount of corpus data is acquired, and the knowledge graph of the network system is constructed based on the full amount of corpus data;

[0018] Based on the knowledge graph of the network system, the professional field knowledge corpus data of each network segment and the business scenario knowledge corpus data of each network segment are extracted;

[0019] The general large language model is pre-trained by using the professional field knowledge corpus data of each network segment to obtain the first large language model with fault network segment positioning function;

[0020] The first large language model is fine-tuned by using the business scenario knowledge corpus data of each network segment to obtain the second large language model with network fault analysis function.

[0021] In one of the embodiments, the first large language model is fine-tuned by using the business scenario knowledge corpus data of each network segment to obtain the second large language model with network fault analysis function, including:

[0022] The first large language model is fine-tuned by using the business scenario knowledge corpus data of each network segment to obtain the fine-tuned large language model;

[0023] The full amount of corpus data is subjected to entity recognition by using the fine-tuned large language model to obtain each entity recognition result;

[0024] The expert review results for each entity recognition result are acquired, and the positive samples and negative samples are separated from each entity recognition result based on the expert review results;

[0025] Based on the positive samples and the negative samples, the fine-tuned large language model is iteratively fine-tuned to obtain a second large language model with network fault analysis function.

[0026] In one of the embodiments, based on the positive samples and the negative samples, the fine-tuned large language model is iteratively fine-tuned to obtain a second large language model with network fault analysis function, comprising:

[0027] Based on the entity recognition result corresponding to the positive sample, the knowledge graph is updated to obtain an updated knowledge graph;

[0028] From the updated knowledge graph, the business scenario knowledge corpus data of each network segment is re-extracted, and the fine-tuned large language model is fine-tuned using the re-extracted business scenario knowledge corpus data of each network segment.

[0029] In one of the embodiments, based on the positive samples and the negative samples, the fine-tuned large language model is iteratively fine-tuned to obtain a second large language model with network fault analysis function, comprising:

[0030] Based on the entity recognition result corresponding to the negative sample, a prompt word sample is constructed;

[0031] The fine-tuned large language model is fine-tuned using the prompt word sample.

[0032] In one of the embodiments, after the step of generating a network fault solution based on the network fault analysis result by the second large language model, the method further comprises:

[0033] Obtain the fault handling result after executing the fault handling order;

[0034] In the case where the fault handling result represents that the network fault triggering the alarm information has not been solved, generate feedback information;

[0035] Input the feedback information into the first large language model and the second large language model to update the first large language model and the second large language model.

[0036] A network fault solution generation device applied to a network system, the network system comprising network segments, the device comprising:

[0037] An alarm information acquisition module for acquiring alarm information of the network system and inputting the alarm information into a first large language model to determine a fault network segment in each network segment by the first large language model; the first large language model is obtained by pre-training a general large language model using professional domain knowledge corpus data of each network segment;

[0038] The analysis information acquisition module is configured to perform fault analysis on the alarm information by using each fault analysis model corresponding to the fault network segment, and obtain network fault analysis information output by each fault analysis model.

[0039] The analysis result generation module is configured to input the network fault analysis information output by each fault analysis model into the second large language model, and generate a network fault analysis result by using the second large language model; the second large language model is obtained by fine-tuning the first large language model by using the service scenario knowledge corpus data of each network segment in advance.

[0040] The solution generation module is configured to generate a network fault solution based on the network fault analysis result by using the second large language model in a case where the network fault analysis result represents that the alarm information is real alarm information; the network fault solution is used to assist in generating a fault handling work order for the alarm information.

[0041] A computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the method when executing the computer program.

[0042] A computer readable storage medium stores a computer program, and the computer program implements the steps of the method when executed by a processor.

[0043] A computer program product includes a computer program, and the computer program implements the steps of the method when executed by a processor.

[0044] The aforementioned network fault solution generation method, apparatus, computer equipment, storage medium, and computer program product acquire alarm information from the network system and input the alarm information into a first large language model. The first large language model then identifies the faulty network segment within each network segment. The first large language model is pre-trained using domain-specific knowledge corpus data from each network segment on a general large language model. Fault analysis models corresponding to the faulty network segment are used to analyze the alarm information, obtaining network fault analysis information output by each fault analysis model. This network fault analysis information is then input into a second large language model. The network fault analysis results are generated using a second language model. This second language model is obtained by fine-tuning the first language model using business scenario knowledge corpora from various network segments. If the network fault analysis results represent real alarm information, a network fault solution is generated based on the network fault analysis results using the second language model. This network fault solution is used to assist in generating fault handling work orders for the alarm information. In this way, the first language model, obtained by pre-training a general language model using professional domain knowledge corpora from various network segments, can perform semantic understanding of alarm information and accurately determine the fault in each network segment. The system first identifies a faulty network segment and then uses the corresponding fault analysis models to output network fault analysis information for the alarm information. Next, a second language model, fine-tuned from the first language model using business scenario knowledge corpus data from each network segment, is used to perform semantic understanding of the network fault analysis information, generating accurate network fault analysis results. If the network fault analysis results represent the alarm information as genuine, the second language model generates network fault solutions based on the network fault analysis results. This approach leverages knowledge graphs to enhance the understanding of the interrelationships between network segments within the network system, while also considering the advantages of knowledge graphs. High accuracy, strong interpretability, and the strong semantic understanding capabilities of large language models facilitate overall detection and root cause analysis of the entire network system. Furthermore, it integrates large language models with different functionalities with traditional fault analysis models, balancing the semantic understanding characteristics of large language models with the professionalism and lightweight nature of traditional fault analysis models. This improves the timeliness and accuracy of anomaly detection in network systems. Overall, it combines the advantages of knowledge graphs, large language models, and traditional fault analysis models, enhancing the efficiency and accuracy of network system fault handling. Attached Figure Description

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the embodiments or related art description will be briefly introduced. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0046] Figure 1 An application environment diagram of a network fault solution generation method in an embodiment;

