Network fault operation and maintenance method and device, equipment and storage medium

Abstract

The invention provides a network fault operation and maintenance method and device, equipment and a storage medium, and relates to the technical field of artificial intelligence. In some embodiments of the invention, an obtained input text containing a fault operation and maintenance service request statement is subjected to vectorization processing to obtain an input text vector; determining a user operation and maintenance demand vector according to the input text vector; performing semantic reasoning operation on the user operation and maintenance demand vector to obtain a structured code; performing field standardization operation on the structured code to obtain a downstream application service calling parameter template capable of being called by a downstream application service; requesting the downstream application service through the API port according to the downstream application service calling parameter template, and receiving code text content returned by the downstream application service; determining a target operation and maintenance suggestion statement according to the code text content; the intelligent degree of network fault operation and maintenance is improved, and the network fault operation and maintenance efficiency is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to a network fault operation and maintenance method, device, equipment and storage medium. Background Art

[0002] In recent years, with the rapid development of 5G network communications, the scale of network elements and the depth of network layers in network infrastructure have increased rapidly.

[0003] In the operation and maintenance work, it is necessary to deal with various faults generated in the network in real time. In the current network business practice, the daily processing of fault data is huge, and there are many sudden faults. When a single network element device fails, the network element devices associated with its topology may also cause associated secondary faults, resulting in large-scale network element decommissioning in a short period of time, and generating large-scale cascading alarms. On the other hand, due to the full cloudification of the 5G core network, the faults generated by network cloud devices will directly lead to the performance degradation and service failure of the 5G core network, which may have a serious impact on the network. The difficulty faced by network cloud operation and maintenance work is that it not only includes physical layer servers, but also nests multiple layers of virtual machines and lower-layer container applications. The occurrence of hardware failures in a cloud server often triggers multi-level and large-scale alarm events from the bottom hardware alarm to the core network application layer alarm, resulting in the superposition of alarm information at multiple network levels and wide topology scales in a short period of time, making it difficult to determine the cause of the fault corresponding to the root alarm. Therefore, operation and maintenance personnel need more accurate and intelligent fault operation and maintenance auxiliary equipment to carry out ultra-large-scale network fault operation and maintenance work in the 5G era.

[0004] At present, the intelligence level of network fault operation and maintenance is low and the efficiency is low. Summary of the invention

[0005] The present disclosure provides a network fault operation and maintenance method, apparatus, device and storage medium to at least solve the problem of low intelligence and low efficiency of existing network fault operation and maintenance.

[0006] The technical solution of the present disclosure is as follows:

[0007] The present disclosure provides a network fault operation and maintenance method, including:

[0008] Vectorize the input text containing the fault operation and maintenance service request statement to obtain an input text vector;

[0009] Determining a user operation and maintenance requirement vector according to the input text vector;

[0010] Performing semantic reasoning operations on the user operation and maintenance requirement vector to obtain a structured code;

[0011] Performing field standardization operation on the structured code to obtain a downstream application service calling parameter template that can be called by the downstream application service;

[0012] According to the downstream application service calling parameter template, requesting the downstream application service through the API port, and receiving the code text content returned by the downstream application service;

[0013] According to the code text content, a target operation and maintenance suggestion statement is determined.

[0014] Optionally, determining a user operation and maintenance requirement vector according to the input text vector includes:

[0015] Matching the input text vector with the candidate fault operation and maintenance knowledge vectors in the fault operation and maintenance knowledge base to obtain a successfully matched target fault operation and maintenance knowledge vector;

[0016] Matching the input text vector with the candidate prompt word vectors in the prompt word library to obtain a successfully matched target prompt word vector;

[0017] The input text vector, the target fault operation and maintenance knowledge vector and the target prompt word vector are concatenated to obtain a user operation and maintenance demand vector.

[0018] Optionally, matching the input text vector with a candidate fault operation and maintenance knowledge vector in a fault operation and maintenance knowledge base to obtain a successfully matched target fault operation and maintenance knowledge vector includes:

[0019] Determine a first vector similarity between the input text vector and each candidate fault operation and maintenance knowledge vector in the fault operation and maintenance knowledge base;

[0020] Selecting the first k target fault operation and maintenance knowledge vectors from the candidate fault operation and maintenance knowledge vectors in descending order of the similarity of the first vectors;

[0021] The step of matching the input text vector with the candidate prompt word vectors in the prompt word library to obtain a successfully matched target prompt word vector includes:

[0022] Determining a second vector similarity between the input text vector and each candidate prompt word vector in the prompt word library;

[0023] In descending order of the similarity of the second vectors, the first k target prompt word vectors are selected from the candidate prompt word vectors, where k is a positive integer.

[0024] Optionally, performing a semantic reasoning operation on the user operation and maintenance requirement vector to obtain a structured code includes:

[0025] Performing intent extraction on the user operation and maintenance demand vector to obtain user intent;

[0026] Generate an operation and maintenance action corresponding to the user intention;

[0027] According to the intelligent operation and maintenance operation rules, the set keywords in the operation and maintenance action are removed to obtain the audited operation and maintenance action;

[0028] A semantic escape operation is performed on the audited operation and maintenance action to obtain a structured code.

[0029] Optionally, extracting the user operation and maintenance demand vector to obtain the user intent includes:

[0030] The user operation and maintenance requirements are input into a pre-trained deep learning model, the user operation and maintenance requirement vector is encoded, and the user intention is extracted using the attention weight.

[0031] Optionally, determining a target operation and maintenance suggestion statement according to the code text content includes:

[0032] Vectorizing the code text content to obtain a preliminary operation and maintenance suggestion statement;

[0033] Generate a fault solution statement according to the preliminary operation and maintenance suggestion statement and the user operation and maintenance requirement vector;

[0034] The fault solution statement is input into a large language model for semantic enhancement to obtain a target operation and maintenance suggestion statement.

[0035] Optionally, generating a fault solution statement according to the preliminary operation and maintenance suggestion statement and the user operation and maintenance requirement vector includes:

[0036] Determine the similarity between the preliminary operation and maintenance suggestion statement and the user operation and maintenance requirement vector according to a similarity algorithm;

[0037] According to the similarity, a fault solution statement matching the input text vector is determined.

