5G private network operation and maintenance method based on large model adaptive capability and related device

Abstract

The invention discloses a 5G private network operation and maintenance method based on the adaptive capacity of a large model and a related device, and relates to the field of fusion of the large model and communication, and the method comprises the steps: obtaining a network operation and maintenance request for a 5G private network; retrieving a network operation and maintenance knowledge base based on a network operation and maintenance problem in the network operation and maintenance request to obtain context information corresponding to the network operation and maintenance problem; searching application perception data corresponding to the network operation and maintenance problem by using a network operation and maintenance intelligent body, and constructing a cue word for the large model based on the application perception data; and inputting the network operation and maintenance problem, the context information and the cue word into the large model so as to generate a solution and an operation and maintenance strategy for the network operation and maintenance problem based on the context information and the cue word by utilizing the large model. According to the invention, the adaptive capability of a large model in the field of 5G industry private network intelligent operation and maintenance is improved, and the efficiency and precision of 5G industry private network intelligent operation and maintenance are improved.

Description

Technical Field

[0001] This application relates to the field of the integration of large models and communications, and particularly to a 5G private network operation and maintenance method and related devices based on the adaptive ability of large models. Background Art

[0002] With the acceleration of the digital transformation of industries, the operation and maintenance requirements of industry 5G (5th Generation Mobile Communication Technology) private networks in the communication field are becoming increasingly complex, and intelligent operation and maintenance has become one of the key factors to maintain competitiveness in the digital age.

[0003] However, with the rapid development of business and the continuous update of technologies, traditional operation and maintenance methods are difficult to meet the needs of communication equipment operation and maintenance. The emergence of large model technology has brought a breakthrough to the field of intelligent operation and maintenance. It can provide a more user-friendly human-computer interaction mode, and at the same time can process a large amount of formatted data, provide high-precision analysis and prediction, and empower intelligent operation and maintenance with technology. Compared with traditional AIOps (Artificial Intelligence for IT Operations), large models can give further capabilities to intelligent operation and maintenance, such as simpler interaction, more comprehensive knowledge coverage, more flexible model architecture, etc., with a lower usage threshold, and continuously generalize operation and maintenance capabilities. At the same time, after industry users deploy 5G networks, business opening and daily operation and maintenance require professional knowledge support and personnel reserve, increasing the network usage cost and investment. Applying large models in the field of 5G operation and maintenance can achieve intelligent network operation and can significantly reduce industry investment.

[0004] Although large models have certain applications in the field of 5G private network intelligent operation and maintenance at present, there are still serious deficiencies in how to improve the adaptive ability of large models in the direction of 5G private network intelligent operation and maintenance to meet the intelligent operation and maintenance needs in the field of 5G private network intelligent operation and maintenance. Summary of the Invention

[0005] In view of this, this application provides a 5G private network operation and maintenance method and related devices based on the adaptive ability of large models, which are used to improve the adaptive ability of large models in the field of 5G industry private network intelligent operation and maintenance, so as to further improve the efficiency and accuracy of 5G industry private network intelligent operation and maintenance.

[0006] The specific technical solutions are as follows: A 5G private network operation and maintenance method based on the adaptive ability of large models, comprising: Obtaining a network operation and maintenance request for the 5G private network; the network operation and maintenance request includes network operation and maintenance problems to be solved in the 5G private network; Retrieve the network operation and maintenance knowledge base based on the network operation and maintenance problem to obtain the context information corresponding to the network operation and maintenance problem; the network operation and maintenance knowledge base is a knowledge base obtained by structuring the original operation and maintenance knowledge data of the 5G private network; Use the pre-constructed network operation and maintenance intelligent body to retrieve the application perception data corresponding to the network operation and maintenance problem, and construct prompt words for the large model based on the application perception data; Input the network operation and maintenance problem, the context information, and the prompt words into the large model, so as to use the large model to generate solutions and operation and maintenance strategies for the network operation and maintenance problem based on the context information and the prompt words.

[0007] Optionally, obtaining the network operation and maintenance request for the 5G private network includes at least one of the following: Obtain the network operation and maintenance request for the 5G private network initiated by the operation and maintenance personnel through natural language interaction; Obtain the network operation and maintenance request matching the service exception event initiated by the network operation and maintenance intelligent body based on the perceived service exception event of the 5G private network; Wherein, the network operation and maintenance intelligent body perceives the service exception event of the 5G private network by dynamically modeling the relationship between the service quality and the user experience of the 5G private network by using the genetic algorithm.

[0008] Optionally, the process of the network operation and maintenance intelligent body perceiving the service exception event of the 5G private network includes: Obtain the service quality index data corresponding to the end-to-end service performance of different categories of 5G services; Based on the correlation relationship between the network service quality and the user's perception of the network performance obtained by dynamically modeling the relationship between the service quality and the user experience of the 5G private network by using the genetic algorithm, and the service quality index data corresponding to the end-to-end service performance of different categories, determine the user's performance perception degree of the end-to-end service performance of different categories as the sub-item perception degrees corresponding to the end-to-end service performance of different categories; According to the sub-item perception degrees corresponding to the end-to-end service performance of different categories, determine the overall perception degree of the user on the network performance of the 5G service; Based on the overall perception degree of the user on the network performance of the 5G service and the network performance perception threshold obtained by the big data learning method, determine whether the service exception event of the 5G service occurs.

[0009] Optionally, before retrieving the network operation and maintenance knowledge base based on the network operation and maintenance problem, the method further includes: Perform at least one of decomposition sub-query, anaphora resolution, and query rewriting on the network operation and maintenance problem; Among them, the decomposition sub-query is used to split the network operation and maintenance problem into multiple sub-problems with independent information; the anaphora resolution is used to replace the anaphoric information in the network operation and maintenance problem with the actual referent content; the query rewriting is used to optimize and adjust the network operation and maintenance problem at the language level, and rewrite the network operation and maintenance problem into multiple rewritten problems.