[0047] Figure 2 A flowchart of a network fault solution generation method in an embodiment;

[0048] Figure 3 A schematic diagram of a network fault solution generation method in an embodiment;

[0049] Figure 4 A relationship diagram of a general large language model, a first large language model and a second large language model in an embodiment;

[0050] Figure 5 A schematic diagram of a server of a network optimization system generating alarm information in an embodiment;

[0051] Figure 6 A schematic diagram of a mutual enhancement process of a knowledge graph and a large language model in an embodiment;

[0052] Figure 7 An application mode schematic diagram of a large language model as a system entrance of a network optimization system in an embodiment;

[0053] Figure 8 An application architecture diagram of a network optimization system in an embodiment;

[0054] Figure 9 A flowchart of a network fault solution generation method in another embodiment;

[0055] Figure 10 A structural block diagram of a network fault solution generation device in an embodiment;

[0056] Figure 11 An internal structure diagram of a computer device in an embodiment. DETAILED DESCRIPTION

[0057] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0058] Intelligent operation and maintenance uses artificial intelligence, automation, big data analysis and other technologies to monitor and analyze network system operation data in real time, which can help to predict network system failures, optimize network system performance, and improve network system stability and reliability. In the current network optimization field, traditional artificial intelligence algorithms and knowledge graphs are often used to improve the intelligence of network systems.

[0059] In recent years, large language models have rapidly risen, and there is no mature technology to realize the professionalization of general large language models, nor is there mature technology to fuse large language models with traditional AI small models and knowledge graphs to fuse multi-source information in network systems and realize high structuring and explainability of knowledge graphs, professionalism and lightweight of traditional AI small models, and generalization and language understanding ability of large language models.

[0060] The network fault solution generation method provided by the embodiments of the present application can fuse multi-source information in network systems, realize high structuring and explainability of knowledge graphs, professionalism and lightweight of traditional AI small models, and generalization and language understanding ability of large language models.

[0061] The network fault solution generation method provided by the embodiments of the present application can be applied to, for example Figure 1The application environment shown. Among them, the terminal 102 communicates with the server 104 of the network optimization system through the network. The data storage system can store the data required by the server 104 of the network optimization system to process. The data storage system can be integrated on the server 104 of the network optimization system, or placed on the cloud or other network servers. The server 104 of the network optimization system obtains the alarm information of the network system, and inputs the alarm information into the first large language model to determine the fault network segment in each network segment through the first large language model; the first large language model is obtained by pre-training a general large language model using professional domain knowledge corpus data of each network segment; the server 104 of the network optimization system uses each fault analysis model corresponding to the fault network segment to analyze the alarm information, and obtains the network fault analysis information output by each fault analysis model; the server 104 of the network optimization system inputs the network fault analysis information output by each fault analysis model into the second large language model, and generates the network fault analysis result through the second large language model; the second large language model is obtained by fine-tuning the first large language model using business scenario knowledge corpus data of each network segment; the server 104 of the network optimization system generates a network fault solution based on the network fault analysis result through the second large language model in the case that the network fault analysis result represents that the alarm information is real alarm information; the fault solution is used to assist in generating a fault handling work order for the alarm information. The server 104 of the network optimization system can send the fault handling work order to the terminal 102. Among them, the terminal 102 can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices, Internet of Things devices can be smart speakers, smart televisions, smart air conditioners, smart vehicle devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server 104 can be realized by an independent server or a server cluster composed of multiple servers.

[0062] In one exemplary embodiment, as shown in Figure 2 , a network fault solution generation method is provided, which is applied to the server 104 of the network optimization system in Figure 1 for example, including the following steps S202 to S208. Among them:

[0063] Step S202, obtaining the alarm information of the network system, and inputting the alarm information into the first large language model to determine the fault network segment in each network segment through the first large language model; the first large language model is obtained by pre-training a general large language model using professional domain knowledge corpus data of each network segment.

[0064] The alarm information can be obtained from customer complaints, alarm system control alarms, and alarm system logs.

[0065] The first large language model is obtained by pre-training a general large language model using professional domain knowledge corpus data of each network segment.

[0066] The professional domain knowledge corpus data of each network segment is generated based on the corresponding graph data of the professional domain knowledge graph of each network segment. The professional domain knowledge graph of any network segment is constructed based on the professional domain knowledge of the network segment. Training the general large language model using the graph data corresponding to the professional domain knowledge graph of each network segment can make the large language model familiar with the professional domain knowledge of each network segment, so that the large language model can quickly determine which network segment the alarm information comes from when receiving the alarm information.

[0067] In practical applications, the professional domain knowledge graph of each network segment is dynamically updated and continuously improved.

[0068] The network system includes network segments such as access networks, backbone networks, core networks, and wireless networks. In traditional network optimization systems, each network segment is independently maintained. However, in reality, each network segment is associated with each other. The alarm generated by the wireless network may be a problem of the core network. It is necessary to perform overall detection and traceability on the entire network when receiving the alarm information. The method provided by the present application uses a knowledge graph to solve the problem that the traditional network optimization system cannot perform overall detection and traceability on the entire network.

[0069] Optionally, with reference to Figure 3 The alarm information of the network system can be obtained through customer complaints, alarm system control alarms, and alarm system logs. The alarm information is input as a prompt word into the first large language model. The first large language model performs semantic understanding on the alarm information to determine the network segment that sent the alarm information, and determines the network segment as the fault network segment.

[0070] In fact, step S202 classifies the alarm information as a problem and determines which network segment the alarm information belongs to.

[0071] In step S204, each fault analysis model corresponding to the fault network segment is used to analyze the alarm information, and network fault analysis information output by each fault analysis model is obtained.

[0072] The fault analysis model corresponding to the fault network segment can include various fault analysis models constructed based on an association rule mining algorithm, a heuristic search algorithm, Prophet (a time series prediction algorithm), LSTM (Long Short-Term Memory), and the like.

[0073] The network fault analysis information can include fault classification, fault probability, fault positioning, fault trend judgment, and traffic early warning information.

