Multi-agent road network quality analysis processing method and device and medium

Abstract

The application provides a multi-agent road network quality analysis processing method and device and medium, and relates to the technical field of network. The method comprises the following steps: a central control agent generates a planning request according to a user query for querying road network quality, and sends the planning request to a planning agent; the planning agent acquires service sub-agent information for analyzing road network quality in a service agent according to the planning request, plans a calling decision for each service sub-agent according to the service sub-agent information, and sends the calling decision to the service agent; the service agent interacts with the planning agent to realize calling each service sub-agent according to the calling decision, acquiring a task execution result of each service sub-agent analyzing road network quality, and sending each task execution result to the central control agent; and the central control agent controls generating a summary report of road network quality analysis according to all the task execution results, and provides the summary report to answer the user query.

Description

Technical Field

[0001] This application relates at least to the field of network technology, and in particular to a method, apparatus and medium for multi-agent road network quality analysis and processing. Background Technology

[0002] Currently, the industry has proposed the concept of wireless network drive test agents. Users raise actual wireless network road problems in the form of natural language. After understanding the intent of the problem, the LLM begins to reason in order to answer the user's question. However, this method adopts a single agent mode, which makes it difficult to accurately identify complex user questions and is prone to illusion problems when calling tools. Summary of the Invention

[0003] To address the aforementioned shortcomings, this application provides a multi-agent road network quality analysis and processing method, apparatus, and medium to solve the following technical problem: how to improve the accuracy of wireless network road test agents in identifying complex user questions and avoid hallucinations.

[0004] In a first aspect, this application provides a method for multi-agent road network quality analysis and processing, the method comprising:

[0005] The central control agent generates a planning request based on the user query used to inquire about the quality of the road network, and sends the planning request to the planning agent;

[0006] The planning agent obtains information about the business sub-agents used to analyze road network quality from the business agent according to the planning request, plans the calling decisions for each business sub-agent according to the business sub-agent information, and sends the calling decisions to the business agent.

[0007] The business intelligence agent interacts with the planning intelligence agent to invoke each business sub-intelligence agent according to the invocation decision, obtain the task execution results of each business sub-intelligence agent analyzing the road network quality, and send the task execution results to the central control intelligence agent.

[0008] The central control intelligent agent generates a summary report on road network quality analysis based on the results of all task executions, and provides the summary report to answer user queries.

[0009] Furthermore, the central control intelligent agent generates a planning request based on the user query used to inquire about road network quality, specifically including:

[0010] The central control intelligent agent receives user queries for inquiring about road network quality and sends the user queries to the pre-processing module;

[0011] The preprocessing module identifies the intent of the user query, identifies the preset intent category corresponding to the intent, determines the key parameters to be extracted and their preset description requirements based on the preset intent category, and obtains the key parameters that meet the preset description requirements by combining the user query and the road network database. The intent and key parameters are then returned to the central control intelligent agent.

[0012] The central control intelligent agent expands the user query based on intent and key parameters. The expanded content includes at least one of the following: querying road network quality indicators, analyzing quality problems, locating root causes of problems, and providing solutions, in order to form a planning request.

[0013] Furthermore, the planning agent obtains information about the business sub-agents used for analyzing road network quality from the business agent based on the planning request, and plans the invocation decisions for each business sub-agent based on the business sub-agent information, specifically including:

[0014] The planning agent obtains the available business sub-agents that meet the planning request and adapts to the business agent. It queries the RAG knowledge base for the functional descriptions of the available business sub-agents that adapt to the business agent and requests the LLM to perform task planning based on the planning request and functional descriptions.

[0015] Based on the planning request and functional description, LLM selects the appropriate business sub-intelligent agents from the available business sub-intelligent agents that meet the planning request, arranges the execution order of each business sub-intelligent agent and formulates the calling requirements for each business sub-intelligent agent, so as to obtain the calling decision for each business sub-intelligent agent and return it to the planning agent.

[0016] Furthermore, the planning agent obtains the available business sub-agents that meet the planning request and then queries the RAG knowledge base for the functional descriptions of the available business sub-agents that meet the planning request, specifically including:

[0017] The planning agent obtains the currently available service sub-agents, including the adapted service agent of all service sub-agents used to analyze road network quality. All service sub-agents used to analyze road network quality include: road problem type analysis sub-agent, road problem master cell analysis sub-agent, alarm analysis sub-agent, neighbor cell analysis sub-agent, and antenna and feeder optimization sub-agent.