[0010] Optionally, retrieving the context information corresponding to the network operation and maintenance problem from the network operation and maintenance knowledge base based on the network operation and maintenance problem includes at least one of the following: Retrieving the network operation and maintenance knowledge base respectively based on multiple sub-problems of the network operation and maintenance problem to obtain sub-retrieval results corresponding to each of the sub-problems; performing a merging process on the sub-retrieval results corresponding to each of the sub-problems to obtain the context information corresponding to the network operation and maintenance problem; Based on different rewritten problems of the network operation and maintenance problem, respectively retrieving different network operation and maintenance knowledge bases through different retrieval methods to obtain retrieval results corresponding to each of the rewritten problems; performing a merging process on the retrieval results corresponding to each of the rewritten problems to obtain the context information corresponding to the network operation and maintenance problem.

[0011] Optionally, the different retrieval methods include some or all of database query, vector retrieval, question-and-answer retrieval, knowledge graph retrieval, plug-in retrieval, and keyword retrieval.

[0012] Optionally, using the pre-constructed network operation and maintenance intelligent body to retrieve the application perception data corresponding to the network operation and maintenance problem includes: Using the network operation and maintenance intelligent body to retrieve the user perception evaluation result, service quality data, and network performance data corresponding to the network operation and maintenance problem; the application perception data includes the user perception evaluation result, the service quality data, and the network performance data.

[0013] A 5G private network operation and maintenance device based on the adaptive ability of a large model includes: An acquisition module, configured to obtain a network operation and maintenance request for the 5G private network; the network operation and maintenance request includes a network operation and maintenance problem to be solved in the 5G private network; A first retrieval module, configured to retrieve the network operation and maintenance knowledge base based on the network operation and maintenance problem to obtain the context information corresponding to the network operation and maintenance problem; the network operation and maintenance knowledge base is a knowledge base obtained by structuring the original operation and maintenance knowledge data of the 5G private network; A second retrieval module, configured to use the pre-constructed network operation and maintenance intelligent body to retrieve the application perception data corresponding to the network operation and maintenance problem, and construct a prompt word for the large model based on the application perception data; An operation and maintenance module, which is used to input the network operation and maintenance problems, the context information, and the prompt words into a large model, so as to use the large model to generate solutions and operation and maintenance strategies for the network operation and maintenance problems based on the context information and the prompt words.

[0014] An electronic device, comprising: A memory, which is used to store computer programs; A processor, which is used to implement the 5G private network operation and maintenance method based on the adaptive ability of the large model as described in any one of the above by calling and executing the computer programs in the memory.

[0015] A computer-readable medium, on which a computer program is stored. When the computer program is executed by a processor, it can be used to implement the 5G private network operation and maintenance method based on the adaptive ability of the large model as described in any one of the above.

[0016] According to the above solutions, the 5G private network operation and maintenance method and related devices provided by this application construct a structured network operation and maintenance knowledge base for the 5G private network by pre-structuring the original operation and maintenance knowledge data of the 5G private network, and pre-construct a network operation and maintenance intelligent body for the 5G private network. On this basis, for the network operation and maintenance requests of the 5G private network, the network operation and maintenance knowledge base is retrieved based on the network operation and maintenance problems in the requests to obtain the context information corresponding to the network operation and maintenance problems, and the application perception data corresponding to the network operation and maintenance problems is retrieved by using the network operation and maintenance intelligent body, and prompt words for the large model are constructed based on the retrieved application perception data, which can provide rich and high-quality context information and prompt words for the network operation and maintenance problems for the large model, facilitating the large model to use the provided context information and prompt words as references to efficiently and accurately generate corresponding solutions and operation and maintenance strategies for the network operation and maintenance problems, thereby enhancing the adaptive ability of the large model in the field of intelligent operation and maintenance of 5G industry private networks and improving the efficiency and accuracy of intelligent operation and maintenance of 5G industry private networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0018] Figure 1 is a 5G private network intelligent operation and maintenance framework diagram provided by the present application based on a large model and related intelligent agents; Figure 2 is a flowchart of the 5G private network operation and maintenance method based on the adaptive ability of the large model provided by the present application; Figure 3 It is a schematic diagram of the business perception evaluation variable structure model of the network operation and maintenance intelligent agent provided by this application; Figure 4 It is a schematic diagram of retrieval enhancement of the network operation and maintenance intelligent agent provided by this application; Figure 5 It is a composition structure diagram of the 5G private network operation and maintenance device based on the adaptive ability of the large model provided by this application; Figure 6 It is a composition structure diagram of the electronic device provided by this application. Specific embodiments

[0019] Next, the technical solutions in the embodiments of this application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this application.

[0020] The embodiments of this application provide a 5G private network operation and maintenance method and related devices based on the adaptive ability of the large model. It mainly realizes the automated intelligent operation and maintenance of the 5G private network by using the large model to construct the intelligent operation and maintenance architecture of the 5G private network intelligent agent, so as to improve the adaptive ability of the large model in the field of intelligent operation and maintenance of the 5G industry private network and improve the efficiency and accuracy of intelligent operation and maintenance of the 5G industry private network.

[0021] The large model and related intelligent agents adopted in the embodiments of this application are used for 5G private network operation and maintenance. Its main goal is to represent the knowledge related to the 5G private network in a form that can be understood by a computer, and enable the computer to understand, reason, and apply this knowledge to achieve the automated and intelligent operation and maintenance of the 5G industry private network.

