Wireless network poor quality analysis processing method and device and medium

Abstract

The invention provides a wireless network poor-quality analysis processing method and device and a medium, and relates to the technical field of networks, and the method comprises the steps: recognizing an inquiry intention for wireless network poor-quality in a user question, and obtaining a plurality of wireless network poor-quality cells which need to be analyzed for answering the user question; compiling a poor-quality root cause diagnosis chain for answering user questions, wherein the poor-quality root cause diagnosis chain comprises a plurality of diagnosis subtasks for each wireless network poor-quality cell; obtaining a plurality of diagnosis results of the plurality of diagnosis sub-tasks, wherein each diagnosis result comprises an anomaly judgment, an anomaly root cause and a processing suggestion of each diagnosis sub-task; and obtaining a comprehensive poor quality root cause and a comprehensive poor quality processing suggestion of the plurality of wireless network poor quality cells according to the plurality of diagnosis results, and outputting an answer to the user question according to the comprehensive poor quality root cause and the comprehensive poor quality processing suggestion. According to the invention, an intelligent agent large model is adopted to provide a network optimization auxiliary tool with a relatively high intelligent degree.

Description

Technical Field

[0001] The present disclosure at least relates to the field of network technology, and in particular to a wireless network quality difference analysis and processing method, a wireless network quality difference analysis and processing device, and a computer-readable storage medium. Background Art

[0002] Existing analysis of poor wireless network quality issues mostly uses traditional data analysis experience and manual participation of network optimization staff. This results in insufficient intelligence in network optimization analysis, and the degree of optimization depends on the level of the optimization staff. The network optimization work is inefficient, has poor accuracy, and is difficult to meet large-scale network optimization requirements. Summary of the invention

[0003] The technical problem to be solved by the present disclosure is to provide a wireless network quality difference analysis and processing method, a wireless network quality difference analysis and processing device and a computer-readable storage medium in response to the above-mentioned shortcomings, so as to solve the problem of how to improve the analysis and optimization efficiency of poor quality cells through automated and intelligent methods.

[0004] In a first aspect, the present disclosure provides a method for analyzing and processing wireless network quality differences, the method being applied to an AI Agent, comprising using a large network model:

[0005] Identify the inquiry intention of the user's question about the poor quality of the wireless network, and obtain several cells with poor wireless network quality that need to be analyzed to answer the user's question;

[0006] A root cause diagnosis chain for poor quality is compiled to answer user questions, the root cause diagnosis chain for poor quality includes several diagnosis subtasks for each poor quality cell of the wireless network;

[0007] Obtaining multiple diagnostic results of multiple diagnostic subtasks, each diagnostic result including an abnormality judgment, abnormal root cause and processing suggestion of each diagnostic subtask;

[0008] The root causes of poor quality and suggestions for handling poor quality of several wireless network poor quality cells are obtained according to several diagnosis results, and answers to user questions are output according to the root causes of poor quality and the suggestions for handling poor quality.

[0009] Furthermore, the inquiry intention about poor wireless network quality in the user's question is identified, and several cells with poor wireless network quality that need to be analyzed to answer the user's question are obtained, specifically including:

[0010] Receive user questions and extract the inquiry time, inquiry area and inquiry category in the user questions through keywords;

[0011] If the inquiry category includes inquiries about wireless network services and / or indicators, determining whether there is a cell with poor wireless network quality in the inquiry area within the inquiry time;

[0012] If so, it is determined that the user's question includes an intention to inquire about poor wireless network quality, and the identifiers of several cells with poor wireless network quality in the inquiry area within the inquiry time and the real-time poor wireless network quality category are obtained.

[0013] Further, judging whether there is a cell with poor wireless network quality in the inquiry area within the inquiry time specifically includes:

[0014] Obtaining a regional wireless network service indicator of the inquiry area within the inquiry time, and judging whether the inquiry area within the inquiry time is a poor wireless network quality area and several real-time poor wireless network quality categories according to the regional wireless network service indicator;

[0015] If the inquiry area is a poor wireless network quality area, obtain the cell-level wireless network service indicator of each wireless network cell in the inquiry area within the inquiry time relative to the several real-time poor wireless network quality categories, and obtain several poor wireless network quality cells in the inquiry area within the inquiry time according to the cell-level wireless network service indicator.

[0016] Furthermore, the method also includes pre-training a large network model, specifically including:

[0017] Collect historical wireless network data, historical wireless network poor quality categories, historical wireless network poor quality diagnosis work orders, wireless network poor quality diagnosis expert knowledge, historical wireless network poor quality diagnosis results, and wireless network API interface for obtaining historical wireless network data;

[0018] Control the network big model, extract the thinking chain question and answer prompt words of the network big model according to the diagnosis process in the historical wireless network poor quality diagnosis work order, combine the historical wireless network data, historical wireless network poor quality categories, thinking chain question and answer prompt words and wireless network poor quality diagnosis expert knowledge to output the historical wireless network poor quality diagnosis results, and fine-tune the parameters of the network big model until the network big model accurately outputs the historical wireless network poor quality diagnosis results;

[0019] Get the wireless network API interface corresponding to the thinking chain question and answer prompt words corresponding to different historical wireless network quality categories and abnormal historical wireless network data in the final network model.

[0020] Furthermore, a root cause diagnosis chain for poor quality is compiled to answer user questions. The root cause diagnosis chain for poor quality includes several diagnosis subtasks for each poor quality cell of the wireless network, specifically including:

[0021] According to the real-time poor wireless network quality category of each poor wireless network quality cell, the corresponding thinking chain question and answer prompt words are obtained;

[0022] According to the question and answer prompts of the thinking chain, a root cause diagnosis chain of poor quality including several diagnosis subtasks for each poor quality cell of the wireless network is compiled;

[0023] Determine the wireless network API interface that needs to be called for each diagnostic subtask.

[0024] Further, wherein:

[0025] The plurality of diagnosis subtasks include at least one of: fault diagnosis, parameter diagnosis, coverage diagnosis, quality diagnosis, interference diagnosis, capacity diagnosis, and hidden fault diagnosis;

[0026] The wireless network API interface includes: at least one of: base station side alarm API interface, cell performance KPI indicator API interface, base station main control board API interface, cell CM configuration file API interface, cell performance KPI indicator API interface, MR indicator API interface, uplink interference diagnosis small model API interface, and cell B domain business traffic indicator API interface.

[0027] Furthermore, several diagnostic results of several diagnostic subtasks are obtained, including:

[0028] Process several diagnostic subtasks in sequence according to the order in the poor quality root cause diagnosis chain;

[0029] Call the wireless network API interface corresponding to each diagnostic subtask to obtain the corresponding real-time wireless network data;

[0030] Combining real-time wireless network data, real-time wireless network poor quality categories, thinking chain question and answer prompts and wireless network poor quality diagnosis expert knowledge, the real-time wireless network poor quality diagnosis results of each diagnosis subtask are output.

[0031] Further, the root causes of poor quality and suggestions for handling poor quality of several wireless network poor quality cells are obtained according to the several diagnosis results, and answers to user questions are output according to the root causes of poor quality and the suggestions for handling poor quality, specifically including:

[0032] Output voice and / or text results of the root causes of poor quality and poor quality treatment suggestions for several wireless network poor quality cells according to the abnormal root causes and treatment suggestions of the diagnosis subtasks whose sequential comprehensive diagnosis results in the poor quality root cause diagnosis chain are abnormal;

[0033] Several wireless network quality-poor cells are displayed on a map and their root causes are marked. Based on the quality-poor treatment suggestions, a page instruction for repairing the root cause is generated and executed automatically or when triggered by the user.

[0034] In a second aspect, the present disclosure provides a wireless network quality difference analysis and processing device, the device includes an intelligent agent AI Agent, the intelligent agent AI Agent includes a network large model, and the network large model includes:

[0035] An intention recognition module is used to recognize the inquiry intention of the user's question about the poor quality of the wireless network, and obtain a number of poor quality wireless network cells that need to be analyzed to answer the user's question;

[0036] A chain arrangement module, connected to the intent recognition module, is used to compile a poor quality root cause diagnosis chain to answer user questions, where the poor quality root cause diagnosis chain includes several diagnosis subtasks for each poor quality cell of the wireless network;

[0037] The diagnosis analysis module is connected to the chain orchestration module and is used to obtain a number of diagnosis results of a number of diagnosis subtasks, each diagnosis result including anomaly judgment, anomaly root cause and processing suggestions of each diagnosis subtask;

[0038] The summarizing output module is connected to the diagnosis and analysis module, and is used to obtain the root causes of poor quality and poor quality treatment suggestions of several wireless network poor quality cells according to several diagnosis results, and output answers to user questions according to the root causes of poor quality and the poor quality treatment suggestions.

[0039] In a third aspect, the present disclosure provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the wireless network quality difference analysis and processing method as described above is implemented.

[0040] The present disclosure provides a wireless network quality poor analysis and processing method, a wireless network quality poor analysis and processing device, and a computer-readable storage medium. An intelligent agent large model is used to provide a highly intelligent network optimization auxiliary tool. By identifying the user's problem intention, the root cause diagnosis of the wireless network quality poor is initiated. By compiling a quality poor root cause diagnosis chain, decomposing the diagnosis tasks, and then synthesizing the diagnosis results, the root cause of the wireless network quality poor and processing suggestions are obtained. The tool can be used to assist network optimization engineers in completing network optimization work and improve network optimization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a flow chart of a wireless network quality difference analysis and processing method according to an embodiment of the present disclosure;

[0042] Figure 2 It is a structural schematic diagram of a wireless network quality difference analysis and processing device according to an embodiment of the present disclosure;

[0043] Figure 3 It is an architecture diagram of a wireless network quality difference analysis and processing system according to an embodiment of the present disclosure;

[0044] Figure 4 is a flow chart of another wireless network quality difference analysis and processing method according to an embodiment of the present disclosure;

[0045] Figure 5 is an example diagram of a poor quality root cause diagnosis chain according to an embodiment of the present disclosure;

[0046] Figure 6 are two example diagrams of alarm diagnosis and analysis results according to an embodiment of the present disclosure;

[0047] Figure 7 is an example diagram of two parameter diagnostic analysis results of an embodiment of the present disclosure;

[0048] Figure 8 are example diagrams of five coverage diagnostic analysis results according to an embodiment of the present disclosure;

[0049] Fig. 9 are example diagrams of three quality diagnosis analysis results of the embodiments of the present disclosure;

[0050] Fig.10 This is a cell hour-level PRB-level waveform example 1 of an embodiment of the present disclosure;

[0051] Fig.11 This is a cell hour-level PRB-level waveform example 2 of an embodiment of the present disclosure;

[0052] Fig.12 This is a cell hour-level PRB-level waveform diagram example 3 of an embodiment of the present disclosure;

[0053] Fig.13 This is a cell hour-level PRB-level waveform example 4 of an embodiment of the present disclosure;

[0054] Fig.14 This is a cell hour-level PRB-level waveform example 5 of an embodiment of the present disclosure;

[0055] Fig.15 is an example diagram of six interference diagnosis analysis results of an embodiment of the present disclosure;

[0056] Fig.16 are example diagrams of four capacity diagnostic analysis results according to an embodiment of the present disclosure;

[0057] Fig.17 are two example diagrams of hidden fault diagnosis and analysis results of the embodiments of the present disclosure;

[0058] Fig.18 is a flowchart of an execution method of an execution body of an embodiment of the present disclosure;

[0059] Fig.19 is a page schematic diagram of the quality difference analysis auxiliary tool according to an embodiment of the present disclosure;

[0060] Fig. 20 This is an example diagram of a page of the quality difference analysis auxiliary tool according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0061] In order to enable those skilled in the art to better understand the technical solution of the present disclosure, the embodiments of the present disclosure will be further described in detail below with reference to the accompanying drawings.

[0062] It should be understood that the specific embodiments and drawings described herein are only used to explain the present disclosure rather than to limit the present disclosure.

[0063] It can be understood that, in the absence of conflict, the various embodiments of the present disclosure and the various features in the embodiments can be combined with each other.