[0018] The functional description of the sub-agent for querying road problem types from the RAG knowledge base includes: inputting the road segment ID and user query time to query key performance indicator data of the road network, analyzing the types of road network quality problems based on the key performance indicator data of the road network, and generating road network quality problem analysis results in a preset format.

[0019] The functional description of the sub-agent for querying the RAG knowledge base for the main control cell analysis of road problems includes: obtaining all coverage cells and sampling point data of the road based on the road segment ID, and identifying the TOPN main control cells of the road from all coverage cells based on the sampling point data;

[0020] The functional description of querying the RAG knowledge base for the alarm analysis sub-agent includes: obtaining alarm information of the road TOPN master cell at the time of user query, analyzing the impact of each alarm message, and identifying alarms that cause road network quality problems;

[0021] The functional description of the neighbor cell analysis sub-agent for querying the RAG knowledge base includes: obtaining neighbor cells that correspond to the ECI of the main control cell of the road TOPN, establishing cell pairs that match the main control cell and neighbor cells, and identifying abnormal cell pairs with missing neighbor cell matching.

[0022] The function description of the antenna and feeder optimization sub-agent retrieved from the RAG knowledge base includes: generating an antenna and feeder optimization and adjustment scheme for the road network based on the type of road network quality problem, the alarms that caused the road network quality problem, and the abnormal cells with missing neighboring cells.

[0023] Furthermore, the invocation requirements for each adapted business sub-agent are defined, specifically including:

[0024] Based on the road type being one of National Highway / Provincial Highway / County Road / Township Road / Expressway / High-speed Railway / Elevated Road, the calling requirements for the road problem type analysis sub-agent include: querying key performance indicator data of road networks of different road segment lengths, and locating road segments with road network quality problems;

[0025] The calling requirements for the main control cell analysis sub-agent for road problems include: obtaining the percentage of sampling points for road segments with road network quality problems, and obtaining the top 8 and top 3 main control cells with the largest percentage of sampling points;

[0026] The requirements for calling the alarm analysis sub-agent include: obtaining alarm information from the top 3 master cells within a preset time period, and diagnosing whether the alarm information affects the services of the master cells;

[0027] The calling requirements for the neighbor cell analysis sub-agent include: obtaining whether any of the top 8 main control cells have been missing from the neighbor cell configuration of the top 3 main control cells;

[0028] The requirements for calling the antenna feeder optimization sub-agent include: formulating an antenna feeder optimization and adjustment scheme based on the sampling contribution of associated cells of road segments with road network quality problems, the distance of the problematic road segments, and the voltage level.

[0029] Furthermore, the business intelligence agent interacts with the planning intelligence agent to invoke various business sub-intelligent agents based on the invocation decision, obtain the task execution results of each business sub-intelligent agent analyzing road network quality, and send the task execution results to the central control intelligence agent. Specifically, this includes:

[0030] The planning intelligent agent controls the business intelligent agent to call each adapted business sub-intelligent agent according to the calling order. The business intelligent agent obtains the task execution results of each business sub-intelligent agent in analyzing road network quality according to the calling requirements. The planning intelligent agent controls the acquisition of each task execution result and sends it to the central control intelligent agent.

[0031] Furthermore, the planning intelligent agent controls the business intelligent agent to invoke each adapted business sub-intelligent agent according to the invocation order. The business intelligent agent obtains the task execution results of each business sub-intelligent agent analyzing road network quality according to the invocation requirements. The planning intelligent agent controls and obtains the task execution results and sends them to the central control intelligent agent, specifically including:

[0032] The planning agent controls the business agent to select the appropriate business sub-agent for the current call based on the calling order;

[0033] The business intelligence agent requests the LLM to generate call parameters including the currently called adaptive business sub-intelligence based on the call requirements. The call parameters are sent to the post-processing module for auditing and calibration. The audited and calibrated call parameters are then used to input the currently called adaptive business sub-intelligence to obtain the task execution result of the currently called adaptive business sub-intelligence.