[0038] The present disclosure also provides a network fault operation and maintenance device, including:

[0039] A vectorization module is used to vectorize the input text containing the fault operation and maintenance business request statement to obtain an input text vector;

[0040] A first determination module, configured to determine a user operation and maintenance requirement vector according to the input text vector;

[0041] A semantic reasoning module, used for performing semantic reasoning operations on the user operation and maintenance demand vector to obtain a structured code;

[0042] A standardization module, used for performing field standardization operation on the structured code to obtain a downstream application service calling parameter template that can be called by a downstream application service;

[0043] A request module, used to call a parameter template according to the downstream application service, request the downstream application service through an API port, and receive the code text content returned by the downstream application service;

[0044] The second determination module is used to determine a target operation and maintenance suggestion statement according to the code text content.

[0045] The present disclosure also provides an electronic device, including:

[0046] processor;

[0047] a memory for storing instructions executable by the processor;

[0048] The processor is configured to execute the instructions to implement each step in the above method.

[0049] The disclosed embodiment further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements each step of the above method when executed by a processor.

[0050] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects:

[0051] In some embodiments of the present disclosure, the input text obtained and containing the fault operation and maintenance business request statement is vectorized to obtain an input text vector; based on the input text vector, a user operation and maintenance demand vector is determined; semantic reasoning operations are performed on the user operation and maintenance demand vector to obtain a structured code; field standardization operations are performed on the structured code to obtain a downstream application service call parameter template that can be called by the downstream application service; based on the downstream application service call parameter template, a downstream application service is requested through an API port, and the code text content returned by the downstream application service is received; based on the code text content, a target operation and maintenance suggestion statement is determined; the present disclosure obtains the user operation and maintenance demand vector based on the input text vector, performs semantic reasoning operations on the user operation and maintenance demand vector, and improves the ability to recognize operation and maintenance intentions; based on the downstream application service call parameter template and the code text content returned by the downstream application service, the target operation and maintenance suggestion statement is determined, and the network fault analysis extrapolation capability is improved to improve the intelligence level of network fault operation and maintenance and improve the efficiency of network fault operation and maintenance.

[0052] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute improper limitations on the present disclosure.

[0054] Figure 1 A flowchart of a network fault operation and maintenance method provided by an exemplary embodiment of the present disclosure;

[0055] Figure 2 A schematic diagram of a network fault operation and maintenance device provided for an exemplary embodiment of the present disclosure;

[0056] Figure 3 A schematic diagram of the structure of a network fault operation and maintenance device provided for an exemplary embodiment of the present disclosure;

[0057] Figure 4 A schematic structural diagram of an electronic device provided by an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0058] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0059] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure.

[0060] It should be noted that the user information involved in this disclosure includes but is not limited to: user device information and user personal information; the collection, storage, use, processing, transmission, provision and disclosure of user information in this disclosure comply with the relevant laws and regulations and do not violate public order and good morals.

[0061] With the rapid development of artificial intelligence technology, the use of communication operation and maintenance question and answer systems to implement fault operation and maintenance has become an important technical exploration in the operation and maintenance industry. The communication operation and maintenance question and answer system is a program that can quickly and accurately provide answers based on the user's natural language questions. At present, the most common intelligent question and answer technology is based on the Transform natural language neural network model built based on deep learning and knowledge graph retrieval. The core is to identify the intent of the question through semantic understanding, and then access the relevant knowledge graph database through the intent, and then obtain the similarity of the question and answer pairs based on the relevance of the knowledge graph. The best answer is selected based on the similarity combined with other rules.

[0062] The existing network fault operation and maintenance methods have the following problems:

[0063] On the one hand, the intent recognition capability is low, there are few application scenarios, and there is a lack of flexibility.

[0064] Existing fault operation and maintenance auxiliary question and answer mainly uses keyword extraction and probability matching to identify user sessions, or establishes fixed business question and answer pairs and then matches user content with existing question and answer pairs. None of the above methods can accurately identify the user's true intention from the user session, nor can they infer how to call, query, and recommend users to access the corresponding downstream application services based on multiple intentions such as user inquiries, questions, and help.

[0065] On the other hand, the dialogue has low intelligence and insufficient extrapolation capabilities, which cannot meet the needs of network device queries with complex topology structures.

[0066] Existing conversational intelligent operation and maintenance devices can only provide simple question-answer matching functions, and are highly dependent on the coverage of knowledge graphs. They also have low matching and success rates for downstream service calls, and provide limited assistance to operation and maintenance personnel, making it difficult to achieve fast, accurate, and large-scale processing. For example, when a user asks "query the cloud storage pool for machines with abnormal performance within 15 minutes", traditional intelligent conversational devices cannot accurately extract the resources, time series, and alarm patterns corresponding to key information such as "cloud storage pool", "within 15 minutes", and "performance abnormalities", and connect them in series to infer and conduct related queries.

[0067] In response to the above technical problems, in some embodiments of the present disclosure, the input text obtained and containing the fault operation and maintenance business request statement is vectorized to obtain an input text vector; based on the input text vector, a user operation and maintenance demand vector is determined; semantic reasoning operations are performed on the user operation and maintenance demand vector to obtain a structured code; field standardization operations are performed on the structured code to obtain a downstream application service call parameter template that can be called by the downstream application service; based on the downstream application service call parameter template, a downstream application service is requested through an API port, and the code text content returned by the downstream application service is received; based on the code text content, a target operation and maintenance suggestion statement is determined; the present disclosure obtains the user operation and maintenance demand vector based on the input text vector, performs semantic reasoning operations on the user operation and maintenance demand vector, and improves the ability to recognize operation and maintenance intentions; based on the downstream application service call parameter template and the code text content returned by the downstream application service, the target operation and maintenance suggestion statement is determined, and the network fault analysis and extrapolation capability is improved, so as to improve the intelligence level of network fault operation and maintenance, and improve the efficiency of network fault operation and maintenance.

[0068] The technical solutions provided by various embodiments of the present disclosure are described in detail below in conjunction with the accompanying drawings.