[0064] It can be understood that, for the convenience of description, the drawings of the present disclosure only show parts related to the present disclosure, while parts irrelevant to the present disclosure are not shown in the drawings.

[0065] It can be understood that each module or unit involved in the embodiments of the present disclosure may correspond to only one entity structure, or may be composed of multiple entity structures, or multiple modules or units may be integrated into one entity structure.

[0066] It will be understood that, in the absence of conflict, the functions and steps marked in the flowcharts and block diagrams of the present disclosure may occur in an order different from that marked in the drawings.

[0067] It is understood that the flowcharts and block diagrams of the present disclosure illustrate the possible architectures, functions, and operations of the systems, devices, equipment, and methods according to the various embodiments of the present disclosure. Each box in the flowchart or block diagram may represent a module, unit, program segment, or code, which contains executable instructions for implementing the specified functions. Moreover, each box or combination of boxes in the block diagram and flowchart may be implemented by a hardware-based device that implements the specified functions, or may be implemented by a combination of hardware and computer instructions.

[0068] It can be understood that the modules and units involved in the embodiments of the present disclosure can be implemented by software or hardware, for example, the modules and units can be located in a processor.

[0069] Embodiment 1:

[0070] like Figure 1 As shown, the present disclosure provides a wireless network quality difference analysis and processing method, the method is applied to an intelligent agent AI Agent, including using a network big model:

[0071] S1. Identify the inquiry intention of the user regarding poor wireless network quality in the question, and obtain several cells with poor wireless network quality that need to be analyzed to answer the user's question;

[0072] S2. compile a root cause diagnosis chain for poor quality to answer user questions, where the root cause diagnosis chain includes several diagnosis subtasks for each poor quality cell of the wireless network;

[0073] S3, obtaining several diagnostic results of several diagnostic subtasks, each diagnostic result including an abnormality judgment, abnormal root cause and processing suggestion of each diagnostic subtask;

[0074] S4. Obtain root causes of poor quality and suggestions for handling poor quality of several wireless network poor quality cells according to several diagnosis results, and output answers to user questions according to the root causes of poor quality and the suggestions for handling poor quality.

[0075] In this embodiment, a large intelligent agent model is used to provide a highly intelligent network optimization auxiliary tool. By identifying the user's problem intention, the root cause diagnosis of poor wireless network quality is initiated. By compiling a root cause diagnosis chain for poor quality, decomposing the diagnosis tasks, and then synthesizing the diagnosis results, the root cause of poor wireless network quality and treatment suggestions are obtained. This can be used to assist network optimization engineers in completing network optimization work and improve network optimization efficiency. The device corresponding to the method is as follows Figure 2 As shown, it includes an intention recognition module 1 for executing step S2, a chain arrangement module 2 for executing step S2, a diagnosis and analysis module 3 for executing step S3 and an inductive output module 4 for executing step S4.

[0076] Specifically, a method and device for processing poor quality cells of wireless network voice services based on AI Agent is provided. AI Agent refers to an intelligent agent, which is a system driven by a large language model as the brain, has the ability to autonomously understand, perceive, plan, remember and use tools, and can automatically perform and complete complex tasks. This embodiment uses AI Agent to locate and analyze poor quality cells in wireless networks, and puts forward suggestions for processing and repair. Poor quality cells refer to cells with one or more poor quality problems in network communication quality. The types of poor quality problems of cells involve multiple aspects, such as voice service connection rate, call drop rate, packet loss rate, data service drop rate, access success rate, user perception rate, etc. The network large model of this embodiment is based on the large language model in AIAgent. After special pre-training, it is suitable for analyzing poor quality cells of wireless networks, locating the poor quality root causes of poor quality cells of wireless networks, and providing treatment suggestions for the poor quality root causes. That is, AI Agent refers to an intelligent agent based on the network large model, and the network large model refers to the large language model trained according to wireless network problems.

[0077] At present, the identification of the problem types of poor quality cells mainly adopts the traditional network optimization (abbreviated as: network optimization) method. Specifically, first collect the key performance indicators (KPI) of the cell, and then the network optimization staff sets thresholds for each KPI based on their own knowledge and experience, and then judge the type of poor quality problems of the cell based on the thresholds of these KPIs and the various KPIs of the cell. Then, the network optimization staff analyzes the data, manually checks various network problems such as coverage, interference, and capacity, and conducts root cause analysis and formulates optimization solutions based on experience, so as to optimize the network problems. From the current situation, the root cause analysis of wireless network problems generally relies on manual experience and combines various tool software for analysis, which has many problems: the lack of personnel with professional skills, and the lack of expertise has become an optimization bottleneck; the network problems are complex, the causes are diverse, and the problems are difficult to locate; the problems are repeatedly handled, the processing timeliness is poor, and the cycle is long; the network problem mining has certain limitations, the real pain point is not clearly located, the human investment cost is high, and the benefit is low.

[0078] With the continuous progress and development of the mobile communication network industry, high-speed and high-quality network application content has put forward higher requirements for network operations. Network operators hope to "discover problems before users". In addition to the conventional manual statistics of cell KPIs and the periodic discovery of poor-quality cells defined by manual rules, the design and application of machine learning-oriented mobile communication network poor-quality cell positioning analysis methods are conducive to discovering poor-quality problems in the network, and can predict the degree of poor quality of the network and whether it affects the user's perception of use. With the help of machine learning, the comprehensive analysis of the positioning of poor-quality cells in the mobile communication network can timely predict and deal with network problems in advance, bringing users a better service experience, which has certain practical significance and value.

[0079] Therefore, this embodiment proposes the concept of intelligent AI Agent, and constructs an intelligent entity for analyzing and processing wireless cells with poor quality based on a large model AI Agent architecture. The front-line network optimization engineer throws out the actual problem of wireless cells with poor quality in the form of natural language. After the agent reads the problem, the LLM (Large Language Model) as the center uses long and short-term memory to plan the problem. After the agent understands the intention of analyzing and processing wireless cells with poor quality, it decomposes the task objectives and coordinates the execution of the scheduling tool set, including real-time data acquisition of various key indicators, expert knowledge base cases and decision-making small model calls, etc. These tools are then called in sequence to obtain the root cause analysis results and intelligent diagnosis solutions of the cells with poor quality. After integrating all the information, the final answer is generated and considered and returned to the network optimization engineer, and finally a full-cycle closed-loop capability for analyzing and processing wireless cells with poor quality based on natural language input is constructed, thereby improving the work efficiency of front-line network optimization engineers and reducing the investment cost of network operation and maintenance.

[0080] The overall design architecture of the system solution is as follows Figure 3 As shown, the network optimization expert Agent classifies the diagnosis results for training the large language model in the diagnostic agent, making it the network large model of this embodiment. The diagnostic agent is connected to the wireless network tool and can call the wireless network's indicators, logs, configurations, reports and other data. The diagnostic agent has memory and planning capabilities, can decompose sub-goals through thinking chains, and has the ability to automatically call tools to execute instructions. When in use, the user raises the problem of poor quality of the actual wireless network language service cell in the form of natural language. After reading the problem, the AI ​​Agent (diagnostic agent) uses long-term and short-term memory to identify and plan the intention of the problem. The generated plan includes which tools need to be called, and then these tools are called in turn to obtain the results (content and execution instructions). After integrating all the information, it is summarized into two types of instructions: visual instructions and page instructions. The executor executes after receiving the instructions, and finally presents the complete answer to the user. Taking the analysis of the poor quality of wireless voice services as an example, the overall design process of the solution is as follows: Figure 4 As shown, the following will be described in detail Figure 4 Each step in .

[0081] In one implementation, S1, identifying the inquiry intention of the user question regarding poor wireless network quality, and obtaining a number of poor wireless network quality cells that need to be analyzed to answer the user question, specifically includes:

[0082] Receive user questions and extract the inquiry time, inquiry area and inquiry category in the user questions through keywords;

[0083] If the inquiry category includes inquiries about wireless network services and / or indicators, determining whether there is a cell with poor wireless network quality in the inquiry area within the inquiry time;

[0084] If so, it is determined that the user's question includes an intention to inquire about poor wireless network quality, and the identifiers of several cells with poor wireless network quality in the inquiry area within the inquiry time and the real-time poor wireless network quality category are obtained.

[0085] In this embodiment, if Figure 4 As shown, S101: using the network big model as an interactive entrance, inputting the user's wireless voice service poor quality cell problem into the dialog box.

[0086] The problem of poor quality of wireless voice service cells that front-line users need to analyze and deal with is input into the dialog box through natural language interaction. After being decomposed by the cell-level voice service perception indicators, if a certain type of poor quality definition condition is met, the big model will do further quality difference analysis and enter S102: After the user problem is submitted to the AI ​​Agent for reading through the network big model, the intent parameter is extracted by calling the plug-in tool, and then submitted to the network big model for intent recognition.

[0087] After receiving the user's question, the network big model submits it to the AI ​​Agent, which calls the plug-in "Capability Orchestration Engine" to realize intent recognition and parameter extraction. It identifies the user's question intention category (such as indicator query category, quality difference analysis category, etc.) and extracts the key parameters in the question, including time, location, and indicator type.

[0088] By calling the plug-in method, the intent recognition and parameter extraction are realized. The tool function is mainly defined by filling in operationId (unique identifier), summary (overview), description (detailed description), and the use of the tool is defined by filling in the name (identifier), namedescription (detailed description), required (whether it is required), and default (default value) of each parameter under parameters. For example, the indicator query of Suixince (a system developed according to the disclosed method) is divided into two fine-grained intents: road indicator (geographical area) and cell indicator, which extract two parameters: road + indicator and poor quality cell name + indicator respectively.

[0089] Parse the results of plugin parameter extraction. Enumerable parameters can be accurately extracted by configuring synonyms in the dictionary library, configuring regular expressions, context extraction and other rules to complete parameter extraction, and configure in the parameter pattern field.

[0090] Common parameter types can be extracted more accurately through the built-in multiple NER (Named Entity Recognition) models, such as time / time period, address, name, organization name, mobile phone, email address, etc. If there is an entity trigger word in the parameter description, the corresponding entity extraction will be triggered.

[0091] According to the parameter description configured in the plug-in, the big model completes parameter extraction by grouping prompts, such as extracting the "city" field, and assembling it into a request json (JavaScript Object Notation) format that can call the service, and the big model then performs intent recognition. After completion, it goes to step 103.

[0092] An example of obtaining data through the random intention recognition is as follows:

[0093] type: string

[0094] title: "Indicators"

[0095] -in:query

[0096] name:road

[0097] required:True

[0098] description: address, road

[0099] schema:

[0100] type: string

[0101] default: "None"

[0102] responses:

[0103] '200'

[0104] description: Report successfully found

[0105] paths:

[0106] / communityIndexSearch:

[0107] get:

[0108] operationId: query the system index of the test

[0109] summary: Query system indicators with your heart

[0110] description: Query the cell index of the free measurement system

[0111] parameters:

[0112] -in:query

[0113] name:Index

[0114] required:True

[0115] description:|

[0116] Operator base station indicators

[0117] schema:

[0118] default: "None"

[0119] type: string

[0120] title: "Indicators"

[0121] -in:query

[0122] name:community

[0123] required:True

[0124] description:|

[0125] Address, neighborhood

[0126] schema:

[0127] type: string

[0128] default: "None"

[0129] responses:

[0130] '200':

[0131] description: Report successfully found

[0132] In one implementation, determining whether there is a cell with poor wireless network quality in the inquiry area within the inquiry time specifically includes:

[0133] Obtaining a regional wireless network service indicator of the inquiry area within the inquiry time, and judging whether the inquiry area within the inquiry time is a poor wireless network quality area and several real-time poor wireless network quality categories according to the regional wireless network service indicator;

[0134] If the inquiry area is a poor wireless network quality area, obtain the cell-level wireless network service indicator of each wireless network cell in the inquiry area within the inquiry time relative to the several real-time poor wireless network quality categories, and obtain several poor wireless network quality cells in the inquiry area within the inquiry time according to the cell-level wireless network service indicator.