[0034] The planning agent determines whether to call the next adaptable business sub-agent based on the current task execution result. If yes, the control business agent calls the next adaptable business sub-agent; otherwise, it sends the currently obtained task execution results to the central control agent.

[0035] Furthermore, the central control intelligent agent generates a summary report on road network quality analysis based on the results of all task executions, and provides this summary report to answer user queries, specifically including:

[0036] The central control agent checks the execution results of each task, assembles all task execution results, and sends them to the summarizing agent.

[0037] The agent requests the LLM to summarize and generalize the results of all task executions, generating a summary report on road network quality analysis.

[0038] The summary agent sends the summary report to the central control agent, which then outputs the summary report to the user in response to the user's query.

[0039] Secondly, this application provides a multi-agent road network quality analysis and processing device, the device comprising:

[0040] The central control intelligent agent is used to generate a planning request based on the user query for querying the quality of the road network, and then send the planning request to the planning intelligent agent.

[0041] The planning agent connects with the central control agent and is used to obtain information on the business sub-agents used to analyze road network quality from the business agent according to the planning request. Based on the information of the business sub-agents, it plans the calling decisions for each business sub-agent and sends the calling decisions to the business agent.

[0042] The business intelligence agent connects with the planning intelligence agent and is used to interact with the planning intelligence agent to invoke each business sub-intelligence agent according to the invocation decision, obtain the task execution results of each business sub-intelligence agent in analyzing road network quality, and send the task execution results to the central control intelligence agent.

[0043] The central control intelligent agent is also used to control the generation of a summary report on road network quality analysis based on the results of all task executions, and to provide the summary report to answer user queries.

[0044] Thirdly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the multi-agent road network quality analysis and processing method described above.

[0045] This application provides a multi-agent road network quality analysis and processing method, device, and medium. It includes at least a central control agent, a planning agent, and business agents. The central control agent addresses collaboration issues among the multiple agents, decomposes complex user problems into planning agents to solve complex task execution path problems, and then, based on the assigned task objectives, multiple business agents precisely invoke multiple tools to achieve end-to-end intelligent decision-making through dynamic task planning and autonomous execution. Finally, the entire detailed analysis process and its results are summarized and provided to frontline network personnel in report form to assist in optimizing road network problems and improve problem-solving efficiency. Attached Figure Description

[0046] Figure 1 This is a flowchart of a multi-agent road network quality analysis and processing method according to an embodiment of this application;

[0047] Figure 2 This is a comparison diagram of traditional road testing and virtual road testing technologies according to an embodiment of this application;

[0048] Figure 3 This is an architecture diagram of a single-agent automatic road test according to an embodiment of this application;

[0049] Figure 4 This is an architecture diagram of a multi-agent road network quality analysis and processing according to an embodiment of this application;

[0050] Figure 5 This is a schematic diagram of the structure of a computer-readable storage medium according to an embodiment of this application;

[0051] Figure 6 This is a schematic diagram of the structure of a computer device according to an embodiment of this application. Detailed Implementation

[0052] To enable those skilled in the art to better understand the technical solution of this application, the embodiments of this application will be further described in detail below with reference to the accompanying drawings.

[0053] It is understood that the specific embodiments and accompanying drawings described herein are merely for explaining this application and are not intended to limit this application.

[0054] It is understood that, without conflict, the various embodiments and features in the embodiments of this application can be combined with each other.

[0055] It is understood that, for ease of description, only the parts relevant to this application are shown in the accompanying drawings, while parts unrelated to this application are not shown in the drawings.

[0056] It is understood that each module or unit involved in the embodiments of this application 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.

[0057] It is understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of this application may occur in a different order than that marked in the accompanying drawings.

[0058] It is understood that the flowcharts and block diagrams of this application illustrate the possible architecture, functions, and operations of systems, apparatuses, devices, and methods according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, unit, program segment, or code, containing executable instructions for implementing the specified function. Furthermore, each block or combination of blocks in the block diagrams and flowcharts may be implemented using a hardware-based device to implement the specified function, or using a combination of hardware and computer instructions.

[0059] It is understood that the modules and units involved in the embodiments of this application can be implemented by software or by hardware. For example, the modules and units can be located in the processor.