[0022] Optionally, the 5G private network operation and maintenance based on the large model and related intelligent agents in the embodiments of this application involves 5G private network professional knowledge representation, knowledge acquisition, knowledge reasoning, knowledge storage, knowledge management, and constructing an Agent for intelligent operation and maintenance of the 5G industry private network to achieve the connection of operation and maintenance requirements, LLM (large language model), and the production environment of the 5G industry private network and other related framework structures.

[0023] See Figure 1 The shown 5G private network intelligent operation and maintenance framework diagram based on the large model and related intelligent agents. The intelligent operation and maintenance of the 5G private network mainly includes the knowledge retrieval executed based on the improved Retrieval-Augmented Generation technology and the large model answer generation based on knowledge retrieval (generating corresponding solutions and operation and maintenance strategies for network operation and maintenance requests) process. In based on Figure 1Before performing intelligent operation and maintenance of the 5G private network using the intelligent operation and maintenance framework shown, it is necessary to pre-complete the construction of operation and maintenance knowledge for the 5G private network.

[0024] Among them, the construction of operation and maintenance knowledge for the 5G private network refers to structuring the original operation and maintenance knowledge data of the 5G private network to build a structured network operation and maintenance knowledge base for the 5G private network. The original operation and maintenance knowledge data of the 5G private network can include, but is not limited to, the internal database of the 5G private network, knowledge graphs, unstructured documents, relevant operation and maintenance data corresponding to internal and external system plugins, and original data such as external operation and maintenance means and Q&A knowledge. See Figure 1 As shown, the structured network operation and maintenance knowledge base can be implemented as, but is not limited to, various structured data formats such as structured knowledge graphs, databases / tables, Q&A knowledge datasets, etc.

[0025] The main steps of the construction process of the network operation and maintenance knowledge base for the 5G private network include knowledge data structuring, index establishment, and knowledge storage. These steps are mainly used to uniformly process the original internal database, knowledge graph, unstructured documents, external operation and maintenance means, Q&A knowledge, etc. of the 5G private network, and store them in a unified form of the operation and maintenance knowledge base of the 5G private network to complete the processing and construction of the intelligent operation and maintenance knowledge data of the 5G private network.

[0026] The knowledge construction of the intelligent operation and maintenance framework for the 5G private network based on large models and agents is responsible for converting the original knowledge data of the 5G private network into structured knowledge that is easy to store and retrieve for use, and storing it in the knowledge base for management. The knowledge sources of the 5G industry private network are rich and diverse, including, but not limited to, unstructured documents, knowledge graphs, databases, original operation and maintenance data corresponding to internal and external system plugins, etc. Optionally, when constructing the structured network operation and maintenance knowledge base of the 5G private network through structured processing, the embodiments of the present application first adopt the chapter and title hierarchy information of the document-type knowledge itself, use a segmentation algorithm that segments according to chapters and titles, and combine the title content and the body content to slice the original operation and maintenance knowledge data. By adding the chapter and / or title information where the slice is located to the head of each slice content, the slice content can better maintain the semantic information in the original document.

[0027] On this basis, further perform unified structured processing on each slice. Taking the structured network operation and maintenance knowledge base of the 5G private network in the form of a knowledge graph as an example, optionally, for each slice, the LLM can be used to perform entity recognition and relationship extraction on the slice content through preset prompts (prompts) to construct relevant structured knowledge of the operation and maintenance knowledge graph of the 5G industry private network. The structured data in the 5G industry private network knowledge graph is stored in the form of triples, and each triple contains an entity, an attribute, and an attribute value, and corresponds to a corresponding triple index.

[0028] Specific implementation methods of the triple index may include, but are not limited to, inverted index, prefix tree, hash table, etc. Among them, the inverted index means that each subject, predicate, and object is used as an index item to record the positions or identifiers of all triples containing the index item, so that relevant triples can be quickly found during query; the prefix tree means that for some specific application scenarios, the prefix tree can be used to organize triple data. Especially when the predicate or object has a certain regularity, this method can significantly improve the query efficiency; the hash table means that each part of the triple is mapped to a specific storage location through a hash function to achieve fast search and update operations.

[0029] Subsequently, the knowledge graph of 5G private network intelligent operation and maintenance can be directly queried by using the triple index to query the knowledge graph. When a network operation and maintenance request for the 5G private network is initiated, the system will query these triples based on the network operation and maintenance problems carried in the network operation and maintenance request to find information related to the problem as the context of the problem to assist the large model in better understanding the problem and generating corresponding solutions and operation and maintenance strategies accurately and efficiently.

[0030] In practical applications, relevant structured databases, tables, and / or knowledge graphs and other structured operation and maintenance knowledge bases can be established for the 5G private network intelligent operation and maintenance framework based on the large model through relational database theory and data warehouse theory, etc. For the case where the original operation and maintenance knowledge data of the 5G private network is in a structured form, it is not necessary to reconstruct this part of the knowledge data in the 5G private network intelligent operation and maintenance knowledge management system. It only needs to incorporate this part of the data into the knowledge management system so that the system can use the existing operation and maintenance knowledge.

[0031] On this basis, the corresponding knowledge retrieval and large model answer generation process based on knowledge retrieval can be executed through the 5G private network operation and maintenance method based on the large model adaptive ability provided by the embodiments of the present application to realize the intelligent operation and maintenance of the 5G private network.

[0032] See Figure 2 As shown in the schematic diagram of the method flow, the 5G private network operation and maintenance method based on the large model adaptive ability provided by the embodiments of the present application may at least include the following steps 201 to 204, and these steps will be described in detail below.