[0135] In this embodiment, the relevant indicator types (wireless network quality category) and quality differences involved in wireless voice services are defined as follows:

[0136] Table 1 Examples of categories of poor wireless network quality and judgment criteria

[0137]

[0138]

[0139] The indicators in the table include all network standard voice service perception indicators for the regional scope (such as provincial and municipal administrative regions), which is completely sufficient for grasping the global network perception. However, how to decompose it into cell-level voice perception indicators under different network standards requires further design and decomposition. After analyzing the components of the overall indicators, it can be further decomposed into cell-level voice service perception indicators. After decomposing the cell-level voice service perception indicators, users can ask the network big model dialog box about the problem of poor quality cells according to the definitions of different indicator categories, such as "connection", "fast connection", "clear hearing", "no dropped calls", etc. If you ask about specific indicator types, the indicator results answered by the big model are different. For example, if you ask about the "connection" indicator situation, the big model will answer the "connection" indicator result, such as the following inquiry method: What is the voice service perception indicator situation in ** District, ** City on July 19? The voice perception situation includes the statistical results of the indicators of "connection", "fast connection", "clear hearing", and "no dropped calls". If any type of indicator meets the definition of poor quality, the big model will do further quality analysis.

[0140] In one embodiment, the method further comprises pre-training a large network model, specifically comprising:

[0141] Collect historical wireless network data, historical wireless network poor quality categories, historical wireless network poor quality diagnosis work orders, wireless network poor quality diagnosis expert knowledge, historical wireless network poor quality diagnosis results, and wireless network API interface for obtaining historical wireless network data;

[0142] Control the network big model, extract the thinking chain question and answer prompt words of the network big model according to the diagnosis process in the historical wireless network poor quality diagnosis work order, combine the historical wireless network data, historical wireless network poor quality categories, thinking chain question and answer prompt words and wireless network poor quality diagnosis expert knowledge to output the historical wireless network poor quality diagnosis results, and fine-tune the parameters of the network big model until the network big model accurately outputs the historical wireless network poor quality diagnosis results;

[0143] Get the wireless network API interface corresponding to the thinking chain question and answer prompt words corresponding to different historical wireless network quality categories and abnormal historical wireless network data in the final network model.

[0144] In this embodiment, the large network model used is a large language model that has been trained and fine-tuned for the quality difference analysis task, that is, the thinking chain ability of the large network model is mainly achieved through incremental pre-training and SFT (supervised fine-tuning) instruction fine-tuning.

[0145] Incremental pre-training: learn new knowledge from a large-scale "unlabeled" text corpus through training, collect and organize more than ten types of front-line business corpora and data including those from the wireless network air interface side, core network side, base station transmission side, bearer network, IP (Internet Protocol) network, backbone network, complaint handling, road testing, service quality analysis, network maintenance, etc., perform text purification, calculate PPL (perplexity) to filter low-quality text, form a network private professional knowledge data set, perform incremental pre-training on the basic model, and inject network expert knowledge.

[0146] SFT instruction fine-tuning: For the specific task of analyzing voice service poor quality cells, the label data such as voice service diagnosis work orders, voice service fault diagnosis knowledge graphs, and voice service performance indicators KPI, perception indicators KQI (Key Quality Indicators) are cleaned into the professional private domain data (network private professional knowledge data set) of the voice service poor quality cell analysis process, including the answer content of the voice service diagnosis work order, etc., to form a structured voice service poor quality diagnosis process text. Subsequently, these texts are used to use the big model in combination with Prompt Engineering to form the ability of thinking chain, and the sub-goals of the task are decomposed based on intent understanding, so that the big model has the ReAct-Agent (reaction intelligent body) ability to analyze, diagnose, perceive, and make flexible decisions on the problem of voice service poor quality cells.

[0147] More specifically, the following empirical knowledge is used to design the network model:

[0148] 1. Experience in judging and handling poor quality communities:

[0149] 1. Ideas for dealing with poor quality communities

[0150] First, define relevant indicators and design KPI monitoring templates on the network management system based on the assessment requirements. Screen out problem cells based on the traffic statistics data. Then, classify and locate different problems. Track and record the process. Track and compare the results after the processing is completed. The loop can only be closed after the problem is solved.

[0151] The following six categories are commonly used in the processing of poor quality cells in LTE (Long Term Evolution) networks. In daily work, voice service / data service performance analysis, resource evaluation, network structure, and neighboring cell parameter verification are given priority, as shown in Table 2 below:

[0152] Table 2 Examples of poor quality cell analysis and processing types

[0153]

[0154]

[0155] 2. Recommendations for judging the threshold of poor quality cells:

[0156] 2.1) RRC (Radio Resource Control) establishment failure: The number of RRC connection establishment failures is greater than 50 times, and the RRC connection establishment success rate is less than 95%;

[0157] 2.2) E-RAB (Evolved Radio Access Bearer) establishment failure: The number of E-RAB establishment failures is greater than 50 times, and the E-RAB establishment success rate is less than 95%;

[0158] 2.3) Drop rate: The number of abnormal releases of UE Context (user equipment context) is greater than 50 times, and the drop rate is greater than 5%;

[0159] 2.4) Handover success rate: The number of cell handover failures is greater than 300 times, and the handover success rate is less than 80%;

[0160] 2.5) Capacity resources: sites involved in key areas

[0161] Condition 1: The average utilization of downlink PRB (Physical Resource Block) is greater than 50%, the average number of valid RRC connections is greater than 30, and the cell downlink busy throughput is greater than 5G (gigabyte, GB);

[0162] Condition 2: The average uplink PRB utilization is greater than 50%, the average number of valid RRC connections is greater than 30, and the cell uplink busy throughput is greater than 1G (gigabyte, GB);

[0163] Condition 3: The maximum number of valid RRC connections is greater than 200;

[0164] Note: Take the average data of the community’s busiest system time for 7 days, and any one of the conditions will suffice.

[0165] 3. Suggestions for checking parameters of poor quality cells:

[0166] Table 3 Poor quality cell parameter check table

[0167]

[0168]

[0169] 4. Ideas and steps for locating the problem of poor quality cells:

[0170] There are usually three main reasons for poor quality of problem cells: radio side parameter configuration problems, core network side configuration problems, and channel environment impacts. Therefore, when you encounter a problem cell, you can roughly follow the steps below to troubleshoot.

[0171] 4.1) Confirm whether the indicators of the entire network have deteriorated. If so, it is necessary to check whether there are any network changes and upgrades in the operations and alarms.

[0172] 4.2) If the indicators of some sites deteriorate and affect the indicators of the entire network, it is necessary to find the TOP poor-quality cells (the cells with the worst quality).

[0173] 4.3) Query the TOP cells (the cells with the worst quality) and the TOP occurrence time period (the time period when the quality difference occurs most frequently) with low RRC connection establishment success rate, low ERAB establishment success rate, low handover success rate, and high dropped call rate.

[0174] 4.4) Check the TOP cell site alarms, board status, RRU (Remote Radio Unit) status, cell operation status, and whether the OMC (Operation & Management Center) operation configuration is normal.

[0175] 4.5) Extract CHR log (log file reflecting the problem of the base station) and analyze the channel quality during access and whether the SINR (Signal to Interference plus Noise Ratio) of SRS (Sounding Reference Signal) is poor (weak coverage) and whether there are TOP users (users with the worst SINR).

[0176] 4.6) Conduct targeted standard signaling tracking and interference detection analysis for TOP cell sites. If there are no abnormalities in standard signaling and interference detection, return the one-click log, standard port tracking and interference detection results to the developer for analysis.

[0177] 4.7) Based on the analysis results of all the above data, the first-line network optimization engineers analyze and eliminate each problem one by one, formulate optimization solutions based on expert experience, and then arrange RF optimization personnel and tower workers to implement the optimization solutions until the problem of poor quality TOP cells is finally solved.

[0178] 2. Experience in designing machine learning algorithms:

[0179] 1. Algorithm level

[0180] The positioning of poor-quality cells in mobile communication networks guided by machine learning needs to be combined with the corresponding KQI data of mobile communication network service perception at the algorithm level, comprehensively analyze the perception portrait of the service, and use machine learning as the guidance algorithm to effectively improve the actual accuracy of the portrait, and build a machine learning-oriented positioning of poor-quality cells in mobile communication networks and a comprehensive perception evaluation and analysis system for poor-quality users. The user portrait is formed and iteratively optimized based on quantitative data such as telephone return visits, user usage perception, and network satisfaction, and the user group and poor-quality content are gradually identified. The impact of poor-quality cells on users and the root cause analysis of poor user perception satisfaction are further optimized, and the optimization solution for poor-quality cells is output. Machine learning pushes the analysis results to the network maintenance unit and combines them with network data analysis, including network performance indicators, network complaint data, coverage and road test quality data, equipment failure rate and impact duration, etc., to enable the mapping of basic elements of customer perception and network elements.

[0181] 1.1) Business perception-related feature portrait

[0182] The characteristic portrait of business perception is mainly based on business models and time models. The network services of mobile communications are mainly streaming media, web browsing, and instant messaging services. The streaming media level services are mainly video and games. The poor perception level scenarios are mainly unable to access the Internet, video freezes, game freezes, and slow web page opening. They have the highest frequency of occurrence in current network complaints, accounting for about 90% of the total complaints. Different services often have different corresponding perception factors W. Among them, the time model implements hour-level perception quantitative comprehensive scoring for different stages of the Internet user's business process to accurately describe the specific situation of user perception.

[0183] 1.2) Basic Quantitative Algorithms for Mobile Communication Network Perception Guided by Machine Learning

[0184] By conducting correlation analysis on user sample data through online KQI and online surveys, we use machine learning as the basic algorithm to effectively construct the indicator set, realize user-business perception coupling, promote the initial formation of the portrait model, implement data verification analysis and model tuning, and finally form a complete perception portrait model, realizing quantitative evaluation under customer perception and effective identification of root causes.

[0185] 2. Identify the level of poor quality users

[0186] 2.1) Identify poor quality services

[0187] For business scenarios such as game freezes, video freezes, inability to access the Internet, slow web page opening, etc., perception indicators are determined separately to achieve a quantitative definition of poor quality services. For services at the level of inability to access the Internet, control plane indicators are used to achieve rapid definition; for services at the perception level related to video freezes, corresponding algorithms are introduced manually, such as playback freezes due to insufficient loading data rate in the client buffer, which contains many perception indicators. Guided by logistic regression machine learning, the threshold of the threshold output by the iterative algorithm is calibrated, and business identification is implemented based on the user's perception of the threshold. For example, if the actual number of video freezes is 0.3 times / minute, the actual waiting time for video playback is more than 5 seconds, and the total duration of video freezes accounts for 30%, it can be determined that the user has a poor perception of one use.

[0188] 2.2) User Perception Rating Model

[0189] Combined with the basic model of wireless Internet perception quantification, the actual number of times that users in the existing network experience poor perception per unit time is obtained, and different dimensions of scoring are implemented in combination with different types of poor quality services to obtain the final score of user perception. For example, the following formula:

[0190] S=∑(K1W1+K2W2+K3W3)

[0191] S represents the user perception score; K1 represents the web browsing service score, and W1 represents the web browsing service weight; K2 represents the streaming media service score, and W2 represents the streaming media service weight; K3 represents the instant messaging service score, and W3 represents the instant messaging service weight.

[0192] From the statistical analysis, we know that user perception varies from good to bad, and most of them are maintained in the middle state, that is, the perception is general or good, which is consistent with the normal distribution or chi-square distribution. In the typical distribution situation, the focus is on the probability distribution. The horizontal axis represents the total number of failures of a single user in various scenarios at the hourly granularity, and the vertical axis is the actual proportion of the number of users corresponding to the number. The vertical axis is the one-dimensional actual variable X, which is the probability of the total number of failures at the user hourly granularity. Combined with the curve of the actual density function change of the random variable X probability, it can be understood that the number of failures can be divided into different intervals, and the corresponding interval users have different perception scores. You can set the user score to be projected into 5 different intervals to obtain the perception score of the current network users. For example, set the filter output to output users with a perception score below 3 points as poor quality users.