[0069] Figure 1 The following is a flow chart of a network fault operation and maintenance method provided by an exemplary embodiment of the present disclosure. Figure 1 As shown, the method includes:

[0070] S101: vectorizing the acquired input text containing the fault operation and maintenance service request statement to obtain an input text vector;

[0071] S102: Determine a user operation and maintenance requirement vector according to the input text vector;

[0072] S103: Perform semantic reasoning on the user operation and maintenance requirement vector to obtain a structured code;

[0073] S104: performing field standardization operations on the structured code to obtain a downstream application service calling parameter template that can be called by the downstream application service;

[0074] S105: calling the parameter template of the downstream application service, requesting the downstream application service through the API port, and receiving the code text content returned by the downstream application service;

[0075] S106: Determine a target operation and maintenance suggestion statement according to the code text content.

[0076] In this embodiment, the execution subject of the above method may be a terminal device or a server.

[0077] Among them, the terminal device includes but is not limited to a mobile station (MS), a mobile terminal, a mobile phone, a handset and a portable equipment, etc. The terminal device can communicate with one or more core networks via a radio access network (RAN). For example, the terminal device can be a mobile phone (or a "cellular" phone), a computer with a wireless communication function, etc. The terminal device can also be a computer with a wireless transceiver function, a virtual reality (VR) terminal device, an AR terminal device, a wireless terminal in industrial control, a wireless terminal in self driving, a wireless terminal in remote medical, a wireless terminal in a smart grid, a wireless terminal in transportation safety, a wireless terminal in a smart city, a wireless terminal in a smart home, etc., and the operating system installed on the terminal device includes but is not limited to: IOS, Android, windows, linux, Mac OS and other operating systems. In different networks, the terminal may be called different names, such as user equipment, mobile station, subscriber unit, station, cellular phone, personal digital assistant, wireless modem, wireless communication device, handheld device, laptop computer, cordless phone, wireless local loop station, television, etc. For the convenience of description, it is referred to as terminal equipment in this embodiment.

[0078] In this embodiment, the implementation form of the server is not limited. For example, the server can be a conventional server, a cloud server, a cloud host, a virtual center, or other server devices. The server mainly includes a processor, a hard disk, a memory, a system bus, etc., and a general computer architecture type.

[0079] The present invention provides a network fault operation and maintenance method based on big model technology, which provides front-line operation and maintenance personnel with operation and maintenance auxiliary capabilities such as status query, business suggestions, and troubleshooting, and improves the communication operation and maintenance personnel's intention recognition ability, fault analysis and extrapolation ability, and fault handling efficiency in troubleshooting.

[0080] Figure 2 Schematic diagram of a network fault operation and maintenance device 20 provided by an exemplary embodiment of the present disclosure. Figure 2As shown, the network fault operation and maintenance device 20 includes: a knowledge integration module 21, a large model driving module 22 and a conclusion enhancement module 23. The present disclosure first obtains the query question input by the user through the knowledge integration module to vectorize the query question, then obtains the question text vector, and compares and matches the question text vector with the text vector in the knowledge base, and then determines whether the knowledge base can meet the accurate knowledge provision of the current query question, and then through the data integration operation, the knowledge base can meet the relevant knowledge of the current query question (that is, the question vector and the text vector in the knowledge base are successfully matched), and finally, according to the matching result, the knowledge text vector is obtained, combined with the query question and the question prompt word, and sent to the large model driving module, and sent to the large model for processing to obtain the final question and answer result. The present disclosure adopts the method of plug-in and embedded knowledge base, which effectively improves the flexibility and memory depth of the large model dialogue ability to professional operation and maintenance knowledge, and enhances the generalization ability of the troubleshooting means for various operation and maintenance scenarios. It solves the disadvantages of the existing operation and maintenance system with low intention recognition ability, insufficient extrapolation ability, limited to question and answer pairs, and low dialogue content matching rate.

[0081] In order to solve the problem of insufficient extrapolation capabilities of traditional network fault operation and maintenance knowledge, the network fault operation and maintenance device disclosed in the present invention uses a pre-trained base large model with certain basic logical capabilities embedded in it to ensure accurate responses to query questions when the capacity of the external knowledge base is limited. Under the premise of optimizing the external knowledge base, the question-and-answer assistance capability of zero-sample or few-sample (Zero-Shot or Few-Shots) operation and maintenance knowledge can be achieved, ensuring that each question and answer query of the operation and maintenance personnel receives a relatively accurate and reasonable response. The final conclusion enhancement module of the disclosed device is used to match troubleshooting solutions and enhance the logic of the output language, which improves the readability and usability of fault operation and maintenance assistance responses.

[0082] The present disclosure also has a fault operation and maintenance expansion capability that is compatible with an external application interface on the above-mentioned basic device. The implementation method is that the large model output text conclusion in this device can provide data transmission and parameter calling capabilities to other application services within the network management system through a pre-established downstream application service adaptation interface template. It enables operation and maintenance personnel to control other application services in the internal system of the network management operation and maintenance domain through dialogue, solving the problem that the existing network management operation and maintenance system can only provide simple questions and answers, but cannot call and drive the existing downstream application resources and capabilities through questions and answers, realizing "AI replacement", improving the degree of network automation capabilities, and providing ideas and evolution directions for the automatic operation and maintenance capabilities of the future new generation of communication networks.

[0083] In this embodiment, the knowledge integration module obtains the input text input by the user containing the fault operation and maintenance business request statement, vectorizes the input text and related input prompt words and fault operation and maintenance knowledge, and integrates them into a text vector to achieve high-dimensional embedding of plug-in knowledge.