[0033] Step 201: Obtain a network operation and maintenance request for the 5G private network; the network operation and maintenance request includes network operation and maintenance problems to be solved in the 5G private network.

[0034] Specifically, it can obtain a network operation and maintenance request for the 5G private network initiated by an operation and maintenance personnel through natural language interaction; or obtain a network operation and maintenance request for the 5G private network that matches the service anomaly event initiated by the network operation and maintenance intelligent agent based on the perceived service anomaly event of the 5G private network. The network operation and maintenance intelligent agent perceives the service anomaly event of the 5G private network by dynamically modeling the relationship between the service quality and user experience of the 5G private network using a genetic algorithm.

[0035] In the embodiment of the present application, a network operation and maintenance intelligent agent for the 5G private network is pre-constructed. The constructed network operation and maintenance intelligent agent supports perceiving the 5G private network service and actively initiating a network operation and maintenance request for the 5G private network when it perceives an anomaly in the 5G private network service.

[0036] In the embodiment of the present application, a genetic algorithm is specifically used to dynamically model the relationship between KQI (service quality) and QoE (user experience) of the 5G private network to achieve the intelligent perception ability of the network operation and maintenance intelligent agent for the 5G private network service. By dynamically modeling the relationship between KQI and QoE of the 5G private network using a genetic algorithm, the potential correlation between the network service quality of the 5G private network and the user's perception of network performance is found. Based on this correlation, user perception evaluation is carried out. When it is identified that the perceived network performance of the user does not reach the threshold or the service indicators deteriorate, an operation and maintenance request is adaptively initiated to solve the corresponding network problem, thereby achieving the goal of true intelligent operation and maintenance of the 5G private network.

[0037] The user perception evaluation of the network operation and maintenance intelligent agent is carried out for the 5G private network service. For any service of the 5G private network, it can be divided into multiple categories of performance according to the user's feelings, such as various service performances / network performances of network jitter, latency, rate, occupancy rate, stability rate, etc. of the service. One category of performance corresponds to one or a group of KQI indicators (service quality indicators), which can also be called the end-to-end performance indicators of the service.

[0038] The end-to-end performance indicators of the service will affect the user's perception of the corresponding category of performance, and the user's perception of each category of performance of the service will determine the user's overall perception of the network performance of the service, that is, the user's overall perception of the service is comprehensively affected by the user's perception of various categories of network performance.

[0039] Based on the above technical idea, a variable structure model for 5G private network service perception evaluation of the network operation and maintenance intelligent agent is constructed. See Figure 3, In the variable influence relationship of the 5G private network service perception evaluation variable structure model, the variables are bottom-up, specifically manifested as: the end-to-end service performance of a category corresponds to a sub-item perception index and a set of KQI indicators. The quality of the KQI indicators directly affects the quality of the corresponding sub-item perception. The quality of the sub-item perception directly affects the quality of the overall perception. Therefore, from the perspective of user perception evaluation, the input variables of the network operation and maintenance intelligent agent model include various categories of KQI indicators to be evaluated, and the output variables include the sub-item perception and overall perception of various categories of this service.

[0040] Correspondingly, the process of the network operation and maintenance intelligent agent perceiving the service exception events of the 5G private network can be realized as follows: obtaining the service quality index data corresponding to the end-to-end service performance of different categories of 5G services; based on the correlation relationship between the network service quality and the user's perception of the network performance obtained by dynamically modeling the relationship between the service quality of the 5G private network and the user experience using the genetic algorithm, and the service quality index data corresponding to the end-to-end service performance of different categories, determining the performance perception of the user for the end-to-end service performance of different categories as the sub-item perception corresponding to the end-to-end service performance of different categories; determining the overall perception of the user for the network performance of the 5G service according to the sub-item perception corresponding to the end-to-end service performance of different categories; and determining whether the service exception event of the 5G service occurs based on the overall perception of the user for the network performance of the 5G service and the network performance perception threshold obtained through the big data learning method.

[0041] Optionally, the network operation and maintenance intelligent agent of the 5G private network can, through big data intelligent analysis of the full-scenario traffic model, intelligent operation and maintenance effect, and KPI trend, etc., continuously iteratively optimize the inflection point between user service perception and network performance online through the enhanced self-learning method, so as to obtain the service perception threshold / network performance perception threshold of the user for the 5G private network service for the identification and judgment of service exception events.

[0042] Step 202, retrieve the network operation and maintenance knowledge base based on the network operation and maintenance problem to obtain the context information corresponding to the network operation and maintenance problem; the network operation and maintenance knowledge base is a knowledge base obtained by structuring the original operation and maintenance knowledge data of the 5G private network.

[0043] In order to improve the accuracy of the RAG, the embodiments of the present application adopt a series of preprocessing technologies and methods to preprocess the network operation and maintenance problems in the network operation and maintenance requests. Therefore, before retrieving the network operation and maintenance knowledge base based on the network operation and maintenance problem, the embodiments of the present application also perform corresponding preprocessing on the network operation and maintenance problem, and the preprocessing performed may include, but is not limited to, some or all of decomposing sub-queries, coreference resolution, and query rewriting.

[0044] In the knowledge retrieval of the 5G industry private network intelligent operation and maintenance framework, a series of preprocessing techniques and methods are used to preprocess network operation and maintenance problems. The main purpose is to optimize network operation and maintenance problems (i.e., optimize queries) to improve retrieval efficiency and enhance retrieval accuracy. The embodiments of this application optimize queries through pre-retrieval processing methods such as decomposing sub-queries, coreference resolution, and query rewriting to adapt to complex retrieval environments, so as to be able to provide more accurate and relevant retrieval results.