[0193] 3. Identify poor quality cells

[0194] 3.1) Basic model of cell perception scoring

[0195] Combined with the number of poor-quality services obtained for users, the user's permanent residence cells are gathered, the actual number of poor-quality services for different users in the cell is input, and the actual number of poor perceptions for each user and the total number of services occurring are obtained. This will reveal that the proportion of poor perception single indicators is higher than X (poor-perception users). Statistical analysis is performed on the number of poor-perception users Ki and the total number of users Mi in each cell, and the proportion of poor-perception users is obtained as Zi = Ki / Mi.

[0196] 3.2) The proportion of users with poor perception can be divided into five different ranges

[0197] The perception scores of users in each corresponding interval are different. Based on the perception algorithm, the cell scores are effectively mapped to 5 different intervals to obtain the cell scores, that is, 5 points are the best perception and 0 points are the worst perception.

[0198] 3.3) Identifying poor quality cells

[0199] Based on this model, the cell scores are obtained, and low-scoring cells with perception scores below 3 points are filtered out, which are poor-quality cells.

[0200] 4. Uncover the root causes of poor quality cells and output solutions

[0201] Network quality and user perception complement each other. User perception can be evaluated comprehensively from aspects such as "station availability, stable residence, and high rate". When a cell has problems such as low access, high disconnection rate, and low rate, the cell is considered to be a poor quality cell that affects user perception. Although these poor quality cells can be identified through indicators such as disconnection rate, wireless access success rate, and rate, it is still very difficult for front-line personnel to locate the causes of poor quality cells because of the variety of reasons that affect user perception, which may even be caused by multiple factors. At present, it is mainly based on expert experience to divide network problems into types such as coverage, interference, capacity, and faults. Combined with CM (Configuration Management), PM (Performance Management), MR (Measurement Report) and other data, the root cause of poor quality cells is marked by formulating corresponding root cause algorithms.

[0202] 4.1) Coverage Class Root Cause Location Algorithm

[0203] Wireless network coverage problems are caused by unreasonable network coverage. In general, network coverage problems in macro base station scenarios can be divided into weak coverage, over-coverage and overlapping coverage. The coverage problems such as overlapping coverage and weak coverage are jointly located through indicators such as RSRP (Reference Signal Received Power), TA (Time Advanced), and overlapping coverage rate.

[0204] ① Analysis of root causes of over-coverage

[0205] At present, the scale construction of 4G (fourth generation mobile communication technology) network has become mature, but the coverage of sites in different scenarios such as urban areas, suburbs / county towns, towns / rural areas is different: urban sites are dense, the distance between suburban / county town sites is large, and the planning of towns / rural sites is mostly wide coverage. Therefore, when formulating the root cause algorithm, different TA thresholds need to be set according to different coverage scenarios. The coverage of urban sites is generally planned to be within 1km. Considering the differences in wireless environment in urban areas, the ratio of TA>1km>25% is used as a prerequisite for determining over-coverage in urban areas; while the coverage area of ​​suburban / county town sites is not too far due to the large distance between sites, but it cannot be too close at the same time. The ratio of TA>2.5km>25% is used as a prerequisite for determining over-coverage in suburban / county towns; although the coverage of towns / rural sites is wide, considering the problem of wireless propagation loss, the farther the signal coverage is, the greater the attenuation and the weaker the signal penetration ability, so the coverage distance should not be too wide, and the ratio of TA>3.5km>25% is used as a prerequisite for determining over-coverage in towns / rural areas.

[0206] ②Analysis of root causes of over-coverage

[0207] Too close coverage means that the signal coverage is far below the planned coverage range. For example, the antenna downtilt angle of the poor quality cell is too large, there are obstructions in the coverage direction, and there are hidden faults in the antenna feed system, which leads to too close coverage. Comprehensive judgment can be made through indicators such as TA and RSRP. The accuracy of using a single indicator to judge the root cause of poor quality is low. Through indicators such as TA and MR, combined with the station spacing and average cell coverage distance in different scenarios, root cause positioning algorithms such as weak coverage, overlapping coverage, over-coverage, and too close coverage can be formulated, as shown in the following table:

[0208] Table 4 Coverage problem judgment basis table

[0209]

[0210] ③Optimize solution output

[0211] For macro base station weak coverage, over-coverage and other coverage problems, there are generally five solutions, provided that the base station and antenna feed system are working properly and the parameters are set reasonably: adjusting the antenna height, adjusting the antenna downtilt angle and azimuth angle, adjusting the RS (Reference Signal) power, relocating the site, and adding RRU. Among them, the solution of adjusting the antenna downtilt angle and azimuth angle is given priority.

[0212] 4.2) Interference root cause location algorithm

[0213] Wireless network interference problems generally include external interference and internal interference. In the root cause algorithm proposed in this paper, the interference problem is located by using the average interference noise power per PRB, which is an indicator that represents the interference intensity.

[0214] ①Interference outside the system

[0215] Generally referred to as external interference, it is caused by external interference sources. External interference is continuous and persistent. When judging external interference, it is necessary to exclude the situation where interference is increased due to factors such as user use during the day. Generally, the interference index at 03:00 when the business is idle is considered for judgment.

[0216] ②Interference within the system

[0217] It refers to the interference caused by co-frequency interference within the system, GPS (Global Positioning System) failure, data configuration error, etc. The time period for interference within the system is not fixed and is greatly affected by factors such as user behavior. Generally, the interference index for the whole day is considered. The root cause location algorithm for external and internal interference is shown in the following table:

[0218] Table 5 Interference problem location basis table

[0219]

[0220] ③Optimize solution output

[0221] For interference problems, there are generally five solutions, provided that the base station and antenna feed system are working properly and the parameters are set reasonably: check for external interference sources, adjust antenna height, adjust antenna downtilt angle and azimuth angle, and adjust RS power. Among them, the solution of checking for external interference sources is given priority.

[0222] 4.3) Root cause location algorithm for poor quality

[0223] Indicators related to poor wireless network quality can directly characterize network quality. For example, CQI (Channel Quality Indicator) characterizes channel quality, and MCS (Modulation and Coding Scheme) characterizes resource scheduling. In general macro base station scenarios, poor network quality problems are mainly caused by weak coverage, over-coverage, and overlapping coverage. In the root cause algorithm based on expert experience, poor quality problems can be jointly located through indicators such as CQI, MCS, and TA.

[0224] ①Quality difference analysis based on coverage type

[0225] The coverage judgment conditions for urban, suburban / county, and township / rural sites in the poor quality category are consistent with the coverage category analysis process. In the poor quality category root cause location algorithm, CQI, MCS, TA indicators, average cell coverage distance, average station spacing in different scenarios and other indicators are combined to locate the root cause of poor quality cells. Please see the table below:

[0226] Table 6 Coverage type quality difference analysis basis table

[0227]

[0228] ②Optimize solution output

[0229] Solving poor quality problems is similar to solving coverage problems. While ensuring that the base station and antenna feed system are working properly and the parameters are set reasonably, optimization is performed by adjusting the antenna height, antenna downtilt angle and azimuth angle, and RS power.

[0230] 4.4) Capacity-based root cause location algorithm

[0231] High load directly affects various network indicators. In the root cause algorithm based on expert experience, load problems such as unbalanced traffic load between carriers and unbalanced traffic load between sectors are jointly located through indicators that characterize the load, such as PRB utilization and RS power.

[0232] ① Load imbalance analysis

[0233] Under the same RRU, the maximum power of two logical cells with the same bandwidth and different frequencies is the same. At this time, if the difference in their PRB utilization is >20%, it is load imbalance; in the actual situation, under the same physical cell, the maximum power of two logical cells with the same coverage may not be the same to avoid interference and other factors, but under normal circumstances, the difference will not exceed 3dB. Combined with the actual situation, the PRB utilization and RS power are used to jointly locate the load balancing problem between carriers. At the same time, this method is also applicable to the load balancing judgment between sectors.

[0234] Based on indicators such as PRB utilization, CCE (Control Channel Element) utilization, number of users, traffic, RS power, and cell merging, root cause algorithms are developed for unbalanced traffic load between carriers, unbalanced traffic load between sectors, large number of users, insufficient bandwidth, high load on logical cells in merged state, unbalanced traffic load between license-restricted sectors, and high CCE utilization, as shown in the following table:

[0235] Table 7 Capacity quality difference analysis basis table

[0236]

[0237] ②Optimize solution output

[0238] In view of the high load problem of macro base stations, there are generally five solutions, provided that the base station and antenna feeder system work normally and the parameters are set reasonably: cell expansion, cell splitting and merging, load balancing between carriers, load balancing between stations, and license expansion. Among them, load balancing between carriers and load balancing between stations are preferred.

[0239] 5. Verification of optimization effect

[0240] An application experiment of the root cause location algorithm was carried out in the pilot area to verify the effectiveness of the above algorithm.

[0241] In one implementation, S2, a root cause diagnosis chain for poor quality is compiled to answer user questions. The root cause diagnosis chain for poor quality includes several diagnosis subtasks for each poor quality cell of the wireless network, specifically including:

[0242] According to the real-time poor wireless network quality category of each poor wireless network quality cell, the corresponding thinking chain question and answer prompt words are obtained;

[0243] According to the question and answer prompts of the thinking chain, a root cause diagnosis chain of poor quality including several diagnosis subtasks for each poor quality cell of the wireless network is compiled;

[0244] Determine the wireless network API interface that needs to be called for each diagnostic subtask.

[0245] In this embodiment, if Figure 4 As shown, S103: using the thinking chain capability of the network big model, plan and arrange the tasks after intent recognition, formulate the analysis tools that need to be called, the thinking chain capability of the network big model has been obtained based on the above pre-training, and determine the docking API (Application Programming Interface) interface of each subtask to obtain real-time data.

[0246] In one embodiment, wherein:

[0247] The plurality of diagnosis subtasks include at least one of: fault diagnosis, parameter diagnosis, coverage diagnosis, quality diagnosis, interference diagnosis, capacity diagnosis, and hidden fault diagnosis;

[0248] The wireless network API interface includes: at least one of: base station side alarm API interface, cell performance KPI indicator API interface, base station main control board API interface, cell CM configuration file API interface, cell performance KPI indicator API interface, MR indicator API interface, uplink interference diagnosis small model API interface, and cell B domain business traffic indicator API interface.

[0249] In this embodiment, when the network model receives the intention result after the Agent calls the plug-in to complete the parameter extraction, it uses the thinking chain ability after incremental pre-training and SFT instruction fine-tuning to plan the analysis task of the problem of poor voice service quality in the cell, such as Figure 5 As shown in the figure, the task planning is mainly decomposed from the following aspects: fault diagnosis, parameter diagnosis, coverage diagnosis, quality diagnosis, interference diagnosis, capacity diagnosis, and hidden fault diagnosis. This is the cell-level diagnosis. After the regional-level indicator analysis, by judging that the regional indicator exceeds a certain threshold, the analysis of the cells in this area is triggered. This is an analysis method from macro to micro. Then enter S104 to analyze the root cause of poor quality.

[0250] In one implementation, obtaining several diagnosis results of several diagnosis subtasks in S3 specifically includes:

[0251] Process several diagnostic subtasks in sequence according to the order in the poor quality root cause diagnosis chain;

[0252] Call the wireless network API interface corresponding to each diagnostic subtask to obtain the corresponding real-time wireless network data;

[0253] Combining real-time wireless network data, real-time wireless network poor quality categories, thinking chain question and answer prompts and wireless network poor quality diagnosis expert knowledge, the real-time wireless network poor quality diagnosis results of each diagnosis subtask are output.