[0084] In some embodiments of the present disclosure, the knowledge integration module determines the user operation and maintenance demand vector based on the input text vector. One achievable method is to match the input text vector with the candidate fault operation and maintenance knowledge vectors in the fault operation and maintenance knowledge base to obtain a successfully matched target fault operation and maintenance knowledge vector; match the input text vector with the candidate prompt word vectors in the prompt word base to obtain a successfully matched target prompt word vector; concatenate the input text vector, the target fault operation and maintenance knowledge vector and the target prompt word vector to obtain the user operation and maintenance demand vector. Among them, firstly, the input text input by the user containing the fault operation and maintenance business request statement is obtained, the input text is vectorized to obtain the input text vector, and then the fault operation and maintenance knowledge vector search and the prompt word vector search are performed. Next, extract the matching between the input text vector and each candidate fault operation and maintenance knowledge vector contained in the fault operation and maintenance knowledge base, that is, calculate the similarity between the input text vector and each candidate fault operation and maintenance knowledge vector respectively, select the first k target fault operation and maintenance knowledge vectors from the candidate fault operation and maintenance knowledge vectors in the order of the first vector similarity from large to small, and associate the keyword with the largest matching degree with the fault operation and maintenance knowledge with the first k matching degrees. The knowledge association step is: according to the input text entered by the user, convert the input text into a numerical vector, that is, an input text vector, which can be of fixed length or variable length; usually the user request does not exceed 2048 characters, but the data structure length of the numerical vector can be increased or reduced according to the use needs. The search step is: perform a similarity search of the input text entered by the user with the words contained in the fault operation and maintenance knowledge base and the prompt word library using the character metric distance method, and return the first k results of the similarity search, where k can be 3-15 according to the matching effect. Finally, the input text vector, target fault operation and maintenance knowledge vector, and target prompt word vector are integrated and input into the knowledge integration module to be spliced ​​into the user operation and maintenance demand vector and input into the next step, namely the large model driving module.

[0085] The following is a detailed description of the processing flow of the knowledge integration module using a specific embodiment. In a specific use case, the fault operation and maintenance knowledge vector search can be retrieved based on the input text. Take the user input "network cloud database service performance degradation" as an example.

[0086] The input text is converted into a 2048-bit input text vector through methods such as word bags and word libraries. Then, similarity search is performed through the knowledge base to match the first 6 corpus vectors such as "network cloud database performance", "cloud server performance inspection process", "cloud server hardware troubleshooting steps", "cloud server account query", and "cloud server process inspection process", which are the target fault operation and maintenance knowledge. And through the prompt word vector, task prompt words such as "hardware equipment monitoring" + "process log query" + "operation steps" + "operation and maintenance plan" are retrieved, which are the target prompt words.

[0087] The knowledge integration module built through the LLM large language model fine-tuning task matches the k results that best match the user's input sentence through lexical attention and semantic similarity, and excludes the corpus vectors generated in the previous step that are irrelevant to the user's input sentence, such as "cloud server account query". The result of LLM knowledge integration is a natural sentence that describes the user's demand process and consists of clear nouns, verbs and descriptive attributes. In this example, "hardware device monitoring", "process log query" and "operation steps" are used as prompt templates for solving the user's request for "network cloud database service performance degradation", and "cloud server performance inspection process", "cloud server hardware troubleshooting steps" and "cloud server process inspection process" are used as the knowledge source of the user's request.

[0088] Therefore, the output result of the knowledge integration module can be similar to: "There is currently a performance degradation problem with the network cloud database service. Please query the cloud server performance inspection process, cloud server hardware troubleshooting steps, cloud server process inspection process, and perform downstream queries to determine whether cloud server hardware replacement operations are required and describe the operation steps for such hardware equipment replacement; and whether cloud server process intervention operations are required and describe the operation steps for such service process debugging and cutover."

[0089] In this embodiment, the big model driving module is used to process, infer, and summarize the user operation and maintenance demand vector output by the knowledge integration module. The big model driving module generates an application service call template that can be called by downstream applications through steps such as intent extraction, content review, and dictionary escape, thereby realizing the ability of the big model to call according to user intent and fault scenarios, and request systems and external application services related to the network management operation and maintenance domain and the fault business.

[0090] In some embodiments of the present disclosure, semantic reasoning operations are performed on user operation and maintenance demand vectors to obtain structured codes. One achievable method is to perform intent extraction on user operation and maintenance demand vectors to obtain user intent; generate operation and maintenance actions corresponding to user intent; remove set keywords in the operation and maintenance actions according to intelligent operation and maintenance operation rules to obtain audited operation and maintenance actions; perform semantic escape operations on the audited operation and maintenance actions to obtain structured codes. Among them, the user operation and maintenance demand is input into a pre-trained deep learning model, the user operation and maintenance demand vector is encoded, and the user intent is extracted using attention weights.

[0091] In the above embodiment, the user operation and maintenance demand vector is subjected to intention extraction to obtain the user intention; and the operation and maintenance action corresponding to the user intention is generated. Among them, the intention is extracted according to the prompt words, input text and fault operation and maintenance knowledge transmitted by the knowledge integration module, and the user intention corresponding to the quantified text is extracted through deep neural network structures including but not limited to Transformer and attention weight algorithms, so as to realize knowledge extrapolation reasoning and generate operation and maintenance actions matching the user intention. Among them, the method of intention extraction is to integrate the input prompt words, user input and fault operation and maintenance knowledge through the attention weight of LLM according to the deep learning model such as Transformer architecture, combined with the attention mechanism (Attention Mechanism), and map the text sequence to the vector representation of the high-dimensional space by using the encoder-decoder framework of the model, and apply the sequence to sequence (Seq2Seq) model or the specific intention recognition algorithm to perform knowledge extrapolation, and organize the operation and maintenance prompt words, user query statements and fault operation and maintenance knowledge statements matched according to the attention into natural language text segments, so as to realize the operation and maintenance action suggestion matching of the user intention.

[0092] In some embodiments of the present disclosure, according to the intelligent operation and maintenance rules, the set keywords in the operation and maintenance actions are removed to obtain the audited operation and maintenance actions. The set keywords are removed by the pre-formulated intelligent operation and maintenance rules, including but not limited to standardized processes, expert experience, troubleshooting manuals, manufacturer technical documents, etc., to audit the operation and maintenance actions generated in the previous step, and prevent the operation and maintenance actions from containing requests that may cause invalid operations.

[0093] Among them, the intelligent operation and maintenance operation rules are built-in knowledge bases containing sensitive words such as dangerous operations, safety hazards, and privacy information based on standardized processes, troubleshooting manuals, and technical documentation rules, which are used to review and filter the generated operation and maintenance actions. The specific implementation method is to use keyword matching and pattern recognition algorithms, and use confidence scoring or verification mechanisms to identify and remove words in the text segment of the operation and maintenance intention generated in the previous step whose confidence scores do not meet the threshold, so as to avoid possible illegal or unsafe operation and maintenance action requests.