[0074] Optionally, with reference to Figure 3 , the alarm information is analyzed by using the fault analysis models corresponding to the fault network segments to obtain fault classification, fault probability, fault positioning, fault trend judgment, and traffic early warning information output by the fault analysis models.

[0075] In step S206, the network fault analysis information output by the fault analysis models is input into a second large language model, and a network fault analysis result is generated by the second large language model. The second large language model is obtained by fine-tuning the first large language model by using the business scenario knowledge corpus data of each network segment in advance.

[0076] The second large language model is obtained by fine-tuning the first large language model by using the business scenario knowledge corpus data of each network segment in advance.

[0077] For the convenience of understanding of those skilled in the art, Figure 4 a relationship diagram of the general large language model, the first large language model, and the second large language model is provided, Figure 4The L0 basic large language model in the L0 basic large language model corresponds to a general large language model, the L1 wireless network field large language model and the L1 core network field large language model correspond to the first large language model, and the L2 wireless network scenario large language model and the L2 core network scenario large language model correspond to the second large language model. The L0 basic model can include a CV large model (a large deep learning model for computer vision tasks), an NLP large model (Natural Language Processing, a natural language processing large model), a multi-modal large model, a scientific computing large model, a graph network large model, ChatGPT (Chat Generative Pre-trained Transformer, based on the GPT system large model), ChatGGLM (a generative pre-training model based on an autoregressive language model), Baichuan (a large language model with more than 100 billion parameters), InternLM (an advanced natural language processing model with strong semantic understanding capabilities, which can realize functions such as automatic classification, abstract extraction, and entity recognition of text by using InternLM), Vicuna (an open-source chat robot), LLaMA (Large Language Model Meta AI), Alpaca (an instruction-following language model based on the Meta LLaMA 7B model), TeleChat (Xingchen Semantic Large Model), and Ziya (a 13 billion parameter bilingual pre-training language model based on LLaMa). The first large language model such as the L1 wireless network field large language model and the L1 core network field large language model has functions such as knowledge understanding, analysis and prediction, automatic optimization, and solution generation. The L2 wireless network scenario large language model has solution generation functions for OMC gateway load abnormalities, network element security, automatic station opening, automatic acceptance, and cell service withdrawal. The L2 core network scenario large language model has solution generation functions for network element faults, capacity overruns, link faults, address pool utilization overruns, and software versions.

[0078] The business scenario knowledge corpus data of each network segment is generated based on the graph data corresponding to the business scenario knowledge graph of each network segment. The business scenario knowledge graph of any network segment is constructed based on the business scenario knowledge of the network segment. The first large language model with fault network segment positioning function is trained using the business scenario knowledge graph of each network segment. The second large language model obtained by training is familiar with the business scenario knowledge of each network segment. Therefore, under the condition that the second large language model is familiar with the business scenario knowledge of the fault network segment, the second large language model can identify whether the alarm information is a false alarm according to the network fault analysis information output by the fault analysis model corresponding to the fault network segment, and can locate the fault under the condition that the alarm information is not a false alarm.

[0079] In practical applications, the service scenario knowledge graph of each network segment is dynamically updated and continuously improved.

[0080] The network fault analysis result can be a network fault analysis conclusion obtained by the second large language model after semantic understanding of the network fault analysis information output by each fault analysis model. The conclusion indicates whether the alarm information is a false alarm, and in the case that the alarm information is not a false alarm, the corresponding fault location result.

[0081] Optionally, with reference to Figure 3 The network fault analysis information output by each fault analysis model is input into the second large language model, and the network fault analysis information output by each fault analysis model is subjected to semantic understanding by the second large language model, and the network fault analysis result is output.

[0082] In step S208, in the case that the network fault analysis result represents that the alarm information is real alarm information, a network fault solution is generated based on the network fault analysis result by the second large language model; the network fault solution is used to assist in generating a fault handling work order for the alarm information.

[0083] The network fault solution can be used as an auxiliary solution to assist in generating a fault handling work order for the alarm information.

[0084] Optionally, with reference to Figure 3 In the case that the network fault analysis result represents that the alarm information is not a false alarm, a specific fault location is determined in the fault network segment according to the fault location provided by the network fault analysis result, and a corresponding network fault solution is generated by the second large language model.

[0085] In practical applications, in the case that the network fault analysis result represents that the alarm information is a false alarm, an expert will further review the alarm information, and if the expert determines that the alarm information is indeed a false alarm, the process is ended.

[0086] In the method for generating the network fault solution, alarm information of a network system is obtained, and the alarm information is input into a first large language model to determine a fault network segment in each network segment by the first large language model; the first large language model is obtained by pre-training a general large language model by using professional domain knowledge corpus data of each network segment; each fault analysis model corresponding to the fault network segment is used to analyze the fault of the alarm information, and network fault analysis information output by each fault analysis model is obtained; the network fault analysis information output by each fault analysis model is input into a second large language model, and a network fault analysis result is generated by the second large language model; the second large language model is obtained by fine-tuning the first large language model by using business scenario knowledge corpus data of each network segment; in the case that the network fault analysis result represents that the alarm information is real alarm information, a network fault solution is generated by the second large language model based on the network fault analysis result; the network fault solution is used to assist in generating a fault handling work order for the alarm information; in this way, the first large language model obtained by pre-training a general large language model by using professional domain knowledge corpus data of each network segment is used to perform semantic understanding on the alarm information, the fault network segment is accurately determined in each network segment, each network fault analysis model corresponding to the fault network segment is used to output network fault analysis information for the alarm information, the second large language model obtained by fine-tuning the first large language model by using business scenario knowledge corpus data of each network segment is used to perform semantic understanding on each network fault analysis information, and an accurate network fault analysis result is generated; in the case that the network fault analysis result represents that the alarm information is real alarm information, the second large language model is used to generate a network fault solution based on the network fault analysis result, the understanding ability of the knowledge graph enhanced large language model for the interrelation between each network segment in the network system is realized, the knowledge graph accuracy and strong explainability are taken into account, and the strong semantic understanding ability of the large language model is taken into account, thereby facilitating the overall detection and traceability of the network system, meanwhile, the large language models with different functions and the traditional fault analysis models with different functions are fused, the semantic understanding characteristics of the large language models with different functions and the professional and lightweight characteristics of the traditional fault analysis models with different functions are taken into account, the timeliness and accuracy of the network system in discovering abnormalities are improved, and overall, the advantages of the knowledge graph, the large language model and the traditional fault analysis model are fused, the processing efficiency and the processing accuracy of the network system fault are improved.