[0254] In this embodiment, if Figure 4 As shown, S104: After the subtask planning is decomposed, the quality problem analysis is carried out on the calling tools formulated in each subtask, and the analysis and diagnosis results are fed back to the AI ​​Agent. The quality problem analysis is carried out based on the different subtasks planned by the thinking chain of the network big model and the calling tools formulated in different subtasks, specifically including:

[0255] 1. Fault diagnosis subtask:

[0256] 1.1) Calling tool:

[0257] ①Base station side alarm API interface

[0258] ②Cell performance KPI indicator API interface

[0259] ③Base station main control board API interface

[0260] 1.2) Quality problem analysis process (such as Figure 6 shown):

[0261] ① Analysis results are normal (6-a)

[0262] Example: The alarm diagnosis analysis is in progress and the analysis result is normal.

[0263] Whether there is a real-time alarm on the wireless side of the cell, and whether the network management is associated with LST ALMAF (a command used to query the current active alarm in the LTE network) alarm: No

[0264] ②Abnormal analysis results (6-b)

[0265] Example: Is there a real-time alarm on the wireless side of the cell, and a real-time associated LST ALMAF alarm on the network management: Yes

[0266] After diagnosis, the community generated a cell unavailability fault alarm at 18:15:20 on April 20, 2024. It is recommended that the maintenance engineer handle the fault and investigate the cause.

[0267] The fault diagnosis subtask calls multiple API interface diagnosis results and then feeds them back to the AI ​​Agent for problem diagnosis classification and induction.

[0268] 2. Parameter diagnosis subtask:

[0269] 2.1) Calling tool:

[0270] ①Cell CM configuration file API interface (including timer parameters, switching parameters and other data)

[0271] ②Cell performance KPI indicator API interface

[0272] 2.2) Quality problem analysis process (such as Figure 7 shown):

[0273] ① Analysis results are normal (7-a)

[0274] Example: During parameter diagnosis analysis, the analysis results are normal.

[0275] a) T300 = 1000 (reasonable range 0 ~ 1000ms), parameter configuration is normal

[0276] b) T304 = 500 (reasonable range 500 ~ 2000ms), parameter configuration is normal

[0277] c) N310 = 10 (reasonable range 6 to 20), parameter configuration is normal

[0278] d) N311 = 1 (reasonable range 1 to 2), parameter configuration is normal

[0279] e) Minimum access level = -126 (reasonable range -110 to -128), parameter configuration is normal

[0280] f) A2 event RSRP threshold (different system data) = -120 (reasonable range -107 to -118), parameter configuration is normal

[0281] g) B2 event RSRP threshold 1 (data) = -122 (reasonable range -110 to -118), parameter configuration is normal

[0282] h) B2 event RSRP threshold 2 (data) = -118 (reasonable range -110 to -118), parameter configuration is normal

[0283] ②Abnormal analysis results (7-b)

[0284] Example: During parameter diagnosis analysis, the analysis result is abnormal

[0285] a) T300 = 2000 (reasonable range 0 ~ 1000ms), parameter configuration is abnormal

[0286] b) T304 = 500 (reasonable range 500 ~ 2000ms), parameter configuration is normal

[0287] c) N310 = 10 (reasonable range 6 to 20), parameter configuration is normal

[0288] d) N311 = 1 (reasonable range 1 to 2), parameter configuration is normal

[0289] e) Minimum access level = -126 (reasonable range -110 to -128), parameter configuration is normal

[0290] f) A2 event RSRP threshold (different system data) = -120 (reasonable range -107 to -118), parameter configuration is normal

[0291] g) B2 event RSRP threshold 1 (data) = -122 (reasonable range -110 to -118), parameter configuration is normal

[0292] h) B2 event RSRP threshold 2 (data) = -118 (reasonable range -110 to -118), parameter configuration is normal

[0293] After diagnosis, it was found that the configuration of cell parameter T300 exceeded the reasonable range, which had a certain impact on the RRC establishment success rate and voice service perception. It is recommended to reset it within a reasonable range.

[0294] The diagnostic subtask calls multiple API interfaces based on the parameter diagnostic results, and then feeds them back to the AI ​​Agent for problem diagnosis classification and induction.

[0295] 3. Coverage diagnosis subtask:

[0296] 3.1) Calling tool:

[0297] ①MR indicator API interface

[0298] ②Cell performance KPI indicator API interface

[0299] 3.2) Quality problem analysis process (such as Figure 8 shown):

[0300] ① Analysis results are normal (8-a)

[0301] Example: Coverage diagnosis analysis is in progress and the analysis result is normal.

[0302] a) Weak coverage ratio (RSRP <= -105 ratio) = 12.6% (reasonable range 0-20%), the indicator is normal

[0303] b) Total tadv sampling points = 9426

[0304] c) Over-coverage sampling points = 2465

[0305] d) Over-coverage rate = 16.15% (reasonable range 0-20%), normal indicators

[0306] e) Near coverage sampling points = 25

[0307] f) Near coverage rate = 1.28% (reasonable range 0-20%), the indicator is normal

[0308] ②Abnormal analysis results

[0309] Example 1: In the coverage diagnosis analysis, the analysis result is abnormal (8-b). (When the weak coverage ratio exceeds the reasonable range, and the over-coverage ratio and near-coverage ratio are within the normal range)

[0310] a) Weak coverage ratio (RSRP <= -105 ratio) = 22.6% (reasonable range 0-10%), there is a weak coverage problem

[0311] b) Total tadv sampling points = 9426

[0312] c) Over-coverage sampling points = 2465

[0313] d) Over-coverage rate = 6.15% (reasonable range 0-20%), normal indicators

[0314] e) Near coverage sampling points = 25

[0315] f) Near coverage rate = 1.28% (reasonable range 0-20%), the indicator is normal

[0316] After diagnosis, the cell has a weak coverage problem, which has a certain impact on the RRC establishment success rate and voice service perception. Query the cell parameters:

[0317] (a) Standing height = 50 meters

[0318] (b) Mechanical downtilt = 2 degrees, electronic downtilt = 0 degrees, azimuth = 60 degrees It is recommended to improve signal strength by properly optimizing the downtilt and azimuth angles.

[0319] The above covered diagnostic subtasks call multiple API interface diagnostic results, and then feed back to the AI ​​Agent for problem diagnosis classification and induction.

[0320] Example 2: In the coverage diagnosis analysis, the analysis result is abnormal (8-c). (When the weak coverage ratio exceeds the reasonable range, the over-coverage ratio also exceeds the reasonable range, but the near coverage ratio is within the normal range)

[0321] a) Weak coverage ratio (RSRP <= -105 ratio) = 22.6% (reasonable range 0-10%), there is a weak coverage problem

[0322] b) Total tadv sampling points = 9426

[0323] c) Over-coverage sampling points = 2465

[0324] d) Over-coverage rate = 26.15% (reasonable range 0-20%), abnormal indicators

[0325] e) Near coverage sampling points = 25

[0326] f) Near coverage rate = 1.28% (reasonable range 0-20%), the indicator is normal

[0327] After diagnosis, the cell has a weak coverage problem. Since the coverage is too far, users at the edge of the coverage access, which has a certain impact on the RRC establishment success rate and voice service perception. Query the cell parameters:

[0328] (a) Standing height = 50 meters

[0329] (b) Mechanical downtilt = 2 degrees, electronic downtilt = 0 degrees, azimuth = 60 degrees. It is recommended to optimize the downtilt and azimuth to control the coverage and enhance the signal strength.

[0330] The above covered diagnostic subtasks call multiple API interface diagnostic results, and then feed back to the AI ​​Agent for problem diagnosis classification and induction.

[0331] Example 3: In the coverage diagnosis analysis, the analysis result is abnormal (8-d). (When the weak coverage ratio exceeds the reasonable range, the near coverage ratio also exceeds the reasonable range, but the over-coverage ratio is within the normal range)

[0332] a) Weak coverage ratio (RSRP <= -105 ratio) = 22.6% (reasonable range 0-10%), there is a weak coverage problem

[0333] b) Total tadv sampling points = 9426

[0334] c) Over-coverage sampling points = 2465

[0335] d) Over-coverage rate = 6.15% (reasonable range 0-20%), abnormal indicators

[0336] e) Near coverage sampling points = 25

[0337] f) Near coverage rate = 21.28% (reasonable range 0-20%), abnormal indicator

[0338] After diagnosis, the cell has a weak coverage problem. It may be blocked by an object in front, resulting in weak signal users accessing the coverage too close, which has a certain impact on the RRC establishment success rate and voice service perception. Query the cell parameters:

[0339] (a) Standing height = 50 meters

[0340] (b) Mechanical downtilt = 2 degrees, electronic downtilt = 0 degrees, azimuth = 60 degrees

[0341] It is recommended to conduct an on-site survey to see if the antenna is blocked, and to appropriately adjust the azimuth or technically modify the antenna position to improve signal strength.

[0342] The above covered diagnostic subtasks call multiple API interface diagnostic results, and then feed back to the AI ​​Agent for problem diagnosis classification and induction.

[0343] Example 4: In the coverage diagnosis analysis, the analysis result is abnormal (8-e). (When the weak coverage ratio exceeds the reasonable range, and the over-coverage and near-coverage ratios also exceed the reasonable range)

[0344] a) Weak coverage ratio (RSRP <= -105 ratio) = 22.6% (reasonable range 0-10%), there is a weak coverage problem

[0345] b) Total tadv sampling points = 9426

[0346] c) Over-coverage sampling points = 2465

[0347] d) Over-coverage rate = 26.15% (reasonable range 0-20%), abnormal indicators

[0348] e) Near coverage sampling points = 25

[0349] f) Near coverage rate = 21.28% (reasonable range 0-20%), abnormal indicator

[0350] After diagnosis, the cell has weak coverage, mainly due to the coverage being too close or too far, which has a certain impact on the RRC establishment success rate and voice service perception. Query the cell parameters:

[0351] (a) Standing height = 50 meters

[0352] (b) Mechanical downtilt = 2 degrees, electronic downtilt = 0 degrees, azimuth = 60 degrees

[0353] Recommendation 1: Improve signal strength by optimizing downtilt and azimuth to control coverage

[0354] Suggestion 2: It is recommended to conduct an on-site survey to see if the antenna is blocked, and appropriately adjust the azimuth or technically modify the antenna position to improve signal strength.

[0355] The above covered diagnostic subtasks call multiple API interface diagnostic results, and then feed back to the AI ​​Agent for problem diagnosis classification and induction.

[0356] 4. Quality diagnosis subtask:

[0357] 4.1) Calling the tool:

[0358] ①MR indicator API interface

[0359] ②Cell performance KPI indicator API interface

[0360] 4.2) Quality problem analysis process (such as Fig. 9 shown):

[0361] ① Analysis results are normal (9-a)

[0362] Example: During quality diagnosis analysis, the analysis results are normal.

[0363] a) MRO-SINR≥0 sampling point ratio = 98.87% (reasonable range 90% to 100%), the indicator is normal

[0364] b) Overlap coverage ratio (RSRP = -105) = 14.78% (reasonable range 0% ~ 20%)

[0365] c) Total MR samples = 9553

[0366] d) Number of MOD30 interference sampling points = 893, MOD30 interference rate = 9.34% (reasonable range 0% to 10%)

[0367] ②Abnormal analysis results

[0368] Example 1: In the quality diagnosis analysis, the analysis result is abnormal (9-b). (When the MRO-SINR≥0 sampling point ratio exceeds the reasonable range, and the overlap coverage ratio exceeds the reasonable range, but the MOD30 interference rate is within the normal range)

[0369] a) MRO-SINR ≥ 0 sampling point ratio = 87.25% (reasonable range 90% to 100%), there is a downlink quality problem

[0370] b) Overlap coverage ratio (RSRP = -105) = 24.78% (reasonable range 90% ~ 100%)

[0371] c) Total MR samples = 9553

[0372] d) Number of MOD30 interference sampling points = 893, MOD30 interference rate = 9.34% (reasonable range 0% to 10%)

[0373] After diagnosis, the cell has downlink quality problems. Due to overlapping coverage, co-channel interference is caused, which has a certain impact on the RRC establishment success rate and voice service perception. Query the cell parameters:

[0374] (a) Standing height = 50 meters

[0375] (b) Mechanical downtilt = 2 degrees, electronic downtilt = 0 degrees, azimuth = 60 degrees

[0376] If the station is too high or the antenna is hung too high, the coverage may be too far and cause interference to surrounding stations. If the station is too close or too far, the main service cell may not be obvious. Due to the overlapping coverage caused by the superposition of multiple signals, it is recommended to solve the downlink quality problem by optimizing the downtilt angle and azimuth angle of this station and surrounding stations.