[0045] Among them, decomposing sub-queries is used to split network operation and maintenance problems into multiple information-independent sub-problems; coreference resolution is used to replace the referential information in network operation and maintenance problems with the actual referents; query rewriting is used to rewrite network operation and maintenance problems into multiple rewritten problems by optimizing and adjusting network operation and maintenance problems at the language level.

[0046] Specifically, decomposing sub-queries can split a complex network operation and maintenance problem into several smaller and more easily processed parts, where each part represents an information-independent sub-problem. To achieve this goal, optionally, a multi-query retriever can be used. This retriever, with the help of an LLM, automatically generates multiple sub-problems for a given network operation and maintenance problem from multiple dimensions. The multiple dimensions can include but are not limited to different dimensions such as the time dimension and the space dimension. For the network operation and maintenance problem in a network operation and maintenance request, it can be regarded as a query (total query), while for each generated sub-problem, it can be regarded as a sub-query of the network operation and maintenance problem, facilitating subsequent retrieval of the network operation and maintenance knowledge base based on each sub-query.

[0047] Optionally, coreference resolution can be implemented using the Few-shot Prompt method and combined with the think-act-observe strategy. By integrating data samples of some common coreference resolution scenarios as Few-shot examples into the Prompt of the LLM and combining the CoT (Chain of Thought) method, the LLM can be enabled to analyze and process more complex coreference resolution problems.

[0048] Query rewriting realizes the rewriting of network operation and maintenance problems by optimizing and adjusting network operation and maintenance problems at the language level. Query rewriting aims to enhance retrieval efficiency and improve the precision of retrieval results. Through the query rewriting system, the user's information needs can be more accurately grasped, and more relevant retrieval results can be returned.

[0049] In implementation, optionally, relying on the powerful capabilities of the LLM, by using high-quality corpora and prompt words, the LLM can be stimulated to effectively rewrite network operation and maintenance problems. To further improve the query rewriting effect, a rewriting component of the service can be introduced. This component is used as an auxiliary component to specifically adjust and rewrite the network operation and maintenance problems in the network operation and maintenance requests, making them better adapt to the processing requirements of the fixed retriever and the LLM.

[0050] On the basis of the preprocessing of the network operation and maintenance problems, the network operation and maintenance knowledge base can be further retrieved based on the network operation and maintenance problems. For the above-mentioned preprocessing performed on the network operation and maintenance problems, the process of retrieving the network operation and maintenance knowledge base based on the network operation and maintenance problems can be implemented as including at least one of the following: 11) Retrieving the network operation and maintenance knowledge base respectively based on multiple sub-problems of the network operation and maintenance problem to obtain sub-retrieval results respectively corresponding to each of the sub-problems; performing a merging process on the sub-retrieval results respectively corresponding to each of the sub-problems to obtain context information corresponding to the network operation and maintenance problem.

[0051] For each sub-problem of the network operation and maintenance problem, it can be regarded as a sub-query. A multi-query retriever is used to retrieve a set of knowledge data related to it from the structured network operation and maintenance knowledge base, and finally a union operation is performed on the documents retrieved based on all sub-queries, so as to construct a more extensive knowledge data set that has a potential correlation with the network operation and maintenance problem, and use this knowledge data set as the context information of the network operation and maintenance problem to assist subsequent large models such as the LLM in accurately understanding and generating answers to the network operation and maintenance problem.

[0052] This embodiment performs knowledge retrieval based on the decomposed sub-query method, which can break through some limitations of the vector distance-based retrieval method, thereby obtaining a more abundant and diverse set of retrieval results.

[0053] 12) Retrieving different network operation and maintenance knowledge bases respectively through different retrieval methods based on different rewritten problems of the network operation and maintenance problem to obtain retrieval results respectively corresponding to each of the rewritten problems; performing a merging process on the retrieval results respectively corresponding to each of the rewritten problems to obtain context information corresponding to the network operation and maintenance problem.

[0054] The different retrieval methods may include, but are not limited to, some or all of database query, vector retrieval, question-answering retrieval (QA retrieval), knowledge graph retrieval, plug-in retrieval, and keyword retrieval.

[0055] After rewriting the network operation and maintenance problem into multiple rewritten problems (i.e., multiple queries) through query rewriting processing, this embodiment adopts a hybrid retrieval strategy, and respectively retrieves different network operation and maintenance knowledge bases through different retrieval methods based on different rewritten problems of the network operation and maintenance problem.

[0056] In implementation, optionally, after rewriting the network operation and maintenance problem into multiple rewritten problems, these different rewritten problems can be distributed to different retrieval method processes through a query routing module. The retrieval methods can include database query, vector retrieval, QA retrieval, knowledge graph retrieval, plug-in retrieval, keyword retrieval, etc. After being retrieved by multiple retrieval methods, each retrieval method will output the top K optimal retrieval results. Subsequently, the final retrieval result corresponding to the network operation and maintenance problem can be obtained by merging these retrieval results, and this final retrieval result is used as the context information of the network operation and maintenance problem.

[0057] Since the scoring and ranking criteria of the top K retrieval results generated based on different retrieval methods are different, when merging the top K retrieval results generated based on different retrieval methods, they cannot be simply combined and sorted. At this time, a new re-ranking algorithm can be introduced to combine and re-rank these top K results, so as to obtain the final top K retrieval results and discard other retrieval results. These finally selected top K retrieval results will be used as the context information of the network operation and maintenance problem and handed over to large models such as LLM together with the network operation and maintenance problem, so that the large model can generate more accurate solutions and operation and maintenance strategies for the network operation and maintenance problem based on the provided context information content.