[0087] In an exemplary embodiment, obtaining alarm information of a network system includes: receiving alarm indicator description information sent by each network segment in the network system; and splicing and fusing the alarm indicator description information sent by each network segment to obtain the alarm information.

[0088] The alarm index description information can be information obtained by describing the abnormal device state index in natural language when the device state of the device is abnormal, and the alarm index description information can also be description information of KPI (Key Performance Indicator), KQI (Key Quality Indicator) and other indicators.

[0089] Optionally, the server of the network optimization system can receive the alarm index description information sent by each network segment in the network system, and splice and fuse the alarm index description information sent by each network segment to obtain alarm information. The alarm information can be input as a prompt word to the first large language model in the subsequent process, so that the first large language model performs semantic understanding on the alarm information to determine the network segment with a fault.

[0090] Reference Figure 5 When each network segment is abnormal, the alarm index description information is sent to the alarm system, and the alarm system sends the alarm index description information to the network optimization system, or the user submits complaint information to the network optimization when the user is aware of the abnormality. The server of the network optimization system splices and fuses each alarm index description information sent by the alarm system and the user complaint information to obtain alarm information, and then executes steps S202 to S208 of the previous embodiment to obtain a network fault solution. Since the second large language model is fine-tuned based on the graph data corresponding to the professional domain knowledge graph of each network segment, the second large language model can output the quality problem positioning, fault cause analysis, solution process and recommended solution for the alarm information based on the learned knowledge to assist decision-making.

[0091] In this embodiment, the alarm index description information sent by each network segment in the network system is received, and the alarm index description information sent by each network segment is spliced and fused to obtain alarm information. In this way, the alarm data from different network segments can be spliced and used as the input of the first large language model, which realizes the fusion of multi-source information across network segments and is beneficial to improving the timeliness and accuracy of problem discovery.

[0092] In an exemplary embodiment, the second large language model generates a network fault solution based on the network fault analysis result, including: calling the knowledge graph and knowledge base of the network system by the second large language model to query structured data matched with the network fault analysis result; and generating the network fault solution by the second large language model based on semantic understanding of the structured data.

[0093] The knowledge graph of the network system can be a professional domain knowledge graph corresponding to the network system, which can represent the association relationship between each network segment of the network system, and the knowledge graph is a stock data.

[0094] The knowledge base of the network system can be a text vector library storing professional knowledge of the network system. In actual applications, the knowledge base can be called by the second large language model in a mounted manner.

[0095] In actual applications, the data of the knowledge graph can be regarded as stock data, and the data of the knowledge base can be regarded as real-time data. The knowledge graph and the knowledge base can be understood as an information retrieval system, which is called by the second large language model. Therefore, the network optimization system can be compatible with stock data processing and real-time data processing. The knowledge graph and the knowledge base can be used as a verification tool in the problem accurate solution scenario to reduce model illusion and improve the accuracy and security of LLM input.

[0096] The structured data can be query results output by the knowledge graph. Since the knowledge graph is a structured semantic knowledge base, the search results output by the knowledge graph are also structured data.

[0097] Optionally, with reference to Figure 3 In the case that the network fault analysis result represents that the alarm information is not a false alarm, the knowledge graph and the knowledge base of the network system are called by the second large language model, the network fault analysis result is input into the knowledge graph and the mounted knowledge base, structured data matched with the network fault analysis result is queried, and the second large language model is used to perform semantic understanding on the structured data to generate a network fault solution.

[0098] In actual applications, after the network fault solution is generated, a pre-order process is entered, and an expert audits the solution of the pre-order process. If the expert determines that the solution is accurate, an order step is performed.

[0099] In the embodiment, the knowledge graph and the knowledge base of the network system are called by the second large language model to query structured data matched with the network fault analysis result. The second large language model is used to perform semantic understanding on the structured data to generate a network fault solution. In this way, in the case that the network fault analysis result is obtained, the structured data matched with the network fault analysis result is queried from the stock knowledge graph and the real-time knowledge base, and the second large language model is used to perform semantic understanding on the structured data to generate a network fault solution. The accuracy of the solution generation is improved.

[0100] In an example embodiment, the method further comprises: obtaining full-volume corpus data, and constructing a knowledge graph of the network system based on the full-volume corpus data; extracting professional domain knowledge corpus data of each network segment and business scenario knowledge corpus data of each network segment based on the knowledge graph of the network system; pre-training the general large language model using the professional domain knowledge corpus data of each network segment to obtain a first large language model having a fault network segment positioning function; and fine-tuning the first large language model using the business scenario knowledge corpus data of each network segment to obtain a second large language model having a network fault analysis function.

[0101] The full-volume corpus data can include corpus data of coarse-grained corpus libraries such as wikis, white papers, and dictionaries, and can also include corpus data of fine-grained corpus libraries such as cases, alarms, and work orders. In actual applications, as shown in Figure 6 The full-volume corpus data is updated in real time according to the collected data, and the knowledge graph of the network system is also in a process of continuous improvement and update, so the professional domain knowledge corpus data of each network segment and the business scenario knowledge corpus data of each network segment are also updated, and the first large language model and the second large language model are continuously updated. In the process of pre-training the first large language model, the input text sequence is encoded as The entity stored in the knowledge graph is encoded as The input text sequence encoding and the entity encoding need to be linearly combined and corrected by the expert knowledge weight matrix W, which is represented as .