[0377] The above covered diagnostic subtasks call multiple API interface diagnostic results, and then feed back to the AI ​​Agent for problem diagnosis classification and induction.

[0378] Example 2: In the quality diagnosis analysis, the analysis result is abnormal (9-c). (When the MRO-SINR≥0 sampling point ratio exceeds the reasonable range, and the MOD30 interference rate exceeds the reasonable range, but the overlap coverage ratio is within the normal range)

[0379] a) MRO-SINR ≥ 0 sampling point ratio = 87.25% (reasonable range 90% to 100%), there is a downlink quality problem

[0380] b) Overlap coverage ratio (RSRP = -105) = 14.78% (reasonable range 90% ~ 100%)

[0381] c) Total MR samples = 9553

[0382] d) Number of MOD30 interference sampling points = 1280, MOD30 interference rate = 13.39% (reasonable range 0% to 10%)

[0383] After diagnosis, the cell has downlink quality problems. Due to the interference caused by MOD30 interference, the RRC establishment success rate and voice service perception are affected to a certain extent. Query the cell parameters:

[0384] (a) Standing height = 50 meters

[0385] (b) Mechanical downtilt = 2 degrees, electronic downtilt = 0 degrees, azimuth = 60 degrees

[0386] Suggestion 1: If the station is high or the antenna is hung high, it is easy to cover too far and cause MOD30 interference to surrounding sites. It is recommended to solve the downlink quality problem by optimizing the downtilt angle and azimuth angle of this site and surrounding sites.

[0387] Recommendation 2: By optimizing the beam coordination strategy, network coverage performance and user service experience can be further improved.

[0388] Recommendation 3: Avoid MOD30 interference by optimizing scrambling codes in neighboring cells

[0389] The above quality diagnosis subtask calls multiple API interface diagnosis results, and then feeds back to the AI ​​Agent for problem diagnosis classification and induction.

[0390] 5. Interference diagnosis subtask:

[0391] 5.1) Calling tool:

[0392] ① Uplink interference diagnosis small model API interface

[0393] ②Cell performance KPI indicator API interface

[0394] 5.2) Interference problem analysis process (such as Figure 10-15 shown):

[0395] ①Analysis results are normal

[0396] Example: During interference diagnosis analysis, the analysis result is normal.

[0397] Average interference noise power per prb (dbm) = -108 (reasonable range <= -105dbm), the indicator is normal

[0398] ②Abnormal analysis results

[0399] Example 1: During interference diagnosis analysis, the analysis result is abnormal (15-a). (When the average interference noise power per prb exceeds the reasonable range, and the uplink interference type = none)

[0400] a) Average interference noise power per prb (dbm) = -98.5 (reasonable range <= -105dbm), abnormal indicator

[0401] b) Uplink Interference Type = None

[0402] After diagnosis, the cell has uplink interference problem (uplink interference type = none), which has a certain impact on the RRC establishment success rate and voice service perception. Query the hourly PRB-level waveform of the cell (implemented by calling the uplink interference diagnosis small model API interface) as follows Fig.10 shown.

[0403] It is recommended to conduct a frequency scan on site to determine the source of interference, and then observe whether the 5G (fifth-generation mobile communication technology) wireless connection rate indicator returns to normal after eliminating the interference.

[0404] The above interference diagnosis subtask calls multiple API interface diagnosis results, and then feeds back to the AI ​​Agent for problem diagnosis classification and induction.

[0405] Example 2: During interference diagnosis analysis, the analysis result is abnormal (15-b). (When the average interference noise power per prb exceeds the reasonable range, and the uplink interference type = spurious interference)

[0406] After diagnosis, it is found that there is an uplink interference problem in the cell, which is spurious interference, which has a certain impact on the RRC establishment success rate and voice service perception. The hourly PRB-level waveform of the cell is queried (by calling the uplink interference diagnosis small model API interface). Fig.11 shown.

[0407] It is recommended to follow the steps below to troubleshoot interference:

[0408] a) Check the isolation between the interference cell and the interference source base station. Generally, the horizontal isolation is changed to vertical isolation, requiring the spatial distance to be at least 3m horizontally or at least 1m vertically.

[0409] b) Check whether a filter is installed at the interference source. You can reduce stray interference by adding a bandpass filter to the interference source.

[0410] The above interference diagnosis subtask calls multiple API interface diagnosis results, and then feeds back to the AI ​​Agent for problem diagnosis classification and induction.

[0411] Example 3: During interference diagnosis analysis, the analysis result is abnormal (15-c). (When the average interference noise power per prb exceeds the reasonable range, and the uplink interference type = blocking interference)

[0412] After diagnosis, it is found that the cell has uplink interference problem, which is blocking interference, and has a certain impact on the RRC establishment success rate and voice service perception. The hourly PRB level waveform of the cell is queried (by calling the uplink interference diagnosis small model API interface). Fig.12 shown.

[0413] It is recommended to follow the steps below to troubleshoot interference:

[0414] a) Check whether the interfered base station is equipped with a filter, and reduce interference by installing a filter

[0415] b) Check the isolation between the two systems and change horizontal isolation to vertical isolation

[0416] c) Check the anti-interference capability of the equipment in the affected cell, and replace the affected RRU with a device with stronger anti-jamming capability

[0417] The above interference diagnosis subtask calls multiple API interface diagnosis results, and then feeds back to the AI ​​Agent for problem diagnosis classification and induction.

[0418] Example 4: During the interference diagnosis analysis, the analysis result is abnormal (15-d). (When the average interference noise power per prb exceeds the reasonable range, and the uplink interference type = GPS lock interference)

[0419] After diagnosis, the cell has uplink interference problem, which is GPS lock interference, and has a certain impact on the success rate of RRC establishment. Query the hourly PRB level waveform of the cell (by calling the uplink interference diagnosis small model API interface) as follows Fig.13 shown.

[0420] It is recommended to follow the steps below to troubleshoot interference:

[0421] a) Check for cells where the base station clock desynchronization alarm coincides with the surrounding large-scale interference and service loss.

[0422] b) Check whether the frame offset configuration of the problem cell is consistent with that of the surrounding base stations

[0423] c) If the frame offsets are consistent, it is necessary to check for base station clock failure and collect the main control board log for further investigation.

[0424] The above interference diagnosis subtask calls multiple API interface diagnosis results, and then feeds back to the AI ​​Agent for problem diagnosis classification and induction.

[0425] Example 5: During interference diagnosis analysis, the analysis result is abnormal (15-e). (When the average interference noise power per prb exceeds the reasonable range, and the uplink interference type = long-distance co-channel interference)

[0426] After diagnosis, the cell has uplink interference problem, which is long-distance co-frequency interference, which has a certain impact on the success rate of RRC establishment. Query the hourly PRB-level waveform of the cell (by calling the uplink interference diagnosis small model API interface) as follows Fig.14 shown.

[0427] It is recommended to follow the steps below to troubleshoot interference:

[0428] 1) Check whether the existing network has completed frequency clearing, and clear the frequencies that have not been completely cleared.

[0429] 2) Check the base station antenna type and replace it

[0430] 3) Check the configuration of the cell parameters for co-frequency interference. You can adjust the antenna transmission power, adjust the reuse distance of the co-frequency adjacent cells, adjust the antenna downtilt angle, azimuth angle, and lower the antenna platform.

[0431] The above interference diagnosis subtask calls multiple API interface diagnosis results, and then feeds back to the AI ​​Agent for problem diagnosis classification and induction.

[0432] 6. Capacity diagnosis subtask:

[0433] 6.1) Calling tool:

[0434] ①Cell B domain service traffic indicator API interface

[0435] ②Cell performance KPI indicator API interface

[0436] 6.2) Capacity problem analysis process (such as Fig.16 shown):

[0437] ①Analysis results are normal

[0438] Example: During capacity diagnostic analysis, the analysis result is normal (16-a).

[0439] a) Uplink PRB utilization = 8.52% (reasonable range 0-50%, take the maximum value in the whole network within 24 hours), the indicator is normal

[0440] b) Downlink PRB utilization = 2.28% (reasonable range 0-50%, take the maximum value in the whole network within 24 hours), the indicator is normal

[0441] c)arfcn=636664

[0442] d) Average number of 5G (fifth generation mobile communication technology) users = 515

[0443] e) 5G (fifth generation mobile communication technology) traffic (5G uplink traffic + 5G downlink traffic) = 8.2Gb

[0444] f) Total traffic of China Unicom (China Unicom uplink traffic + China Unicom downlink traffic) = 6Gb

[0445] g) Total telecom traffic (telecom uplink traffic + telecom downlink traffic) = 2.2Gb

[0446] ②Abnormal analysis results

[0447] Example 1: During capacity diagnosis analysis, the analysis result is abnormal (16-b). (When the uplink PRB utilization rate exceeds the reasonable range, but the downlink PRB utilization rate is within the normal range)

[0448] a) Uplink PRB utilization = 88.56% (reasonable range 0-50%, take the maximum value in the whole network within 24 hours), the indicator is abnormal

[0449] b) Downlink PRB utilization = 12.49% (reasonable range 0-50%, take the maximum value in the whole network within 24 hours), the indicator is normal

[0450] c)arfcn=636664

[0451] d) Average number of 5G users = 928

[0452] e) 5G traffic (5G uplink traffic + 5G downlink traffic) = 86.2Gb

[0453] f) Total traffic of China Unicom (China Unicom uplink traffic + China Unicom downlink traffic) = 50.1Gb

[0454] g) Total telecom traffic (telecom uplink traffic + telecom downlink traffic) = 36.1Gb

[0455] After diagnosis, it was found that the cell had an overload problem in the uplink, which had a certain impact on the 5G voice perception service. It was recommended to optimize and adjust from the following aspects:

[0456] a) Check whether the uplink dynamic scheduling algorithm is enabled

[0457] b) Check the load balancing parameters, mainly from the three aspects of power parameters, switching reselection parameters and MLB (Mobility Load Balance) parameters, and configure the parameters reasonably

[0458] c) Neighborhood integrity check, mainly analyzing the integrity of different frequency points, the accuracy of external neighbor definition and the integrity of internal neighbor addition.

[0459] d) Structural rationality analysis, mainly from the accuracy of engineering parameters, azimuth, downtilt angle and TA coverage rationality four aspects of verification

[0460] The capacity diagnosis subtask calls multiple API interface diagnosis results and then feeds them back to the AI ​​Agent for problem diagnosis classification and induction.

[0461] Example 2: During capacity diagnosis analysis, the analysis result is abnormal (16-c). (When the downlink PRB utilization exceeds the reasonable range, but the uplink PRB utilization is within the normal range)

[0462] a) Uplink PRB utilization = 9.33% (reasonable range 0-50%, take the maximum value in the whole network within 24 hours), the indicator is normal

[0463] b) Downlink PRB utilization = 82.49% (reasonable range 0-50%, take the maximum value in the whole network within 24 hours), the indicator is abnormal

[0464] c)arfcn=636664

[0465] d) Average number of 5G users = 1,008

[0466] e) 5G traffic (5G uplink traffic + 5G downlink traffic) = 109.5Gb

[0467] f) Total traffic of China Unicom (China Unicom uplink traffic + China Unicom downlink traffic) = 80.2Gb

[0468] g) Total telecom traffic (telecom uplink traffic + telecom downlink traffic) = 29.3Gb

[0469] After diagnosis, it was found that the downlink load in the cell was too high, which had a certain impact on the 5G voice perception service. It was recommended to optimize and adjust from the following aspects:

[0470] a) Check whether the downlink dynamic scheduling algorithm is enabled

[0471] b) Load balancing parameter verification, mainly from the power parameters, switching reselection parameters and MLB parameters, and reasonable parameter configuration

[0472] c) Expansion of cells with the same or different frequencies

[0473] d) Newly built base stations share the load

[0474] The capacity diagnosis subtask calls multiple API interface diagnosis results and then feeds them back to the AI ​​Agent for problem diagnosis classification and induction.