[0060] Example 1:

[0061] like Figure 1 As shown, this application provides a multi-agent road network quality analysis and processing method, the method comprising:

[0062] S1. The central control intelligent agent generates a planning request based on the user query used to inquire about the quality of the road network and sends the planning request to the planning intelligent agent.

[0063] S2. The planning agent obtains information about the business sub-agents used to analyze road network quality from the business agent according to the planning request, plans the calling decisions for each business sub-agent according to the business sub-agent information, and sends the calling decisions to the business agent.

[0064] S3. The business intelligence agent interacts with the planning intelligence agent to invoke each business sub-intelligence agent according to the invocation decision, obtain the task execution results of each business sub-intelligence agent in analyzing the road network quality, and send the task execution results to the central control intelligence agent.

[0065] S4, the central control intelligent agent generates a summary report of road network quality analysis based on the results of all task executions, and provides the summary report to answer user queries.

[0066] In this embodiment, the provided method sets up at least a central control intelligent agent, a planning intelligent agent, and a business intelligent agent. The central control intelligent agent solves the collaboration problem between multiple intelligent agents, decomposes the user's complex problems into the planning intelligent agent, solves the complex task execution path problem, and then, based on the multiple business intelligent agents, completes the precise invocation of multiple tools according to the assigned task objectives, realizing full-link intelligent decision-making for dynamic task planning and autonomous execution. Finally, the entire detailed analysis process and its results are summarized and provided to front-line network personnel in the form of a report to assist in guiding the optimization of road network problems and improve the efficiency of problem closure.

[0067] Specifically, this embodiment provides a method and apparatus for road network quality analysis and processing based on wireless network drive testing multi-agents.

[0068] With the rapid development of communication services, the network scale is constantly expanding, the types of services are increasing, and the number of users is rising. Network optimization work is becoming more and more complex and the costs are constantly increasing. Traditional network optimization models can no longer meet the current needs of network development, and there is an urgent need for a revolutionary solution to change the status quo of network optimization work.

[0069] like Figure 2The diagram shows existing road testing technologies, including traditional road testing and virtual road testing. Virtual road testing is mainly achieved through the following steps: 1) Route definition: By importing maps and combining manual definition, urban roads are rasterized and defined using historical road testing data; 2) Data extraction: Pedestrians and vehicles are constantly moving on urban roads, and their mobile phones continuously report MR data to the network. The system filters and selects data containing AGPS information, associates this data with call detail records (CDRs), and matches it to the road raster; 3) Result display: Statistical analysis is performed on the selected data, and GIS is used to render the map raster with different colors to present coverage, anomalies, and other information, providing a precise assessment of network performance.

[0070] While using virtual road testing systems to analyze road problems offers some efficiency improvements compared to traditional manual road testing, it places certain demands on the learning level of optimization personnel. When the road testing scope expands and more problems are discovered, it becomes difficult to meet the requirements for timely analysis and optimization loops. Therefore, automated and intelligent methods are needed to improve the efficiency of road problem analysis and optimization.

[0071] like Figure 3 As shown, to address the shortcomings of the MDT (Multi-Level Device) virtual road test system, this paper proposes a concept based on intelligent agents for wireless network road testing. Intelligent agents are introduced into the self-intelligent business process of perception, cognition, decision-making, execution, and evaluation in the virtual automated road test system. By synergizing large models and toolset capabilities, the proportion of manual operation during the operation and updating of the self-intelligent business process is reduced, thereby improving the level of self-intelligence. Users pose actual wireless network road problems in natural language. After understanding the intent of the problem, the LLM (Limited Learning Model) begins reasoning. Through interaction with the knowledge base, a planning task is generated, primarily obtaining a toolset list, specific functions, and descriptions of the tools from the knowledge base. The LLM then decides which tools to use, observes the results after execution, and repeatedly considers and reasons until the diagnostic analysis and recommendation goals for the road problem are met, ultimately presenting a complete answer to the user. The aim is to enable dynamic task planning through interaction between the LLM and the knowledge base, autonomously deciding which tools to use, observing the results after execution, and repeatedly considering and reasoning until the diagnostic analysis and recommendation goals for the road problem are met. This constructs a full-cycle closed-loop capability for analyzing and processing road problems in automated wireless network road testing, driven by LLM natural language interaction. However, this single-agent model is simplistic and struggles to accurately identify complex user problems. It is also prone to illusions when invoking tools, exhibiting the following main shortcomings:

[0072] 1. Architecture Comparison: Only one LLM agent is responsible for the entire process, including intent understanding, reasoning, tool invocation, execution observation and result generation. All decisions and operations are concentrated on one agent.