[0102] The fault network segment positioning function can refer to the function of determining the fault network segment in each network segment of the network system. In actual applications, the fault network segment positioning function refers to that the first large language model can perform semantic understanding on the received alarm information based on the learned professional domain knowledge of each network segment, so as to determine the fault network.

[0103] The network fault analysis function can refer to that the second large language model performs semantic understanding on the network fault analysis information output by each fault analysis model of the fault network segment based on the learned professional domain knowledge of each network segment and the learned business scenario knowledge of each network segment, so as to determine whether the alarm information is a false alarm, and in the case that the alarm information is not a false alarm, to locate the fault.

[0104] Optionally, reference is made to Figure 6The server obtains full corpus data, and constructs a knowledge graph of the network system based on the full corpus data. The server extracts professional domain knowledge corpus data of each network segment and business scenario knowledge corpus data of each network segment based on the knowledge graph of the network system. The server pre-trains a general large language model (corresponding to L0 general LLM) using the professional domain knowledge corpus data of each network segment (corresponding to coarse-grained corpus data such as wiki, white paper, and dictionary), and the pre-training process includes sampling, preprocessing, and using a RoPE method (Rotary Position Embedding, rotary position embedding technology) and the like. A first large language model (corresponding to L1 field LLM) with fault network segment positioning function is obtained. The first large language model is fine-tuned using the business scenario knowledge corpus data of each network segment (corresponding to fine-grained corpus data such as cases, alarms, and work orders), and a second large language model (corresponding to L2 scenario LLM) with network fault analysis function is obtained.

[0105] In the embodiment, full corpus data is obtained, and a knowledge graph of the network system is constructed based on the full corpus data. Professional domain knowledge corpus data of each network segment and business scenario knowledge corpus data of each network segment are extracted based on the knowledge graph of the network system. A general large language model is pre-trained using the professional domain knowledge corpus data of each network segment, and a first large language model with fault network segment positioning function is obtained. The first large language model is fine-tuned using the business scenario knowledge corpus data of each network segment, and a second large language model with network fault analysis function is obtained. In this way, the knowledge graph can be used to strengthen the training of the large language model, and the first large language model with fault network segment positioning function and the second large language model with network fault analysis function are obtained through training, thereby facilitating accurate positioning of fault network segments and accurate judgment of the authenticity of alarm information.

[0106] In one exemplary embodiment, fine-tuning the first large language model using the business scenario knowledge corpus data of each network segment to obtain the second large language model with network fault analysis function includes: fine-tuning the first large language model using the business scenario knowledge corpus data of each network segment to obtain a fine-tuned large language model; performing entity recognition on the full corpus data using the fine-tuned large language model to obtain each entity recognition result; obtaining expert review results for each entity recognition result, and separating positive samples and negative samples from each entity recognition result based on the expert review results; and based on the positive samples and the negative samples, iteratively fine-tuning the fine-tuned large language model to obtain the second large language model with network fault analysis function.

[0107] The entity recognition result can correspond to a specific matter in the full corpus data, for example, entity recognition of certain corpus data can obtain specific matters such as "equipment", "index", "fault" and the like involved in the corpus data.

[0108] The positive sample can refer to an entity recognition result that is correct according to the output of the large language model, and the negative sample can refer to an entity recognition result that is incorrect according to the output of the large language model.

[0109] Optionally, referring to Figure 6 , the business scenario knowledge corpus data (corresponding to case, alarm, work order and the like fine-grained corpus data) of each network segment is used to fine-tune the first large language model (corresponding to L1 scene LLM), the fine-tuning can use the LoRA method (a method for fine-tuning large pre-trained language models, the core idea is to add small, low-rank matrices in the key layers of the model to adjust the behavior of the model, rather than directly changing the structure of the entire model) to obtain a fine-tuned large language model (corresponding to L2 scene LLM), then the fine-tuned large language model (corresponding to L2 scene LLM) is used to perform entity recognition on the full corpus data to obtain each entity recognition result, RLHF (Reinforcement Learning from Human Feedback) method is used for reinforcement learning, each entity recognition result is audited by an expert, and based on the expert audit result, the positive sample and the negative sample are separated from each entity recognition result, and the positive sample and the negative sample are used to iteratively fine-tune the fine-tuned large language model (corresponding to L2 scene LLM) to obtain a second large language model (corresponding to L2 scene LLM) with network fault analysis function.

[0110] It should be noted that the "fine-tuned large language model" and "second large language model with network fault analysis function" in the present embodiment are both second large language models with network fault analysis function, and correspond to the state of the second large language model at different times. The second large language model is always in the process of iterative fine-tuning.

[0111] In the embodiment, the first large language model is fine-tuned by using the service scenario knowledge corpus data of each network segment to obtain a fine-tuned large language model; entity recognition is performed on the full corpus data by using the fine-tuned large language model to obtain each entity recognition result; expert review results for each entity recognition result are obtained, and positive samples and negative samples are separated from each entity recognition result based on the expert review results; the fine-tuned large language model is iteratively fine-tuned based on the positive samples and the negative samples to obtain a second large language model having a network fault analysis function; in this way, the accuracy of the fine-tuned large language model can be verified by using the entity recognition on the full corpus data, and in the process of verifying the accuracy of the fine-tuned large language model, the reinforcement learning technology based on human feedback is used, that is, the expert review feedback is used to capture subtle differences and subjectivity, which is beneficial to the self-learning of the large language model more effectively, can accelerate the fine-tuning process of the large language model, and improve the performance of the large language model.

[0112] In one exemplary embodiment, the fine-tuned large language model is iteratively fine-tuned based on the positive samples and the negative samples to obtain a second large language model having a network fault analysis function, including: updating the knowledge graph based on the entity recognition result corresponding to the positive sample to obtain an updated knowledge graph; reextracting the service scenario knowledge corpus data of each network segment from the updated knowledge graph, and fine-tuning the fine-tuned large language model by using the reextracted service scenario knowledge corpus data of each network segment.