[0475] Example 3: During capacity diagnosis analysis, the analysis result is abnormal (16-d). (When both the uplink PRB utilization and the downlink PRB utilization are beyond the reasonable range)

[0476] a) Uplink PRB utilization = 85.50% (reasonable range 0-50%, take the maximum value in the whole network within 24 hours), the indicator is abnormal

[0477] b) Downlink PRB utilization = 72.85% (reasonable range 0-50%, take the maximum value in the whole network within 24 hours), the indicator is abnormal

[0478] c)arfcn=636664

[0479] d) Average number of 5G users = 1,180

[0480] e) 5G traffic (5G uplink traffic + 5G downlink traffic) = 509.8Gb

[0481] f) Total traffic of China Unicom (China Unicom uplink traffic + China Unicom downlink traffic) = 428.5Gb

[0482] g) Total telecom traffic (telecom uplink traffic + telecom downlink traffic) = 81.3Gb

[0483] After diagnosis, it was found that the downlink load in the cell was too high, which had a certain impact on the 5G voice perception service. It was recommended to optimize and adjust from the following aspects:

[0484] a) Check whether the uplink and downlink dynamic scheduling algorithms are enabled

[0485] b) Load balancing parameter verification, mainly from the power parameters, switching reselection parameters and MLB parameters, and reasonable parameter configuration

[0486] c) Neighborhood integrity check, mainly analyzing the integrity of different frequency points, the accuracy of external neighbor definition and the integrity of internal neighbor addition.

[0487] d) Structural rationality analysis, mainly from the accuracy of engineering parameters, azimuth, downtilt angle and TA coverage rationality four aspects of verification

[0488] e) Expansion of cells with the same or different frequencies

[0489] f) Newly built base stations share the load

[0490] The capacity diagnosis subtask calls multiple API interface diagnosis results and then feeds them back to the AI ​​Agent for problem diagnosis classification and induction.

[0491] 7. Hidden fault diagnosis subtask:

[0492] 7.1) Calling the tool:

[0493] ①Cell B domain (business domain, business support system) business traffic indicator API interface

[0494] ②Cell performance KPI indicator API interface

[0495] ③Base station side alarm API interface

[0496] ④MR indicator API interface

[0497] 7.2) Hidden fault problem analysis process (such as Fig.17 shown):

[0498] ①Analysis results are normal

[0499] Example: In the hidden fault diagnosis analysis, the analysis result is normal (17-a).

[0500] All indicators of the community are normal, and no hidden faults have occurred.

[0501] ②Abnormal analysis results

[0502] Example 1: In the analysis of hidden fault diagnosis, the analysis result is abnormal (17-b). (When the uplink PRB utilization rate exceeds the reasonable range, but the downlink PRB utilization rate is within the normal range)

[0503] After diagnosis, the community is suspected to have hidden faults. It is recommended to check one by one according to the following steps:

[0504] a) If the RRU software crashes, it is recommended to power off and reset the RRU in the background and front end before observing the indicators.

[0505] b) If the problem is still not solved after resetting, it is necessary to go to the site to check the antenna and feeder system, including whether the antenna is placed in a reasonable position, whether the waterproof tape at the antenna and feeder head is damaged, whether the feeder is damaged, etc.

[0506] c) Replace the antenna or exchange the antenna.

[0507] d) If no hidden faults are found after on-site inspection, it is recommended to obtain relevant indicators on the DNS (Domain Name System), CDN (Content Delivery Network), core network and SP (Service Provider) server side according to the end-to-end analysis method, and analyze them section by section to find the above problems on the wireless air interface side.

[0508] e) Check the base station terminal compatibility parameters and change the base station SRS reconfiguration mode switch to 0 (OFF).

[0509] The implicit diagnosis subtask calls multiple API interface diagnosis results and then feeds them back to the AI ​​Agent for problem diagnosis classification and induction.

[0510] In one implementation, S4, obtaining root causes of poor quality and poor quality processing suggestions for several wireless network poor quality cells according to several diagnosis results, and outputting answers to user questions according to the root causes of poor quality and the poor quality processing suggestions, specifically including:

[0511] Output voice and / or text results of the root causes of poor quality and poor quality treatment suggestions for several wireless network poor quality cells according to the abnormal root causes and treatment suggestions of the diagnosis subtasks whose sequential comprehensive diagnosis results in the poor quality root cause diagnosis chain are abnormal;

[0512] Several wireless network quality-poor cells are displayed on a map and their root causes are marked. Based on the quality-poor treatment suggestions, a page instruction for repairing the root cause is generated and executed automatically or when triggered by the user.

[0513] In this embodiment, if Figure 4 As shown, S105: the AI ​​Agent performs comprehensive summarization processing on the diagnosis results to obtain two types of instructions: visualization instructions and page instructions (the two instructions exist at the same time, one is the interface that needs to be presented, and the other is the instruction operation). The executor executes the instructions after receiving them, and performs the actions through the RPA (Robotic Process Automation) intelligent agent, and finally presents the complete answer to the user.

[0514] In the overall architecture, the executor is responsible for the use of external tools as well as the execution of client instructions. After AIAgent identifies and plans the user's question intention, it then identifies and plans the processing result, determines the type of operation and the object of the operation involved, and then extracts the object unique identifier from the instruction library. After processing all the data in the format agreed upon with the instruction execution SDK (Software Development Kit), two types of instruction data are obtained: page instructions and visual instructions. Fig.18 shown.

[0515] Page instructions are instructions for the agent to present the specific steps of the agent's operation on the existing system page. After receiving the user's instructions, the client uses the page instructions to execute the SDK tool ("complete the user's query / operation goal through a question"). According to the received page instructions, the SDK tool is executed in the order in which the instructions are arranged. There is an error correction mechanism to ensure the correct and complete execution of the instructions. In the event of an instruction execution failure, the error correction mechanism will use a backup plan to correct the error and execute the instruction to ensure that each instruction can be fully executed.

[0516] The visualization instructions are used to visualize the content summarized and summarized by the large model intelligent agent planning on the client. The client extracts the corresponding components from the visualization component library according to the visualization type, inputs the visualization content, and forms a user-friendly visualization page.

[0517] After all instructions are executed, the client will provide feedback to the AI ​​Agent, including the operation results and any abnormal situations. The AI ​​Agent learns based on the feedback and continuously optimizes the accuracy of operation steps and intent recognition. At the same time, in this process, it will also ensure that the entire process complies with data protection regulations and business compliance requirements to protect user data security. The above instruction types and operation objects are all stored in the instruction library.

[0518] Instruction library: instruction operation type, operable objects, and visual components. The operable objects in the instruction library include all operable objects in the system page. In the early stage, all operable objects are stored in the instruction library through manual entry assisted by automated tools. Each operable object is mapped accurately to the business operation process or function name to facilitate the intelligent agent to identify the intent.

[0519] After the AI ​​Agent plans the operation steps (these steps are detailed to each specific operation action), it will identify the intent based on the specific nouns in the specific operation steps, query the instruction library to obtain the object unique identifier corresponding to the semantics, assemble it into the data format agreed with the client and return it to the client. After the client receives the instruction data, if the instruction is a page instruction, it will find the operated object on the page based on the object unique identifier, and perform the corresponding action on the operated object according to the agent operation type.

[0520] The instruction library contains two major categories of instructions: page instructions and visual instructions.

[0521] Page instructions, instructions for agent users to operate the page. The entire instruction content consists of agent operation type (required) + agent operation object (required) + agent content (optional), among which agent operation type and agent operation object are required items, and agent content is optional.

[0522] The proxy operation types include: 1) Click: Use the mouse to click a specific object on the web system page, including left click, right click, and double click. 2) Text filling: Enter text in an editable control, imitating keyboard behavior. 3) Copy. 4) Cut. 5) Paste. 6) Get text: Get the content of the operated object. 7) Get object: Get the operated object; 8) Scroll: Scroll the control item vertically or horizontally, imitating the behavior of the mouse wheel, so that the covered content that is not in the current visible area is visible. 9) Drag: Imitate the behavior of the user dragging the component. 10) Jump: Imitate the user jumping page, including jumping the current page, jumping to a new tab, and jumping to a new window. 11) Page matching: Match to determine whether the current page is the target page; 12) Screenshot: Screenshot the current screen / browser view interface. 13) Confirmation prompt: It can be used to provide a secondary confirmation prompt before performing some irreversible or high-risk operations as a protection mechanism.

[0523] The proxy operation objects include: 1) All operable component objects on the system page: single selection, multiple selection, input box, date, map, chart, icon, table, button, etc. 2) Page scroll bar: horizontal scroll bar and vertical scroll bar.

[0524] Proxy content: some data used to fill in the proxy operation object and assign values ​​to the proxy operation object. The above proxy operation types, proxy operation objects, and proxy contents can be freely combined according to business needs.

[0525] Visualization instructions, including visualization type + visualization content.

[0526] Visualization types include: 1) Chart instructions, which can be subdivided into: bar chart, line chart, discounted bar chart, pie chart, multi-dimensional chart... 2) Tables, which can be subdivided into: no paging, front-end paging, remote paging; support filtering, filtering types include single selection, multiple selection, text, date, label, etc.;... 3) Single selection. 4) Multiple selection. 5) Button. 6) List. 7) Text. 8) Download. 9) Image...

[0527] Visualization content is generally transmitted in the form of JSON data. The specific data fields and hierarchical relationships are related to the visualization type.

[0528] The specific page composition is as follows Fig.19 As shown, the page rendering effect is as follows Fig. 20 shown.

[0529] This embodiment uses the definition of poor quality cells for wireless network 4 / 5G voice perception services and constructs a poor quality wireless cell analysis and processing agent based on a large model AI Agent architecture. This method can be designed based on any network or higher evolution network voice perception service.

[0530] Embodiment 2:

[0531] like Figure 2 As shown, the present disclosure provides a wireless network quality difference analysis and processing device, the device includes an intelligent agent AI Agent, the intelligent agent AI Agent includes a network large model, and the network large model includes:

[0532] Intention recognition module 1, used to identify the inquiry intention of the user's question about the poor quality of the wireless network, and obtain a number of poor quality wireless network cells that need to be analyzed to answer the user's question;

[0533] The chain arrangement module 2 is connected to the intention recognition module 1 and is used to compile a poor quality root cause diagnosis chain to answer user questions. The poor quality root cause diagnosis chain includes several diagnosis subtasks for each poor quality cell of the wireless network;

[0534] The diagnosis analysis module 3 is connected to the chain arrangement module 2 and is used to obtain a number of diagnosis results of a number of diagnosis subtasks, each diagnosis result including an abnormality judgment, abnormal root cause and processing suggestion of each diagnosis subtask;

[0535] The summarizing and outputting module 4 is connected to the diagnostic and analyzing module 3, and is used to obtain the root causes of poor quality and poor quality treatment suggestions of several wireless network poor quality cells according to several diagnostic results, and output answers to user questions according to the root causes of poor quality and the poor quality treatment suggestions.

[0536] In one embodiment, the intention recognition module 1 specifically includes:

[0537] An extraction unit, used to receive user questions and extract the inquiry time, inquiry area and inquiry category in the user questions through keywords;

[0538] A judgment unit connected to the extraction unit, configured to judge whether there is a poor quality wireless network cell in the inquiry area within the inquiry time if the inquiry category includes an inquiry about wireless network services and / or indicators;

[0539] The acquisition unit is connected to the judgment unit and is used to determine if the user's question includes an inquiry intention about poor wireless network quality, and obtain the identifiers of several poor wireless network quality cells in the inquiry area within the inquiry time and the real-time poor wireless network quality category.