[0094] In some embodiments of the present disclosure, semantic escape operation is performed on the audited operation and maintenance actions to obtain structured code. Among them, the dictionary escape module converts the operation and maintenance action information in the form of numerical vectors into a data form including but not limited to JSON fields containing structured data and natural language function descriptions using forms including but not limited to dictionaries, word bags, vector databases, etc., to achieve semantic escape. Specifically, the method of the dictionary escape module is to use prompt word technology to guide the pre-trained large language model, and convert the above-mentioned operation and maintenance action information mentioned in text into structured code in JSON or other data exchange formats through attention matching and other methods, including execution code, encapsulated structured data, and code comments described in natural language.

[0095] In some embodiments disclosed herein, the structured code is subjected to field standardization operations to obtain a downstream application service call parameter template that can be called by the downstream application service. Among them, according to the escaped structured operation and maintenance data, the corresponding downstream application service call parameter template is generated through field standardization operations, and is input into the next step, that is, the conclusion enhancement module to perform application calls and requests. Specifically, the downstream application service call parameter template is a pre-arranged API call or service orchestration rule template. In this module, according to the structured code obtained in the previous step, the parameter fields involved in the structured code are mapped to the standardized parameter field dictionary of the pre-arranged rules through the combination of large model prompts and fine-tuning and the like, combined with the pre-arranged API call or service orchestration rule template. The purpose of the above operation is to standardize the parameter transmission and calling methods, so that the parameters of the execution code can be smoothly passed to the corresponding application service, triggering the subsequent automated operation and maintenance action execution and calling of the downstream application service.

[0096] The processing flow of the large model driving module is described in detail below with reference to a specific embodiment.

[0097] A cloud service data center received feedback from downstream users that its cloud database service had performance degradation issues.

[0098] Step 1: The large model intent extraction module extracts intent from user feedback on the cloud database. The user submits a question through the customer support system: "The database query response time suddenly slows down." and receives the prompt word template input by the knowledge integration module: "Database", "Database performance", "Database query response event". The user inputs: "The database query response time suddenly slows down", and the relevant information in the fault operation and maintenance knowledge base, including text sentences such as "Database performance" and "Database response slows down" in the "Database Operation and Maintenance Manual". Use a pre-trained Transformer model, such as BERT or GPT, to encode the input text and use attention weights to capture key information. The model decoder generates an intent representation in vector form and infers the user intent through a Seq2Seq model or intent recognition algorithm: "The user needs to troubleshoot the problem of database performance degradation." Generate operation and maintenance action suggestions, including: "Check the database CPU usage" and "Analyze the slow query log."

[0099] Step 2: Intelligent O&M action review. O&M actions go through an intelligent review system that has built-in standardized processes, expert experience, troubleshooting manuals, and technical documentation. The review system uses keyword matching and pattern recognition algorithms, and through a confidence scoring mechanism, removes or modifies O&M actions that do not meet security thresholds to ensure that all recommendations are safe and compliant. Check whether the recommended O&M actions contain sensitive operations, such as "restarting the database service" which may be considered a potential high-risk operation and needs to be removed from the O&M actions. In this use case, "restarting the database service" is not included, so there is no need to perform the high-risk operation removal process.

[0100] Step 3: Convert the instruction into execution code according to the dictionary escape module. Audited operation and maintenance actions, including but not limited to "check database CPU usage" and "check database memory usage", are converted by the dictionary escape module. This module is implemented by using a predefined dictionary or bag-of-words model to convert the text description into structured code, such as JSON format: {"action":"check_cpu_usage", "check_ram_usage", "target":"database"}. The structured code may include but not be limited to execution code, data, and code comments described in natural language for subsequent processing.

[0101] Step 4: Call the downstream application services related to operation and maintenance according to the generated call template module. The function of this module is to structure the operation and maintenance data as API call parameters and standardize them through the downstream application service call parameter template. The implementation method is to use the pre-arranged API call template to map the parameter fields to the standardized parameter fields of the downstream service. Specifically, the "check_cpu_usage" and "check_ram_usage" corresponding to the previous step correspond to the "check_15min_cpu_usage" and "check_15min_ram_usage" with specific meaning instructions in the downstream service, workbench, and control platform in the application service call parameter template. In addition, according to the API call template, the cloud server accessed by the user can be mapped to the cloud server referred to by a specific parameter, such as "cloud_dbase25_monitoring". Therefore, through the standardized parameter field, the structured code generated by the dictionary escape module can be translated into:

[0102] {"service":"cloud_dbase25_monitoring", "parameters":{"action":"check_15min_cpu_usage", "check_15min_ram_usage"}}.

[0103] Subsequently, by executing the above code, automated operation and maintenance actions can be triggered, such as calling the database monitoring service through the API, collecting CPU and memory usage data, and returning the results.

[0104] In this embodiment, the conclusion enhancement module is used to process the downstream application service call parameter template output by the large model driving module, and call the network operation and maintenance resources and applications involved according to the above structured operation and maintenance data. Through the large model semantic enhancement and solution matching technology, the readability and usability of the output conclusion are improved.

[0105] In some embodiments of the present disclosure, a target operation and maintenance suggestion statement is determined based on the code text content. One achievable method is to vectorize the code text content to obtain a preliminary operation and maintenance suggestion statement; generate a fault solution statement based on the preliminary operation and maintenance suggestion statement and the user operation and maintenance requirement vector; input the fault solution statement into a large language model for semantic enhancement to obtain a target operation and maintenance suggestion statement.

[0106] The conclusion enhancement module uses the downstream application service call parameter template to call the relevant network management operation and maintenance application, and integrates its external query data into the operation and maintenance information to supplement the conclusion. The implementation method is as follows:

[0107] In this method, according to the downstream application service call parameter template, the downstream application service is requested through the API port, and the code text content returned by the downstream application service is received. Among them, according to the downstream application service call parameter template generated by the large model driver module, the corresponding downstream application service is requested through the API port, and the application return content is accepted. After the request is executed, the output results of the downstream application service are collected, and the results are vectorized to finally generate the external application query conclusion text vector.

[0108] For example, the technical method of the conclusion enhancement module is to call the parameter template of the downstream application service, initiate a synchronous or asynchronous request to the downstream operation and maintenance application service through the defined application program interface (API) endpoint, and receive and parse the returned code text content: including status code, header information and body content.