[0113] Optionally, referring to Figure 6 , the knowledge graph of the network system is updated based on the entity recognition result corresponding to the positive sample, the service scenario knowledge corpus data of each network segment is reextracted from the updated knowledge graph, and the fine-tuned large language model is fine-tuned by using the reextracted service scenario knowledge corpus data of each network segment.

[0114] In the embodiment, the knowledge graph is updated based on the entity recognition result corresponding to the positive sample to obtain an updated knowledge graph; the service scenario knowledge corpus data of each network segment is reextracted from the updated knowledge graph, and the fine-tuned large language model is fine-tuned by using the reextracted service scenario knowledge corpus data of each network segment; in this way, the knowledge graph of the network system can be updated by using the positive sample, and on this basis, the service scenario knowledge corpus data of each network segment can be reextracted from the updated knowledge graph, and the fine-tuned large language model can be fine-tuned, realizing the iterative fine-tuning of the second large language model.

[0115] In an exemplary embodiment, based on the positive samples and the negative samples, the fine-tuned large language model is iteratively fine-tuned to obtain a second large language model with network fault analysis function, including: based on the entity recognition result corresponding to the negative sample, constructing a prompt word sample; using the prompt word sample, fine-tuning the fine-tuned large language model.

[0116] Optionally, referring to Figure 6 , based on the entity recognition result corresponding to the negative sample, constructing a prompt word sample, and using the prompt word sample to fine-tune the fine-tuned large language model.

[0117] In this embodiment, by constructing a prompt word sample based on the entity recognition result corresponding to the negative sample, fine-tuning the fine-tuned large language model using the prompt word sample, the negative sample can be used to construct a prompt word sample, which is fed back to the fine-tuned large language model, and the iterative fine-tuning of the second large language model is realized.

[0118] In an exemplary embodiment, after the step of generating a network fault solution based on the network fault analysis result by the second large language model, the method further comprises: obtaining a fault handling result after executing the fault handling order; in the case that the network fault triggering the alarm information is not solved, generating feedback information; inputting the feedback information to the first large language model and the second large language model to update the first large language model and the second large language model.

[0119] Wherein, the fault handling result can be that the network fault triggering the alarm information has been solved and no longer appears the alarm information, or the network fault triggering the alarm information has not been solved and still appears the same alarm information.

[0120] Optionally, referring to Figure 3 , after the second large language model generates a network fault solution based on the network fault analysis result, the fault handling result is obtained based on the fault handling order, and in the case that the network fault triggering the alarm information is not solved and still appears the same alarm information, the feedback information is generated, since the first large language model and the second large language model participate in data processing from obtaining the alarm information to generating the network fault solution, inputting the feedback information to the first large language model and the second large language model can make the first large language model and the second large language model more accurate in the next data processing process.

[0121] In this embodiment, the fault handling result after executing the fault handling work order is obtained; the feedback information is generated in the case that the fault handling result represents that the network fault triggering the alarm information is not solved; the feedback information is input to the first large language model and the second large language model to update the first large language model and the second large language model; in this way, the semantic understanding accuracy of the first large language model and the second large language model can be verified according to the fault handling result after executing the fault handling work order, and in the case that the fault handling result represents that the network fault triggering the alarm information is not solved, the fault handling result can be fed back to the first large language model and the second large language model to improve the semantic understanding ability of the first large language model and the second large language model in the next semantic understanding process.

[0122] In actual application, as shown in Figure 7 The large language model can also be used as a system entrance of the network optimization system, understand the user intent, and generate an answer according to the user intent. When the network optimization system receives the user intent, after the intent understanding of the user intent, the network optimization system determines the query operation to be executed, for example, executes the cloud network operation situation index query, enters the big data lake index system to query to obtain a first query result, executes the disaster influence query, enters the knowledge base to query to obtain a second query result, executes the solution query, enters the knowledge base to query and then calls the GIS interface to query to obtain a third query result, and generates an answer based on the first query result, the second query result and the third query result. At the same time, when the network optimization system receives the user intent, after the intent understanding of the user intent, the network optimization system logs in to the 5GR system, queries the poor quality cell, the distributed file and the system data, and generates the corresponding query result to be displayed on the 5GR system.

[0123] In order to facilitate the understanding of those skilled in the art, Figure 8An application architecture of a network optimization system is provided, and the application architecture of the network optimization system includes a data collection and processing module, a knowledge management module, an AI capability training model, a large language model training and fine-tuning module, and an application module. The data collection and processing module is configured to collect data of the network optimization system, data in a data lake, wiki, web pages, white papers, and open datasets, and perform preprocessing to obtain processed data. The knowledge management module is configured to construct a knowledge graph for the network system based on the processed data to obtain a knowledge graph that can be used for knowledge retrieval, intelligent recommendation, intelligent question answering, knowledge extraction, and intelligent evaluation. The knowledge graph can be used to generate corpus data to pretrain and fine-tune a large language model. The AI capability training model is configured to train a traditional AI model using the processed data to obtain various fault analysis models that can be used for fault identification, fault location, intent understanding, text summarization, and trend prediction. The various fault analysis models can be used to work collaboratively with the large language model in a prediction reasoning stage. The large language model training and fine-tuning module is configured to pretrain and fine-tune the large language model using corpus data generated by the knowledge graph, the processed data, and various trained fault analysis models, and to supplement the knowledge graph of the network system with new knowledge generated by the large language model during pretraining and fine-tuning. The application module is configured to use a pre-trained first large language model and a fine-tuned second large language model to perform network optimization, fault identification, intelligent question answering, and intelligent recommendation.

[0124] The network optimization system of the present application combines a large language model with traditional knowledge graphs, knowledge bases, and AI models, thereby improving the utilization rate and application flexibility of various resources, improving the accuracy of fault location in existing network optimization scenarios, enhancing the early warning capability for network faults, strengthening the inheritance of professional knowledge in the network optimization process, and achieving intelligent operation and maintenance.