[0058] Combined with Figure 4 Referring to the retrieval enhancement schematic diagram of the 5G private network network operation and maintenance intelligent agent shown, the relevant processing performed on the retrieval results corresponding to each sub-problem or rewritten problem of the network operation and maintenance problem can be regarded as post-processing of the retrieval. The post-processing stage is mainly responsible for further optimizing and adjusting the retrieval results of each sub-problem or rewritten problem to improve the performance of the retrieval system and the quality of the retrieval results. Among them, mainly screening, compressing and re-ranking the retrieval results of each sub-problem or rewritten problem. The purpose of performing these operations is to refine and sort out a set of final retrieval results for the network operation and maintenance problem. These final retrieval results will then be handed over to the large model to assist in understanding the network operation and maintenance problem and generating more accurate answers.

[0059] By adopting a variety of pre- and post-processing technologies and hybrid knowledge retrieval in this application embodiment to perform the required knowledge retrieval on the network operation and maintenance problem, the accuracy of RAG can be effectively improved, and further the accuracy of the answers generated by the final large model for the network operation and maintenance problem can be improved.

[0060] Step 203: Use the pre-constructed network operation and maintenance intelligent body to retrieve the application perception data corresponding to the network operation and maintenance problem, and construct prompt words for the large model based on the application perception data.

[0061] Specifically, the network operation and maintenance intelligent body can be used to retrieve the user perception evaluation results, service quality data, and network performance data corresponding to the network operation and maintenance problem. And the retrieved user perception evaluation results, service quality data, and network performance data related to the network operation and maintenance problem are used as the application perception data corresponding to the network operation and maintenance problem. Then, based on these data, prompt words for the large model are constructed to prompt the large model to mine and analyze the root cause of the network operation and maintenance problem from aspects such as user perception, service quality, and network performance, so as to quickly and efficiently generate accurate answers for the network operation and maintenance problem.

[0062] Step 204: Input the network operation and maintenance problem, the context information, and the prompt words into the large model, so that the large model generates a solution and an operation and maintenance strategy for the network operation and maintenance problem based on the context information and the prompt words.

[0063] Finally, the network operation and maintenance problem, the context information, and the prompt words can be input into the large model together, such as input into the LLM. The large model combines the input context information and prompt words to understand the input network operation and maintenance problem, analyze the root cause of the problem, and then generate a corresponding answer for the network operation and maintenance problem.

[0064] Among them, the generated answer includes a solution and an operation and maintenance strategy for the network operation and maintenance problem.

[0065] Optionally, the solution is a set of high-level decision-making solutions that cannot be directly understood and executed by the device, while the operation and maintenance strategy is responsible for implementing these high-level decisions into specific configuration rules or commands and other operation and maintenance operation information that can be directly understood and executed by the device, so as to realize the intelligent operation and maintenance of the 5G private network and automatically solve the network operation and maintenance problems indicated by the initiated network operation and maintenance requests.

[0066] According to the above solution, the 5G private network operation and maintenance method based on the adaptive ability of the large model provided by this application constructs a structured network operation and maintenance knowledge base for the 5G private network by pre-structuring the original operation and maintenance knowledge data of the 5G private network, and pre-constructs a network operation and maintenance intelligent body for the 5G private network. On this basis, for the network operation and maintenance requests of the 5G private network, the network operation and maintenance knowledge base is retrieved based on the network operation and maintenance problems in the request to obtain the context information corresponding to the network operation and maintenance problems, and the application perception data corresponding to the network operation and maintenance problems is retrieved by using the network operation and maintenance intelligent body. A prompt word for the large model is constructed based on the retrieved application perception data, which can provide rich and high-quality context information and prompt words for the network operation and maintenance problems to the large model, facilitating the large model to generate corresponding solutions and operation and maintenance strategies for the network operation and maintenance problems efficiently and accurately with the provided context information and prompt words as references, thereby enhancing the adaptive ability of the large model in the field of intelligent operation and maintenance of 5G industry private networks and improving the efficiency and accuracy of intelligent operation and maintenance of 5G industry private networks.

[0067] In addition, this application also embeds a 5G private network operation and maintenance intelligent body obtained by dynamically and adaptively modeling the KQI-QoE relationship based on the genetic algorithm on the basis of the large model framework. Based on this intelligent body, the correlation between service quality and user perception of network performance can be mined. When the corresponding performance indicators of the user perception do not reach the threshold or the service indicators deteriorate, the intelligent body can adaptively initiate an operation and maintenance request to solve the corresponding network operation and maintenance problems, thus achieving the goal of true intelligent operation and maintenance of 5G private network networks.

[0068] Corresponding to the above method, the embodiment of this application also provides a 5G private network operation and maintenance device based on the adaptive ability of the large model. Refer to Figure 5 the schematic diagram of the composition structure shown in, this device includes: An acquisition module 501, configured to obtain a network operation and maintenance request for a 5G private network; the network operation and maintenance request includes network operation and maintenance problems to be solved in the 5G private network; A first retrieval module 502, configured to retrieve the network operation and maintenance knowledge base based on the network operation and maintenance problems to obtain the context information corresponding to the network operation and maintenance problems; the network operation and maintenance knowledge base is a knowledge base obtained by structuring the original operation and maintenance knowledge data of the 5G private network; A second retrieval module 503, configured to retrieve the application perception data corresponding to the network operation and maintenance problems by using a pre-constructed network operation and maintenance intelligent body, and construct a prompt word for the large model based on the application perception data; An operation and maintenance module 504, configured to input the network operation and maintenance problems, the context information, and the prompt word into the large model, so as to use the large model to generate solutions and operation and maintenance strategies for the network operation and maintenance problems based on the context information and the prompt word.