[0073] 2. Task decomposition and processing capability: LLM needs to handle all task steps on its own, from intent understanding to final output. There is no clear task decomposition mechanism. For complex wireless network drive testing problems (such as multi-index analysis and cross-regional diagnosis), a single agent may not be able to effectively decompose tasks, resulting in low processing efficiency or incomplete decision-making.

[0074] 3. Context Management and State Maintenance: When LLM performs tool calls and observations in a loop, it needs to manage the context and historical interactions itself, but it lacks a dedicated context management mechanism. In long dialogues or multi-step tasks, the context may be lost or confused, affecting the accuracy of inference.

[0075] 4. Error handling and parameter auditing: It relies on the self-observation and reasoning of LLM to correct errors, but there is no dedicated post-processing inspection mechanism. The results of tool calls may be directly used for the next step. The lack of parameter verification makes it easy to spread errors and affect the quality of road test diagnosis.

[0076] 5. Resource efficiency and scalability: Concentrating all computation and inference loads on a single LLM can lead to high response latency, especially with concurrent user requests. System expansion requires retraining or optimizing individual agents, which is costly and results in poor scalability, making it difficult to cope with large-scale or real-time drive test requirements.

[0077] 6. Knowledge Retrieval and Dynamic Programming: LLM interacts with the knowledge base to obtain toolsets, but the retrieval process may be one-off or simple and cannot dynamically adapt to changes in tasks. Tool selection depends on the real-time reasoning of LLM, which may miss key tools and affect the accuracy of drive test analysis.

[0078] 7. User experience and result quality: The final result relies on the repeated reasoning of LLM, but due to too many iterations or limitations in reasoning, the answer may be incomplete or the delay may be high. Users may have to wait for a long time. At the same time, the output results may lack in-depth summary, which reduces the work efficiency of network optimization engineers.

[0079] Therefore, to address the shortcomings of single-agent automated road testing, a multi-agent road network quality analysis and processing method based on wireless network automated road testing is proposed. This method uses a central control agent to solve the collaboration problem among multiple agents, decomposing complex user problems into planning agents to address complex task execution path issues. Then, based on the assigned task objectives, multiple business agents precisely invoke multiple tools through React's "think-act-observe loop" mechanism, achieving end-to-end intelligent decision-making for dynamic task planning and autonomous execution. Finally, a summary agent completes the entire detailed analysis process, providing a report to frontline network personnel to assist in optimizing road network issues and improve problem-solving efficiency.

[0080] The overall design of the system solution is as follows: Figure 4 As shown, the user raises a practical wireless network road problem in natural language. The process involves: user request → pre-processing enhancement → central agent context management → planning agent decomposition → RAG (Retrieval-augmented Generation) to obtain business components → LLM (Large Language Model) to generate a plan → iterative execution of subtasks (business agent RAG acquisition toolset) → ReAct (Reasoning+Acting, a large model agent architecture paradigm) mechanism LLM interaction → post-processing parameter audit → tool call execution → agent summary → return of complete solution to correctly return the user's problem.

[0081] In one implementation, the central control agent generates a planning request based on a user query for road network quality, specifically including:

[0082] The central control intelligent agent receives user queries for inquiring about road network quality and sends the user queries to the pre-processing module;

[0083] The preprocessing module identifies the intent of the user query, identifies the preset intent category corresponding to the intent, determines the key parameters to be extracted and their preset description requirements based on the preset intent category, and obtains the key parameters that meet the preset description requirements by combining the user query and the road network database. The intent and key parameters are then returned to the central control intelligent agent.

[0084] The central control intelligent agent expands the user query based on intent and key parameters. The expanded content includes at least one of the following: querying road network quality indicators, analyzing quality problems, locating root causes of problems, and providing solutions, in order to form a planning request.