[0109] The returned text content is vectorized by using the pre-made prompt word template of the pre-trained large language model. The method includes Word2Vec, BERT or its variants to convert the text into a fixed-dimensional vector representation, including word segmentation, part-of-speech tagging, named entity recognition (NER), etc., to extract preliminary operation and maintenance suggestion sentences related to the operation and maintenance conclusion.

[0110] Step 2: The large model semantic enhancement module receives the user operation and maintenance demand vector and the preliminary operation and maintenance suggestion statement output by the knowledge integration module. The two vectors represent the semantic representation of the user query and compares and matches the user operation and maintenance demand vector with the obtained one.

[0111] In the above embodiment, a fault solution statement is generated based on the preliminary operation and maintenance suggestion statement and the user operation and maintenance requirement vector. One achievable method is to determine the similarity between the preliminary operation and maintenance suggestion statement and the user operation and maintenance requirement vector based on a similarity algorithm; and determine the fault solution statement that matches the input text vector based on the similarity. For example, the large model semantic enhancement module uses cosine similarity, Euclidean distance or other similarity measurement methods to compare the user operation and maintenance requirement vector and the preliminary operation and maintenance suggestion statement with the solution template knowledge base to evaluate the similarity between the two vectors, and determines to retrieve the fault solution statement that matches the user demand and the external application service output according to the measured matching degree.

[0112] Step 3: The conclusion output module inputs the troubleshooting solution statement into the large language model again through the prompt word fine-tuning method for text post-processing, and performs semantic enhancement on the text statement, including but not limited to text summary, keyword extraction, text expansion and rewriting, etc., to enhance the readability and accuracy of the troubleshooting solution. Secondly, the large language model is used to optimize the grammar and sentence structure to ensure the fluency and professionalism of the text, and through sentiment analysis and tone adjustment, ensure that the tone of the troubleshooting solution is consistent with the expected communication style. The conclusion output after semantic enhancement is displayed to the user as the final response of the system.

[0113] The processing flow of the conclusion enhancement module is described in detail below with reference to a specific embodiment.

[0114] A cloud service data center received feedback from downstream users that its cloud database service had performance degradation issues.

[0115] Step 1: According to the request content on the generated call template, send a request to the operation and maintenance application service through the API port and receive the returned content. In this use case, the request content of the generated call template is:

[0116] {"service":"cloud_dbase25_monitoring","parameters":{"action":"check_15min_cpu_usage","check_15min_ram_usage"}}

[0117] Send a request to the operation and maintenance application service through the API port, receive the returned content results, and collect the status code, header information, and body content returned by the API. Specifically, the execution return results can be in the form of strings, arrays, and lists, as follows:

[0118]

[0119]

[0120] Here, timestamp indicates the timestamp of the data record.

[0121] cpuUsageData is an object that contains 15 minutes of usage related data.

[0122] Among them, 15minCpuUsage contains detailed information of the 15-minute period, including: intervalStart and intervalEnd define the time interval for usage measurement.

[0123] averageUsage indicates the average usage.

[0124] peakUsage indicates the peak usage.

[0125] usageTrend describes the usage trend (e.g. "increasing", "decreasing", "steady").

[0126] samples is an array containing the usage sample points within the time interval.

[0127] systemStatus provides a brief report of the system status, such as "normal", "warning", and "critical", corresponding to normal, warning, and failure respectively.

[0128] alerts is an array that can be used to log any relevant warning or alert information.

[0129] The metadata provides additional information about the CPU usage data, such as units and description.

[0130] Next, the above content can be input into the big model, and the fields closely related to the operation and maintenance intentions can be searched through the attention method and the text distance method, and deleted and summarized. In this use case, the field in the above cpuUsageData object that refers to the system CPU utilization within 15 minutes can be translated into "cpu_usage" by the big model; the field in the memoryUsageData object that refers to the system CPU utilization within 15 minutes can be translated into "memory_usage" by the big model; the status codes of the above two fields can be retained as a data structure that transmits the above field information in structured data.

[0131] Therefore, the states in the above use case can be summarized as:

[0132] {"status":"success","data":{"cpu_usage":"normal","memory_usage":"high"}}

[0133] Next, the data collected by executing the code in the previous step and the results of the operation and maintenance actions are input into the big model again, and all the contents of the fields contained in the object are vectorized and retrieved, sorted, and output in the form of natural language text through attention. The outline of the above content is extracted, summarized, expanded, and the preliminary operation and maintenance suggestion statement after wording optimization is output.

[0134] The generated preliminary operation and maintenance suggestion statements are presented to the user in the form of natural language text, for example: "After inspection, the CPU usage of the cloud database service is normal, and the memory usage is high. It is recommended to further analyze the process log query of the integrated resource center to determine the cause of the performance degradation."

[0135] Step 2: The large model knowledge integration module integrates the user's operation and maintenance intentions and generates operation and maintenance suggestions for the call template.

[0136] After receiving user feedback, we use natural language processing technology to extract key information and convert it into vector form. For example, the user feedback "cloud database service performance deteriorates" is converted into a vector, represented as

[0137] v usr = <x1,x2,...x n >

[0138] The preliminary operation and maintenance suggestions obtained in the previous step are converted into another vector through the same NLP technology and vectorization process, expressed as:

[0139] v conc = <y1,y2,...y n >

[0140] The large model semantic enhancement module receives the user operation and maintenance demand vector v usr and external application query conclusion text vector v conc Calculate the similarity between two vectors using cosine similarity or Euclidean distance:

[0141]

[0142] According to the above distance measurement method, the text vectors are sorted again, and the vector texts that indicate that the user needs and the preliminary operation and maintenance suggestions are highly correlated are selected.

[0143] Step 3: Input the matched solution statement into the large language model for semantic enhancement. Perform text summarization, keyword extraction, text expansion and rewriting. Ensure the grammatical correctness and fluency of the text. Adjust the tone of the text to ensure that the tone is consistent with the expected communication style. Display the conclusion output after semantic enhancement to the user. In this use case, it is:

[0144] "According to our analysis, your cloud database service currently has a normal CPU usage, but a high memory usage. This may be the cause of the performance degradation. We recommend that you conduct an in-depth analysis of the database server's memory usage and check for memory leaks or query optimization issues. At the same time, we recommend reviewing recent system and application logs to identify any unusual activities or configuration issues that may affect performance. If you need further assistance, please feel free to contact our technical support team."