[0069] In an alternative embodiment, the obtaining module 501 is specifically configured to: Obtain a network operation and maintenance request for the 5G private network initiated by an operation and maintenance personnel through natural language interaction; Obtain a network operation and maintenance request matching the service exception event initiated by the network operation and maintenance agent based on the perceived service exception event of the 5G private network; Wherein, the network operation and maintenance agent perceives the service exception event of the 5G private network by dynamically modeling the relationship between the service quality and user experience of the 5G private network by using a genetic algorithm.

[0070] In an alternative embodiment, the process by which the network operation and maintenance agent perceives the service exception event of the 5G private network includes: Obtain service quality index data corresponding to different types of end-to-end service performances of 5G services; Based on the correlation relationship between the network service quality obtained by dynamically modeling the relationship between the service quality and user experience of the 5G private network by using a genetic algorithm and the user's perception of network performance, and the service quality index data corresponding to different types of end-to-end service performances, determine the performance perception degree of the user for different types of end-to-end service performances, as the sub-item perception degrees corresponding to different types of end-to-end service performances; According to the sub-item perception degrees corresponding to different types of end-to-end service performances, determine the overall perception degree of the user for the network performance of the 5G service; Based on the overall perception degree of the user for the network performance of the 5G service and the network performance perception threshold obtained through big data learning, determine whether a service exception event of the 5G service occurs.

[0071] In an alternative embodiment, the apparatus further includes a preprocessing module, configured to perform at least one of decomposed sub-query, anaphora resolution, and query rewriting on the network operation and maintenance problem before retrieving the network operation and maintenance knowledge base based on the network operation and maintenance problem; Wherein, the decomposed sub-query is used to split the network operation and maintenance problem into multiple information-independent sub-problems; the anaphora resolution is used to replace the anaphoric information in the network operation and maintenance problem with the actual anaphoric content; the query rewriting is used to rewrite the network operation and maintenance problem into multiple rewritten problems by optimizing and adjusting the network operation and maintenance problem at the language level.

[0072] In an alternative embodiment, the first retrieval module 502 is specifically configured to: Retrieve the network operation and maintenance knowledge base respectively for multiple sub-issues based on the network operation and maintenance problem, and obtain sub-retrieval results corresponding to each of the sub-issues; perform a merging process on the sub-retrieval results corresponding to each of the sub-issues to obtain context information corresponding to the network operation and maintenance problem. Based on different rewritten problems of the network operation and maintenance problem, retrieve different network operation and maintenance knowledge bases respectively through different retrieval methods, and obtain retrieval results corresponding to each of the rewritten problems; perform a merging process on the retrieval results corresponding to each of the rewritten problems to obtain context information corresponding to the network operation and maintenance problem.

[0073] In an alternative embodiment, the different retrieval methods include some or all of database query, vector retrieval, question and answer retrieval, knowledge graph retrieval, plug-in retrieval, and keyword retrieval.

[0074] In an alternative embodiment, the second retrieval module 503 is specifically configured to: Use network operation and maintenance intelligent body to retrieve the user perception evaluation result, service quality data, and network performance data corresponding to the network operation and maintenance problem; the application perception data includes the user perception evaluation result, the service quality data, and the network performance data.

[0075] For the 5G private network operation and maintenance device based on the adaptive ability of the large model disclosed in the embodiments of the present application, since it corresponds to the 5G private network operation and maintenance method based on the adaptive ability of the large model disclosed in the above method embodiments, the description is relatively simple. For relevant similarities, please refer to the descriptions of the above method embodiments, and details are not repeated here.

[0076] The embodiments of the present application also disclose an electronic device, and the composition structure of the electronic device is as Figure 6 shown, and at least includes: A memory 10 for storing a computer program.

[0077] A processor 20 for implementing the 5G private network operation and maintenance method based on the adaptive ability of the large model provided in any of the above method embodiments by calling and executing the computer program in the memory.

[0078] The processor 20 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a neural network processor (NPU), a deep learning processor (DPU), or other programmable logic devices, etc.

[0079] In addition, the electronic device may further include components such as a communication interface and a communication bus. The memory, the processor, and the communication interface complete communication with each other through the communication bus.

[0080] The communication interface is used for communication between the electronic device and other devices. The communication bus may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus may be divided into an address bus, a data bus, a control bus, etc.

[0081] In addition, the present application also provides a computer-readable medium, on which a computer program is stored. The computer program includes program codes for executing the 5G private network operation and maintenance method based on the adaptive ability of the large model disclosed in any of the above method embodiments. Correspondingly, when the computer program is executed by the processor, it can be used to implement the 5G private network operation and maintenance method based on the adaptive ability of the large model disclosed in any of the above method embodiments.

[0082] In the context of the present application, a computer-readable medium (machine-readable medium) may be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium would include electrical connections based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0083] It should be noted that the embodiments in this specification are all described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other.

[0084] For the convenience of description, when describing the above system or device, it is divided into various modules or units according to functions for description. Of course, when implementing the present application, the functions of each unit can be implemented in one or more software and / or hardware.

[0085] From the description of the above embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of this application, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0086] Finally, it should also be noted that in this text, relational terms such as first, second, third, and target are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0087] The above are only the preferred embodiments of this application. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of this application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of this application.

Claims

1. A 5G private network operation and maintenance method based on the adaptive ability of large models, characterized in that, Including: Obtaining a network operation and maintenance request for a 5G private network; the network operation and maintenance request includes network operation and maintenance problems to be solved in the 5G private network; Retrieving a network operation and maintenance knowledge base based on the network operation and maintenance problems to obtain context information corresponding to the network operation and maintenance problems; the network operation and maintenance knowledge base is a knowledge base obtained by structuring the original operation and maintenance knowledge data of the 5G private network; Using a pre-constructed network operation and maintenance intelligent body to retrieve application perception data corresponding to the network operation and maintenance problems, and constructing prompt words for a large model based on the application perception data; Inputting the network operation and maintenance problems, the context information, and the prompt words into the large model, so as to use the large model to generate solutions and operation and maintenance strategies for the network operation and maintenance problems based on the context information and the prompt words.