[0085] In this embodiment, as Figure 4 As shown:

[0086] Step 101: Input the user's complex task request query into the dialog box, which will be processed by the central control intelligent agent.

[0087] Frontline users can input the wireless network issues they need to analyze and resolve into the dialog box using natural language interaction. The following types of queries are considered as the same question:

[0088] 1) Please help me analyze the problematic section of Changning Road 21789-S1 in Shanghai during the second week of August.

[0089] 2) How was the problem on the section of Changning Road 21789-S1 in Shanghai resolved last week?

[0090] After the user enters their question into the dialog box, step 101 ends and the process proceeds to step 102.

[0091] Step 102: After receiving the user's question, the central control intelligent agent directly sends it to the preprocessing unit for intent recognition and parameter extraction. After completion, it proceeds to step 103.

[0092] Step 103: Preprocessing After receiving the user query, perform entity intent recognition and question expansion.

[0093] For example, if the question is "Please help me analyze the problematic section of Changning Road 21789-S1 in Shanghai during the second week of April", the specific steps are as follows:

[0094] Based on the input text, intent recognition is performed, indicating that the path required is appid=sxc-XX. The complete question and answer is then sent to the toolchain's preprocessing. Here, appid=scx-XX refers to the toolchain's definition of the content for different intent queries. For example, XX=00000 represents a query for "province / regional road indicators", XX=00001 represents a query for "prefecture-level city / regional road indicators", XX=00002 represents a query for "county-level city / regional road indicators", XX=00003 represents a query for "single road indicator", and XX=00004 represents a query for "single road problem section indicator". Other intent recognition identifiers are shown in Table 1 below.

[0095] Table 1: Preset Intent IDs and Examples of Corresponding Intents

[0096]

[0097] Therefore, it is necessary to identify the user's intent category (such as single road indicator query, single road problem segment analysis, etc.) from the user's input query, and to preprocess and extract key parameters from the question, including time, location, indicator type, and problem analysis. That is, from the complete question text, an address parsing tool is used to obtain the city name, single road name, and specific time, such as "Shanghai", "Changning Road", "Second week of August", "Road problem", and "Problem analysis".

[0098] Intent recognition and parameter extraction are achieved through a plugin approach. The tool's functionality is defined by filling in the operationId (unique identifier), summary, and description. The usage of the tool is defined by filling in the name, namedescription, required, and default values ​​for each parameter under parameters.

[0099] The results of parameter extraction from the large model are parsed. Enumerable parameters can be accurately extracted by configuring a thesaurus in the dictionary, and parameters can be extracted by configuring rules such as regular expressions and contextual parameter extraction, which are configured in the parameter pattern field. Commonly used parameter types can be extracted with relatively high accuracy through various built-in NER models, such as time / time period, address, person name, organization name, mobile phone number, email address, etc. If there are entity trigger words in the parameter description, the corresponding entity extraction will be triggered.

[0100] According to the parameter description configured in the plugin, the large model completes parameter extraction through group prompts. For example, in the question "Please help me analyze the problem section of Changning Road 21789-S1 in Shanghai during the second week of August?", the extracted "city" field is "Shanghai", which is converted to "Shanghai" and assembled into a request JSON format that can call the service.

[0101] This JSON file is used to call a data API interface to obtain the corresponding results, such as querying the database for an issue at a specific time and location. Further analysis is then performed. The parameters and format required by the interface agreed upon between the large model and the database (id=sxc-04) are as follows:

[0102] The agreed-upon JSON data format for large models and databases:

[0103] [{city_name=Shanghai}, {road_name=Changning Road}, {sdate=2025-08-Week 2}, {abnormal_section_id=Problem section ID}]

[0104] After the back-end obtains the required parameters provided by the large model through interface id=sxc-04, that is the required parameters for the standard interface sxc-00: city_name, roaod_name, sdate, abnormal_section_id

[0105] After preprocessing assembles the SQL statement corresponding to id=sxc-04, pushes it to the database for execution, obtains the result, and constructs it into the format agreed with the front-end. The SQL statement is as follows:

[0106] select key_indicators from "table_name" where city_name=Shanghai and road_name=Changning Road and sdate=Second week of August 2025 and abnormal_section_id=21789-S1

[0107] Intent execution action after parameter extraction. After parameter extraction is completed, it is in the agreed format that is easy for the large model to understand, as shown below (standard JSON format):