[0125] In another embodiment, as shown in Figure 9 , a method for generating a network fault solution is provided, which is applied to the server 104 of the network optimization system in Figure 1 , and includes the following steps:

[0126] Step S902, receiving alarm indicator description information sent by each network segment in the network system.

[0127] Step S904, splicing and fusing the alarm indicator description information sent by each network segment to obtain alarm information.

[0128] Step S906, inputting the alarm information into a first large language model to determine a fault network segment in each network segment through the first large language model; the first large language model is obtained by pretraining a general large language model using professional domain knowledge corpus data of each network segment.

[0129] Step S908, using each fault analysis model corresponding to the fault network segment to perform fault analysis on the alarm information, to obtain network fault analysis information output by each fault analysis model.

[0130] Step S910, inputting the network fault analysis information output by each fault analysis model into the second large language model, and generating a network fault analysis result through the second large language model; the second large language model is obtained by fine-tuning the first large language model using the service scenario knowledge corpus data of each network segment.

[0131] Step S912, in a case where the network fault analysis result represents that the alarm information is real alarm information, querying structured data matched with the network fault analysis result through the second large language model calling a knowledge graph and a knowledge base of the network system.

[0132] Step S914, performing semantic understanding on the structured data through the second large language model, and generating a network fault solution; the fault solution is used for assisting in generating a fault handling work order for the alarm information.

[0133] It should be noted that the specific limitations of the above steps can be referred to the specific limitations of the method for generating a network fault solution described above.

[0134] It should be understood that, although each step in the flowchart involved in each embodiment described above is displayed in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in the flowchart involved in each embodiment described above can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or steps or stages in other steps.

[0135] Based on the same inventive concept, the embodiments of the present application also provide a network fault solution generation apparatus for implementing the network fault solution generation method described above. The implementation scheme for solving the problem provided by the apparatus is similar to the implementation scheme described in the above method, and therefore the specific limitations in one or more network fault solution generation apparatus embodiments provided below can be referred to the limitations of the network fault solution generation method described above, which will not be repeated here.

[0136] In one exemplary embodiment, as Figure 10As shown, a network fault solution generation apparatus is provided, applied to a network system including network segments, and the network fault solution generation apparatus includes an alarm information acquisition module 1002, an analysis information acquisition module 1004, an analysis result generation module 1006, and a solution generation module 1008, wherein:

[0137] The alarm information acquisition module 1002 is configured to acquire alarm information of the network system, and input the alarm information to a first large language model to determine a fault network segment in each network segment by the first large language model; the first large language model is obtained by pre-training a general large language model by using professional domain knowledge corpus data of each network segment;

[0138] The analysis information acquisition module 1004 is configured to perform fault analysis on the alarm information by using each fault analysis model corresponding to the fault network segment to obtain network fault analysis information output by each fault analysis model;

[0139] The analysis result generation module 1006 is configured to input the network fault analysis information output by each fault analysis model to a second large language model to generate a network fault analysis result by the second large language model; the second large language model is obtained by fine-tuning the first large language model by using business scenario knowledge corpus data of each network segment;

[0140] The solution generation module 1008 is configured to generate a network fault solution based on the network fault analysis result by the second large language model in a case where the network fault analysis result indicates that the alarm information is real alarm information; the network fault solution is used to assist in generating a fault handling work order for the alarm information.

[0141] In one embodiment, the alarm information acquisition module 1002 is specifically configured to receive alarm indicator description information sent by each network segment in the network system; and splice and fuse the alarm indicator description information sent by each network segment to obtain the alarm information.

[0142] In one embodiment, the solution generation module 1008 is specifically configured to call a knowledge graph and a knowledge base of the network system by the second large language model, query structured data matched with the network fault analysis result, and generate the network fault solution by performing semantic understanding on the structured data by the second large language model.

[0143] In one of the embodiments, the apparatus further comprises a training module configured to obtain full-volume corpus data, and construct a knowledge graph of the network system based on the full-volume corpus data; extract professional domain knowledge corpus data of each network segment and business scenario knowledge corpus data of each network segment based on the knowledge graph of the network system; pre-train the general large language model using the professional domain knowledge corpus data of each network segment to obtain a first large language model with a fault network segment positioning function; fine-tune the first large language model using the business scenario knowledge corpus data of each network segment to obtain a second large language model with a network fault analysis function.

[0144] In one of the embodiments, the training module is further configured to fine-tune the first large language model using the business scenario knowledge corpus data of each network segment to obtain a fine-tuned large language model; perform entity recognition on the full-volume corpus data using the fine-tuned large language model to obtain each entity recognition result; obtain expert review results for each entity recognition result, and separate positive samples and negative samples from each entity recognition result based on the expert review results; and iteratively fine-tune the fine-tuned large language model based on the positive samples and the negative samples to obtain the second large language model with the network fault analysis function.

[0145] In one of the embodiments, the training module is further configured to update the knowledge graph based on the entity recognition result corresponding to the positive sample to obtain an updated knowledge graph; re-extract the business scenario knowledge corpus data of each network segment from the updated knowledge graph, and fine-tune the fine-tuned large language model using the re-extracted business scenario knowledge corpus data of each network segment.

[0146] In one of the embodiments, the training module is further configured to construct a prompt word sample based on the entity recognition result corresponding to the negative sample; and fine-tune the fine-tuned large language model using the prompt word sample.

[0147] In one of the embodiments, the apparatus further comprises a feedback module configured to obtain a fault handling result after executing the fault handling order; generate feedback information in a case where the fault handling result represents that the network fault triggering the alarm information has not been solved; and input the feedback information to the first large language model and the second large language model to update the first large language model and the second large language model.

[0148] Each of the modules in the network fault solution generation apparatus described above can be realized by software, hardware, and combinations thereof, in whole or in part. Each of the modules described above can be embedded in or independent of a processor in a computer device in hardware form, or stored in a memory in a computer device in software form, so as to be called and executed by a processor to perform operations corresponding to each of the modules.