2. The 5G private network operation and maintenance method based on the adaptive ability of the large model according to claim 1, wherein, The obtaining of the network operation and maintenance request for the 5G private network includes at least one of the following: Obtaining a network operation and maintenance request for the 5G private network initiated by an operation and maintenance personnel through a natural language interaction method; Obtaining a network operation and maintenance request matching the service exception event initiated by the network operation and maintenance intelligent body based on perceiving a service exception event in the 5G private network; Wherein, the network operation and maintenance intelligent body perceives the service exception event in the 5G private network by dynamically modeling the relationship between the service quality and the user experience of the 5G private network by using a genetic algorithm.

3. The 5G private network operation and maintenance method based on the adaptive ability of the large model according to claim 2, wherein, The process by which the network operation and maintenance intelligent body perceives the service exception event in the 5G private network includes: Obtaining service quality index data corresponding to different categories of end-to-end service performances of 5G services; Based on the correlation relationship between the network service quality obtained by dynamically modeling the relationship between the service quality and the user experience of the 5G private network by using a genetic algorithm and the user's perception of the network performance, and the service quality index data corresponding to different categories of end-to-end service performances, determining the performance perception degree of the user for different categories of end-to-end service performances as sub-item perception degrees corresponding to different categories of end-to-end service performances; Determining the overall perception degree of the user for the network performance of the 5G service according to the sub-item perception degrees corresponding to different categories of end-to-end service performances; Based on the overall perception degree of the user for the network performance of the 5G service and the network performance perception threshold obtained by a big data learning method, determining whether a service exception event of the 5G service occurs.

4. The 5G private network operation and maintenance method based on the adaptive ability of the large model according to claim 1, wherein, Before retrieving the network operation and maintenance knowledge base based on the network operation and maintenance problems, it further includes: Performing at least one of decomposition sub-query, anaphora resolution, and query rewriting on the network operation and maintenance problems; Wherein, the decomposition sub-query is used to split the network operation and maintenance problems into multiple information-independent sub-problems; the anaphora resolution is used to replace the anaphoric information in the network operation and maintenance problems with actual anaphoric content; the query rewriting is used to rewrite the network operation and maintenance problems into multiple rewritten problems by optimizing and adjusting the network operation and maintenance problems at the language level.

5. The 5G private network operation and maintenance method based on the adaptive ability of the large model according to claim 4, wherein, The retrieving of the network operation and maintenance knowledge base based on the network operation and maintenance problems to obtain context information corresponding to the network operation and maintenance problems includes at least one of the following: Retrieve the network operation and maintenance knowledge base respectively for multiple sub-questions based on the network operation and maintenance problem, and obtain the sub-retrieval results corresponding to each of the sub-questions; perform a merging process on the sub-retrieval results corresponding to each of the sub-questions to obtain the context information corresponding to the network operation and maintenance problem. Based on different rewritten questions of the network operation and maintenance problem, retrieve different network operation and maintenance knowledge bases respectively through different retrieval methods, and obtain the retrieval results corresponding to each of the rewritten questions; perform a merging process on the retrieval results corresponding to each of the rewritten questions to obtain the context information corresponding to the network operation and maintenance problem.

6. The 5G private network operation and maintenance method based on the adaptive ability of the large model according to claim 5, wherein, The different retrieval methods include some or all of database query, vector retrieval, question and answer retrieval, knowledge graph retrieval, plug-in retrieval, and keyword retrieval.

7. The 5G private network operation and maintenance method based on the adaptive ability of the large model according to claim 1, characterized in that, The use of the pre-constructed network operation and maintenance intelligent body to retrieve the application perception data corresponding to the network operation and maintenance problem includes: Use the network operation and maintenance intelligent body to retrieve the user perception evaluation result, service quality data, and network performance data corresponding to the network operation and maintenance problem; the application perception data includes the user perception evaluation result, the service quality data, and the network performance data.

8. A 5G private network operation and maintenance device based on the adaptive ability of a large model, characterized in that, Include: An acquisition module, configured to obtain a network operation and maintenance request for the 5G private network; the network operation and maintenance request includes a network operation and maintenance problem to be solved in the 5G private network. A first retrieval module, configured to retrieve the network operation and maintenance knowledge base based on the network operation and maintenance problem, and obtain the context information corresponding to the network operation and maintenance problem; the network operation and maintenance knowledge base is a knowledge base obtained by structuring the original operation and maintenance knowledge data of the 5G private network. A second retrieval module, configured to use the pre-constructed network operation and maintenance intelligent body to retrieve the application perception data corresponding to the network operation and maintenance problem, and construct a prompt word for the large model based on the application perception data. An operation and maintenance module, configured to input the network operation and maintenance problem, the context information, and the prompt word into the large model, so as to use the large model to generate a solution and an operation and maintenance strategy for the network operation and maintenance problem based on the context information and the prompt word.

9. An electronic device, characterized in that, Include: A memory, configured to store a computer program. A processor, configured to implement the 5G private network operation and maintenance method based on the adaptive ability of the large model as described in any one of claims 1-7 by calling and executing the computer program in the memory.

10. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it can be used to implement the 5G private network operation and maintenance method based on the adaptive ability of the large model as described in any one of claims 1-7.

Images