[0145] The present disclosure can accurately identify the user's true intention through vectorized fault knowledge, large model intent extraction and semantic enhancement, and call the corresponding downstream fault operation and maintenance services according to the intention, thereby improving the fault detection rate and accuracy of the communication operation and maintenance question and answer system, and shortening the processing time and accuracy of conversational fault operation and maintenance.

[0146] Specifically, there are the following aspects:

[0147] The scene adaptability is increased by using a pre-trained base model with certain basic logical capabilities embedded in it to ensure accurate answers to query questions when the capacity of the external knowledge base is limited.

[0148] Usability improvement: Under the premise of optimizing the plug-in knowledge base, the question-answering assistance capability of zero-shot or few-shot operation and maintenance knowledge can be realized, ensuring that each question-answering query of the operation and maintenance personnel receives a relatively accurate and reasonable answer. The final conclusion enhancement module of the disclosed device is used to match the fault solution and enhance the logic of the output language, which improves the readability and usability of the fault operation and maintenance auxiliary reply.

[0149] Improved scalability: The present invention also has fault operation and maintenance expansion capabilities that are compatible with external application interfaces on the above-mentioned basic devices.

[0150] Figure 3 FIG. 1 is a schematic diagram of a network fault operation and maintenance device 30 provided by an exemplary embodiment of the present disclosure. Figure 3 As shown, the network fault operation and maintenance device 30 includes: a vectorization module 31, a first determination module 32, a semantic reasoning module 33, a standardization module 34, a request module 35 and a second determination module 36.

[0151] The vectorization module 31 is used to vectorize the acquired input text containing the fault operation and maintenance service request statement to obtain an input text vector;

[0152] A first determination module 32, configured to determine a user operation and maintenance requirement vector according to an input text vector;

[0153] The semantic reasoning module 33 is used to perform semantic reasoning operations on the user operation and maintenance demand vector to obtain a structured code;

[0154] A standardization module 34 is used to perform field standardization operations on the structured code to obtain a downstream application service call parameter template that can be called by the downstream application service;

[0155] The request module 35 is used to call the parameter template according to the downstream application service, request the downstream application service through the API port, and receive the code text content returned by the downstream application service;

[0156] The second determination module 36 is used to determine the target operation and maintenance suggestion statement according to the code text content.

[0157] Optionally, when determining the user operation and maintenance requirement vector according to the input text vector, the first determining module 32 is configured to:

[0158] Match the input text vector with the candidate fault operation and maintenance knowledge vector in the fault operation and maintenance knowledge base to obtain a successfully matched target fault operation and maintenance knowledge vector;

[0159] Match the input text vector with the candidate prompt word vectors in the prompt word library to obtain the successfully matched target prompt word vector;

[0160] The input text vector, the target fault operation and maintenance knowledge vector, and the target prompt word vector are concatenated to obtain the user operation and maintenance demand vector.

[0161] Optionally, when the first determination module 32 matches the input text vector with the candidate fault operation and maintenance knowledge vector in the fault operation and maintenance knowledge base to obtain a successfully matched target fault operation and maintenance knowledge vector, it is used to:

[0162] Determine a first vector similarity between the input text vector and each candidate fault operation and maintenance knowledge vector in the fault operation and maintenance knowledge base;

[0163] Select the first k target fault operation and maintenance knowledge vectors from the candidate fault operation and maintenance knowledge vectors in descending order of the similarity of the first vector;

[0164] When the first determination module 32 matches the input text vector with the candidate prompt word vector in the prompt word library to obtain a successfully matched target prompt word vector, it is used to:

[0165] Determine a second vector similarity between the input text vector and each candidate prompt word vector in the prompt word library;

[0166] In descending order of the similarity of the second vectors, the first k target prompt word vectors are selected from the candidate prompt word vectors, where k is a positive integer.

[0167] Optionally, when the semantic reasoning module 33 performs semantic reasoning on the user operation and maintenance requirement vector to obtain a structured code, it is used to:

[0168] Extract the user's operation and maintenance demand vector to obtain the user's intention;

[0169] Generate operation and maintenance actions corresponding to user intentions;

[0170] According to the intelligent operation and maintenance rules, the set keywords in the operation and maintenance actions are removed to obtain the audited operation and maintenance actions;

[0171] Perform semantic escape operations on the audited operation and maintenance actions to obtain structured code.

[0172] Optionally, when the semantic reasoning module 33 extracts the user operation and maintenance demand vector and obtains the user intention, it is used to:

[0173] The user operation and maintenance requirements are input into the pre-trained deep learning model, the user operation and maintenance requirement vector is encoded, and the user intention is extracted using the attention weight.

[0174] Optionally, when determining the target operation and maintenance suggestion statement according to the code text content, the second determination module 36 is used to:

[0175] Vectorize the code text content to obtain preliminary operation and maintenance suggestion statements;

[0176] Generate a fault solution statement based on the preliminary operation and maintenance suggestion statement and the user operation and maintenance requirement vector;

[0177] The fault solution statement is input into the large language model for semantic enhancement to obtain the target operation and maintenance recommendation statement.

[0178] Optionally, when generating a fault solution statement according to the preliminary operation and maintenance suggestion statement and the user operation and maintenance requirement vector, the second determination module 36 is used to:

[0179] Determine the similarity between the preliminary operation and maintenance suggestion statement and the user operation and maintenance requirement vector according to the similarity algorithm;

[0180] Based on the similarity, the fault scenario statements that match the input text vector are determined.

[0181] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0182] Figure 4 FIG. 1 is a schematic diagram of a structure of an electronic device provided by an exemplary embodiment of the present disclosure. Figure 4As shown, the electronic device includes: a memory 41 and a processor 42. In addition, the electronic device also includes a power supply component 43 and a communication component 44.

[0183] The memory 41 is used to store computer programs and can be configured to store various other data to support operations on the electronic device. Examples of such data include instructions for any application program or method operating on the electronic device.

[0184] The memory 41 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0185] The communication component 44 is used for data transmission with other devices.