[0149] In an example embodiment, a computer device, which can be a server, is provided, and an internal structure diagram of the computer device can be as shown in FIG. 1. Figure 11 The computer device includes a processor, a memory, an input / output interface, and a communication interface. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store generation data of a network fault solution. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with terminals outside through a network connection. The computer program is executed by the processor to implement a network fault solution generation method.

[0150] Those skilled in the art can understand that Figure 11 The structure shown in FIG. 1 is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0151] In an example embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program. The computer program is executed by the processor to cause the processor to perform the steps of the network fault solution generation method. The steps of the network fault solution generation method can be the steps in the network fault solution generation method of each of the above embodiments.

[0152] In an example embodiment, a computer readable storage medium is provided, which stores a computer program. The computer program is executed by the processor to cause the processor to perform the steps of the network fault solution generation method. The steps of the network fault solution generation method can be the steps in the network fault solution generation method of each of the above embodiments.

[0153] In an example embodiment, a computer program product is provided, which includes a computer program. The computer program is executed by the processor to cause the processor to perform the steps of the network fault solution generation method. The steps of the network fault solution generation method can be the steps in the network fault solution generation method of each of the above embodiments.

[0154] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0155] The technical features of the above embodiments can be combined arbitrarily. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0156] The above-described embodiments are merely illustrative of several embodiments of the present application, and the description is relatively specific and detailed, but should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A method of generating a network fault resolution solution, the method comprising: The method is applied to a network system comprising network segments, and comprises the following steps: obtaining alarm information of the network system and inputting the alarm information into a first large language model to determine a fault network segment in each network segment by the first large language model; the first large language model is obtained by pre-training a general large language model using professional domain knowledge corpus data of each network segment; performing fault analysis on the alarm information using each fault analysis model corresponding to the fault network segment to obtain network fault analysis information output by each fault analysis model; inputting the network fault analysis information output by each fault analysis model into a second large language model to generate a network fault analysis result by the second large language model; the second large language model is obtained by fine-tuning the first large language model using business scenario knowledge corpus data of each network segment; in a case where the network fault analysis result represents that the alarm information is real alarm information, generating a network fault solution based on the network fault analysis result by the second large language model; the network fault solution is used to assist in generating a fault handling work order for the alarm information.

2. The method of claim 1, wherein, The method further comprises the following steps: receiving alarm indicator description information sent by each network segment in the network system; splicing and fusing the alarm indicator description information sent by each network segment to obtain the alarm information.

3. The method of claim 1, wherein, The method further comprises the following steps: calling a knowledge graph and a knowledge base of the network system by the second large language model to query structured data matched with the network fault analysis result; performing semantic understanding on the structured data by the second large language model to generate the network fault solution.

4. The method of claim 1, wherein, The method further comprises the following steps: obtaining full corpus data and constructing a knowledge graph of the network system based on the full corpus data; extracting professional domain knowledge corpus data of each network segment and business scenario knowledge corpus data of each network segment based on the knowledge graph of the network system; pre-training the general large language model using the professional domain knowledge corpus data of each network segment to obtain the first large language model having a fault network segment positioning function; fine-tuning the first large language model using the business scenario knowledge corpus data of each network segment to obtain the second large language model having a network fault analysis function.

5. The method of claim 4, wherein, The method further comprises the following steps: fine-tuning the first large language model using the business scenario knowledge corpus data of each network segment to obtain a fine-tuned large language model; performing entity recognition on full corpus data using the fine-tuned large language model to obtain each entity recognition result; obtaining expert review results for each entity recognition result and separating positive samples and negative samples from each entity recognition result based on the expert review results. Based on the positive samples and the negative samples, the fine-tuned large language model is iteratively fine-tuned to obtain the second large language model with the network fault analysis function.

6. The method of claim 5, wherein, The method further comprises: Based on the entity recognition results corresponding to the positive samples, the knowledge graph is updated to obtain an updated knowledge graph; From the updated knowledge graph, the business scenario knowledge corpus data of each network segment is re-extracted, and the fine-tuned large language model is fine-tuned using the re-extracted business scenario knowledge corpus data of each network segment.

7. The method of claim 5, wherein, The method further comprises: Based on the entity recognition results corresponding to the negative samples, prompt word samples are constructed; The fine-tuned large language model is fine-tuned using the prompt word samples.

8. The method of claim 1, wherein, After the step of generating a network fault solution based on the network fault analysis result by the second large language model, the method further comprises: Obtaining a fault handling result after executing the fault handling work order; In the case where the fault handling result indicates that the network fault triggering the alarm information has not been solved, feedback information is generated; The feedback information is input into the first large language model and the second large language model to update the first large language model and the second large language model.

9. An apparatus for generating a network fault resolution solution, the apparatus comprising: The device is applied to a network system, and the network system comprises network segments. An alarm information obtaining module is configured to obtain alarm information of the network system and input the alarm information into a first large language model to determine a fault network segment in each network segment by the first large language model; the first large language model is obtained by pre-training a general large language model using professional domain knowledge corpus data of each network segment; An analysis information obtaining module is configured to perform fault analysis on the alarm information using each fault analysis model corresponding to the fault network segment to obtain network fault analysis information output by each fault analysis model; An analysis result generating module is configured to input the network fault analysis information output by each fault analysis model into a second large language model to generate a network fault analysis result by the second large language model; the second large language model is obtained by fine-tuning the first large language model using business scenario knowledge corpus data of each network segment; A solution generating module is configured to generate a network fault solution based on the network fault analysis result by the second large language model in the case where the network fault analysis result indicates that the alarm information is real alarm information; the network fault solution is used to assist in generating a fault handling work order for the alarm information. 10.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-9. The processor executes the computer program to implement the steps of the method of any one of claims 1 to 8.

11. A computer readable storage medium having stored thereon a computer program, characterized in that The computer program, which when executed by a processor, implements the steps of the method of any one of claims 1 to 8.

12. A computer program product comprising a computer program, characterized in that, The computer program, which when executed by a processor, implements the steps of the method of any one of claims 1 to 8.

Images