[0540] In one implementation, the determination unit specifically includes:

[0541] A regional level judgment subunit is used to obtain the regional level wireless network service index of the inquiry area within the inquiry time, and judge whether the inquiry area within the inquiry time is a poor wireless network quality area according to the regional level wireless network service index, and several real-time poor wireless network quality categories;

[0542] The cell-level judgment subunit is connected to the area-level judgment subunit and is used to obtain the cell-level wireless network service indicators of each wireless network cell in the inquiry area within the inquiry time relative to the several real-time wireless network quality poor categories if the inquiry area is a wireless network quality poor area, and obtain several wireless network quality poor cells in the inquiry area within the inquiry time according to the cell-level wireless network service indicators.

[0543] In one embodiment, the device further includes a network large model pre-training module, specifically including:

[0544] A collection unit, used to collect historical wireless network data, historical wireless network poor quality categories, historical wireless network poor quality diagnosis work orders, wireless network poor quality diagnosis expert knowledge, historical wireless network poor quality diagnosis results, and a wireless network API interface for obtaining historical wireless network data;

[0545] The fine-tuning unit is connected to the acquisition unit and is used to control the network large model, extract the thinking chain question and answer prompt words of the network large model according to the diagnosis process in the historical wireless network poor quality diagnosis work order, combine the historical wireless network data, the historical wireless network poor quality category, the thinking chain question and answer prompt words and the wireless network poor quality diagnosis expert knowledge to output the historical wireless network poor quality diagnosis result, and fine-tune the parameters of the network large model until the network large model accurately outputs the historical wireless network poor quality diagnosis result;

[0546] The corresponding unit is connected to the fine-tuning unit and is used to obtain the thinking chain question and answer prompt words corresponding to different historical wireless network quality categories in the final network model and the wireless network API interface corresponding to the abnormal historical wireless network data.

[0547] In one embodiment, the chain arrangement module 2 specifically includes:

[0548] The chain word unit is used to obtain corresponding thinking chain question and answer prompt words according to the real-time poor wireless network quality category of each poor wireless network quality cell;

[0549] The chain task unit is connected with the chain word unit and is used to compile a poor quality root cause diagnosis chain including a plurality of diagnosis subtasks for each poor quality cell of the wireless network according to the thought chain question and answer prompt words;

[0550] The interface calling unit is connected to the chain task unit and is used to determine the wireless network API interface that needs to be called by each diagnosis subtask.

[0551] In one embodiment, wherein:

[0552] The plurality of diagnosis subtasks include at least one of: fault diagnosis, parameter diagnosis, coverage diagnosis, quality diagnosis, interference diagnosis, capacity diagnosis, and hidden fault diagnosis;

[0553] The wireless network API interface includes: at least one of: base station side alarm API interface, cell performance KPI indicator API interface, base station main control board API interface, cell CM configuration file API interface, cell performance KPI indicator API interface, MR indicator API interface, uplink interference diagnosis small model API interface, and cell B domain business traffic indicator API interface.

[0554] In one embodiment, the diagnostic analysis module 3 specifically includes:

[0555] A diagnostic sequence unit, used to process several diagnostic subtasks in sequence according to the sequence in the poor quality root cause diagnosis chain;

[0556] The real-time data unit is connected to the diagnostic sequence unit and is used to call the wireless network API interface corresponding to each diagnostic subtask to obtain the corresponding real-time wireless network data;

[0557] The diagnosis output unit is connected to the real-time data unit and is used to combine the real-time wireless network data, the real-time wireless network quality poor category, the thinking chain question and answer prompt words and the wireless network quality poor diagnosis expert knowledge to output the real-time wireless network quality poor diagnosis result of each diagnosis subtask.

[0558] In one embodiment, the output module 4 specifically includes:

[0559] A language output unit, configured to output voice and / or text results of root causes of poor quality and poor quality treatment suggestions for several wireless network poor quality cells according to abnormal root causes and treatment suggestions of the diagnosis subtasks whose sequential comprehensive diagnosis results in the poor quality root cause diagnosis chain are abnormal;

[0560] The instruction output unit is connected to the language output unit, and is used to display several wireless network poor quality cells on the map, mark their poor quality root causes, generate page instructions for repairing the poor quality root causes according to the poor quality processing suggestions, and execute the page instructions automatically or under user triggering.

[0561] Embodiment 3:

[0562] Embodiment 3 of the present disclosure provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the wireless network quality difference analysis and processing method as described in Embodiment 1 is implemented, or the wireless network quality difference analysis and processing device as described in Embodiment 2 is implemented.

[0563] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program elements or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable read only memory), flash memory or other memory technology, CD-ROM (Compact Disc Read-Only Memory), digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer.

[0564] In addition, the present disclosure may also provide a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor runs the computer program stored in the memory, the processor executes the wireless network quality difference analysis processing method as described in Example 1. The computer device may be the wireless network quality difference analysis processing device as described in Example 2.

[0565] The memory is connected to the processor, the memory may be a flash memory or a read-only memory or other memory, and the processor may be a central processing unit or a single-chip microcomputer.

[0566] Embodiments 1-3 of the present disclosure provide a wireless network quality difference analysis and processing method, a wireless network quality difference analysis and processing device, and a computer-readable storage medium, and adopt a large intelligent agent model to provide a network optimization auxiliary tool with a high degree of intelligence. By identifying the user's problem intention, the root cause diagnosis of the wireless network quality difference is initiated, and by compiling a root cause diagnosis chain for the quality difference, decomposing the diagnosis tasks, and then synthesizing the diagnosis results, the root cause of the wireless network quality difference and processing suggestions are obtained. This can be used to assist network optimization engineers in completing network optimization work and improve network optimization efficiency.

[0567] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present disclosure, but the present disclosure is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and substance of the present disclosure, and these modifications and improvements are also considered to be within the scope of protection of the present disclosure.

Claims

1. A wireless network quality analysis and processing method, characterized in that: The method is applied to an AI Agent, including using a large network model: Identify the inquiry intention of the user's question about the poor quality of the wireless network, and obtain several cells with poor wireless network quality that need to be analyzed to answer the user's question; A root cause diagnosis chain for poor quality is compiled to answer user questions, the root cause diagnosis chain for poor quality includes several diagnosis subtasks for each poor quality cell of the wireless network; Obtaining multiple diagnostic results of multiple diagnostic subtasks, each diagnostic result including an abnormality judgment, abnormal root cause and processing suggestion of each diagnostic subtask; The root causes of poor quality and suggestions for handling poor quality of several wireless network poor quality cells are obtained according to several diagnosis results, and answers to user questions are output according to the root causes of poor quality and the suggestions for handling poor quality.

2. The method according to claim 1, characterized in that: Identify the user's intention to inquire about poor wireless network quality in the question, and obtain several cells with poor wireless network quality that need to be analyzed to answer the user's question, including: Receive user questions and extract the inquiry time, inquiry area and inquiry category in the user questions through keywords; If the inquiry category includes inquiries about wireless network services and / or indicators, determining whether there is a cell with poor wireless network quality in the inquiry area within the inquiry time; If so, it is determined that the user's question includes an intention to inquire about poor wireless network quality, and the identifiers of several cells with poor wireless network quality in the inquiry area within the inquiry time and the real-time poor wireless network quality category are obtained.

3. The method according to claim 2, characterized in that Determine whether there is a cell with poor wireless network quality in the inquiry area during the inquiry time, specifically including: Obtaining a regional wireless network service indicator of the inquiry area within the inquiry time, and judging whether the inquiry area within the inquiry time is a poor wireless network quality area and several real-time poor wireless network quality categories according to the regional wireless network service indicator; If the inquiry area is a poor wireless network quality area, obtain the cell-level wireless network service indicator of each wireless network cell in the inquiry area within the inquiry time relative to the several real-time poor wireless network quality categories, and obtain several poor wireless network quality cells in the inquiry area within the inquiry time according to the cell-level wireless network service indicator.

4. The method according to any one of claims 1 to 3, characterized in that: The method also includes pre-training a large network model, specifically including: Collect historical wireless network data, historical wireless network poor quality categories, historical wireless network poor quality diagnosis work orders, wireless network poor quality diagnosis expert knowledge, historical wireless network poor quality diagnosis results, and wireless network API interface for obtaining historical wireless network data; Control the network big model, extract the thinking chain question and answer prompt words of the network big model according to the diagnosis process in the historical wireless network poor quality diagnosis work order, combine the historical wireless network data, historical wireless network poor quality categories, thinking chain question and answer prompt words and wireless network poor quality diagnosis expert knowledge to output the historical wireless network poor quality diagnosis results, and fine-tune the parameters of the network big model until the network big model accurately outputs the historical wireless network poor quality diagnosis results; Get the wireless network API interface corresponding to the thinking chain question and answer prompt words corresponding to different historical wireless network quality categories and abnormal historical wireless network data in the final network model.

5. The method according to claim 4, characterized in that A root cause diagnosis chain for poor quality is compiled to answer user questions. The root cause diagnosis chain includes several diagnostic subtasks for each poor quality cell of the wireless network, including: According to the real-time poor wireless network quality category of each poor wireless network quality cell, the corresponding thinking chain question and answer prompt words are obtained; According to the question and answer prompts of the thinking chain, a root cause diagnosis chain of poor quality including several diagnosis subtasks for each poor quality cell of the wireless network is compiled; Determine the wireless network API interface that needs to be called for each diagnostic subtask.

6. The method according to claim 5, wherein: The plurality of diagnosis subtasks include at least one of: fault diagnosis, parameter diagnosis, coverage diagnosis, quality diagnosis, interference diagnosis, capacity diagnosis, and hidden fault diagnosis; The wireless network API interface includes: at least one of: base station side alarm API interface, cell performance KPI indicator API interface, base station main control board API interface, cell CM configuration file API interface, cell performance KPI indicator API interface, MR indicator API interface, uplink interference diagnosis small model API interface, and cell B domain business traffic indicator API interface.

7. The method according to claim 5 or 6, characterized in that: Obtain several diagnostic results of several diagnostic subtasks, including: Process several diagnostic subtasks in sequence according to the order in the poor quality root cause diagnosis chain; Call the wireless network API interface corresponding to each diagnostic subtask to obtain the corresponding real-time wireless network data; Combining real-time wireless network data, real-time wireless network poor quality categories, thinking chain question and answer prompts and wireless network poor quality diagnosis expert knowledge, the real-time wireless network poor quality diagnosis results of each diagnosis subtask are output.

8. The method according to claim 7, characterized in that The root causes of poor quality and suggestions for handling poor quality of several wireless network poor quality cells are obtained based on several diagnosis results, and answers to user questions are output based on the root causes of poor quality and suggestions for handling poor quality, including: Output voice and / or text results of the root causes of poor quality and poor quality treatment suggestions for several wireless network poor quality cells according to the abnormal root causes and treatment suggestions of the diagnosis subtasks whose sequential comprehensive diagnosis results in the poor quality root cause diagnosis chain are abnormal; Several wireless network quality-poor cells are displayed on a map and their root causes are marked. Based on the quality-poor treatment suggestions, a page instruction for repairing the root cause is generated and executed automatically or when triggered by the user.

9. A wireless network quality analysis and processing device, characterized in that: The device includes an intelligent agent AIAgent, the intelligent agent AIAgent includes a network model, and the network model includes: An intention recognition module is used to recognize the inquiry intention of the user's question about the poor quality of the wireless network, and obtain a number of poor quality wireless network cells that need to be analyzed to answer the user's question; A chain arrangement module, connected to the intent recognition module, is used to compile a poor quality root cause diagnosis chain to answer user questions, where the poor quality root cause diagnosis chain includes several diagnosis subtasks for each poor quality cell of the wireless network; The diagnosis analysis module is connected to the chain orchestration module and is used to obtain a number of diagnosis results of a number of diagnosis subtasks, each diagnosis result including anomaly judgment, anomaly root cause and processing suggestions of each diagnosis subtask; The summarizing output module is connected to the diagnosis and analysis module, and is used to obtain the root causes of poor quality and poor quality treatment suggestions of several wireless network poor quality cells according to several diagnosis results, and output answers to user questions according to the root causes of poor quality and the poor quality treatment suggestions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the wireless network quality difference analysis and processing method according to any one of claims 1 to 8 is implemented.

Images