[0186] The processor 42 can execute computer instructions stored in the memory 41 to: vectorize the obtained input text containing the fault operation and maintenance business request statement to obtain an input text vector; determine the user operation and maintenance demand vector based on the input text vector; perform semantic reasoning operations on the user operation and maintenance demand vector to obtain structured code; perform field standardization operations on the structured code to obtain a downstream application service call parameter template that can be called by the downstream application service; request the downstream application service through the API port according to the downstream application service call parameter template, and receive the code text content returned by the downstream application service; determine the target operation and maintenance suggestion statement based on the code text content.

[0187] Accordingly, the present disclosure also provides a computer-readable storage medium storing a computer program. When the computer-readable storage medium stores the computer program and the computer program is executed by one or more processors, the one or more processors execute Figure 1 Each step in the method embodiment.

[0188] Accordingly, the present disclosure also provides a computer program product, the computer program product includes a computer program / instruction, the computer program / instruction is executed by a processor Figure 1 Each step in the method embodiment.

[0189] Above Figure 4The communication component in is configured to facilitate wired or wireless communication between the device where the communication component is located and other devices. The device where the communication component is located can access a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G / LTE, 5G and other mobile communication networks, or a combination thereof. In an exemplary embodiment, the communication component receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0190] Above Figure 4 The power supply component in the device provides power to various components of the device where the power supply component is located. The power supply component may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device where the power supply component is located.

[0191] The electronic device also includes a display screen and an audio component.

[0192] The display screen includes a screen, and the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from a user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure associated with the touch or slide operation.

[0193] The electronic device mentioned above also includes an audio component.

[0194] The audio component may be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC), and when the device where the audio component is located is in an operating mode, such as a call mode, a recording mode, and a speech recognition mode, the microphone is configured to receive an external audio signal. The received audio signal may be further stored in a memory or sent via a communication component. In some embodiments, the audio component also includes a speaker for outputting an audio signal.

[0195] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0196] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0197] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0198] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0199] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0200] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0201] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0202] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0203] The above are only specific embodiments of the present disclosure, so that those skilled in the art can understand or implement the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to these embodiments herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A network fault operation and maintenance method, characterized in that: include: Vectorize the input text containing the fault operation and maintenance service request statement to obtain an input text vector; Determining a user operation and maintenance requirement vector according to the input text vector; Performing semantic reasoning operations on the user operation and maintenance requirement vector to obtain a structured code; Performing field standardization operation on the structured code to obtain a downstream application service calling parameter template that can be called by the downstream application service; According to the downstream application service calling parameter template, requesting the downstream application service through the API port, and receiving the code text content returned by the downstream application service; According to the code text content, a target operation and maintenance suggestion statement is determined.

2. The method according to claim 1, characterized in that: The step of determining the user operation and maintenance requirement vector according to the input text vector includes: Matching the input text vector with the candidate fault operation and maintenance knowledge vectors in the fault operation and maintenance knowledge base to obtain a successfully matched target fault operation and maintenance knowledge vector; Matching the input text vector with the candidate prompt word vectors in the prompt word library to obtain a successfully matched target prompt word vector; The input text vector, the target fault operation and maintenance knowledge vector and the target prompt word vector are concatenated to obtain a user operation and maintenance demand vector.

3. The method according to claim 1, characterized in that The step of matching the input text vector with the candidate fault operation and maintenance knowledge vector in the fault operation and maintenance knowledge base to obtain a successfully matched target fault operation and maintenance knowledge vector includes: Determine a first vector similarity between the input text vector and each candidate fault operation and maintenance knowledge vector in the fault operation and maintenance knowledge base; Selecting the first k target fault operation and maintenance knowledge vectors from the candidate fault operation and maintenance knowledge vectors in descending order of the similarity of the first vectors; The step of matching the input text vector with the candidate prompt word vectors in the prompt word library to obtain a successfully matched target prompt word vector includes: Determining a second vector similarity between the input text vector and each candidate prompt word vector in the prompt word library; In descending order of the similarity of the second vectors, the first k target prompt word vectors are selected from the candidate prompt word vectors, where k is a positive integer.

4. The method according to claim 1, characterized in that: The performing semantic reasoning operation on the user operation and maintenance requirement vector to obtain a structured code includes: Performing intent extraction on the user operation and maintenance demand vector to obtain user intent; Generate an operation and maintenance action corresponding to the user intention; According to the intelligent operation and maintenance operation rules, the set keywords in the operation and maintenance action are removed to obtain the audited operation and maintenance action; A semantic escape operation is performed on the audited operation and maintenance action to obtain a structured code.

5. The method according to claim 4, characterized in that The extracting the intent of the user operation and maintenance demand vector to obtain the user intent includes: The user operation and maintenance requirements are input into a pre-trained deep learning model, the user operation and maintenance requirement vector is encoded, and the user intention is extracted using the attention weight.

6. The method according to claim 1, characterized in that Determining a target operation and maintenance suggestion statement according to the code text content includes: Vectorizing the code text content to obtain a preliminary operation and maintenance suggestion statement; Generate a fault solution statement according to the preliminary operation and maintenance suggestion statement and the user operation and maintenance requirement vector; The fault solution statement is input into a large language model for semantic enhancement to obtain a target operation and maintenance suggestion statement.

7. The method according to claim 6, characterized in that The generating a fault solution statement according to the preliminary operation and maintenance suggestion statement and the user operation and maintenance requirement vector includes: Determine the similarity between the preliminary operation and maintenance suggestion statement and the user operation and maintenance requirement vector according to a similarity algorithm; According to the similarity, a fault solution statement matching the input text vector is determined.

8. A network fault operation and maintenance device, characterized in that: include: A vectorization module is used to vectorize the input text containing the fault operation and maintenance business request statement to obtain an input text vector; A first determination module, configured to determine a user operation and maintenance requirement vector according to the input text vector; A semantic reasoning module, used for performing semantic reasoning operations on the user operation and maintenance demand vector to obtain a structured code; A standardization module, used for performing field standardization operation on the structured code to obtain a downstream application service calling parameter template that can be called by a downstream application service; A request module, used to call a parameter template according to the downstream application service, request the downstream application service through an API port, and receive the code text content returned by the downstream application service; The second determination module is used to determine a target operation and maintenance suggestion statement according to the code text content.

9. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement each step in the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the method according to any one of claims 1 to 7 is implemented.

Images