[0108] {city_name=Shanghai}, {road_name=Changning Road}, {sdate=Second week of August 2025}, {abnormal_section_id=21789-S1}, {rsrp=-108dbm, SINR=1.5}

[0109] The content presented by the provided planning agent is as follows:

[0110] 1) City name: Shanghai

[0111] 2) Road name: Changning Road

[0112] 3) Problem road section id: 21789-S1

[0113] 4) Time: Second week of August 2025

[0114] 5) Main indicators of the problematic road section: Average level: -108dBm, downlink SINR=1.5

[0115] After the query statement is executed successfully, the query "Please help me analyze the problematic road section 21789-S1 on Changning Road, Shanghai in the second week of August?" is expanded as follows:

[0116] Please first help me query the indicators of the problematic road section with id=21789-S1 on Changning Road, Shanghai in the second week of August 2025, then determine whether the result meets the requirements for a poor-quality road section. If the conditions are met, please continue to analyze the root cause of the problem and the solution. No analysis is required if the conditions are not met.

[0117] Once completed, the preprocessing unit sends feedback to the central control intelligent agent. After completion, proceed to step 104.

[0118] Step 104: Send the content after recognizing and expanding the user query intent to the planning agent.

[0119] After receiving and expanding the user query intent from the preprocessing stage, the system concatenates and organizes prompt words.

[0120] ## Role Definition

[0121] <role_definition>

[0122] You are a professional wireless network automated drive test consultant with the following expertise:

[0123] - In-depth analysis of road network problems

[0124] - Design an approach to analyze the root causes of problem sections in automated road testing.

[0125] - Invoking different sub-agents and specialized tools for system diagnosis

[0126] - Provide coherent suggestions based on historical interactions

[0127] - Refer to highly relevant expert knowledge

[0128] < / role_definition>

[0129] ## Execution Logic

[0130] <execution_logic>

[0131] 1. Problem Understanding Phase

[0132] - Analyze the user's true intent in the current query, which can be expanded as follows:

[0133] 1) Please help me find the index for the problematic section of Changning Road, Shanghai, with ID=21789-S1, in the second week of August 2025.

[0134] 2) Determine whether the road section meets the criteria for poor road quality based on the query results.

[0135] 3) If the conditions are met, please continue to analyze the root causes and solutions to the problem.

[0136] 4) If the conditions are not met, no analysis is required; simply provide the indicator results and end the analysis.

[0137] 2. Sub-agent selection phase

[0138] - Select one or more sub-agents based on problem complexity.

[0139] - Prioritize diagnostic sub-agents to identify the root cause of the problem.

[0140] - Use a toolchain invocation strategy for complex problems.

[0141] 3. Solution Generation Phase

[0142] - Provide actionable step-by-step implementation suggestions based on the output schemes of multiple sub-agents.

[0143] - Clearly define expected results and risk warnings

[0144] 4. Follow-up Action Plan

[0145] - Define clear follow-up steps

[0146] - Provide methods for verifying the effectiveness of the solution.

[0147] - Establish a feedback mechanism for continuous optimization.

[0148] < / execution_logic>

[0149] ## Output Format

[0150] <output_format>

[0151] - Problem Analysis Summary

[0152] [The process of understanding and analyzing the problem]

[0153] {JSON format for intelligent agent or tool calls}

[0154] < / output_format>

[0155] ## Specific Requirements

[0156] <requirements>

[0157] - Response quality requirements: All recommendations must be based on expert knowledge or tool analysis results, and technical solutions must be specific to the feasibility level.

[0158] - Requirements for using intelligent agents or tools: Each invocation must clearly state the reason for the invocation, including the functionality of the intelligent agent and that the parameters for invoking the tool must be reasonable and complete.

[0159] - User experience requirements: Technical explanations should balance professionalism and comprehensibility, maintaining consistency with historical context.

[0160] < / requirements>

[0161] ## Precautions

[0162] <attention_points>

[0163] - Technical Risk Warning: Architectural changes may introduce new complexities, and performance optimizations may affect system stability.

[0164] - Implementation constraints: Consider the team's technical capabilities and resource limitations, and balance short-term results with long-term maintainability.

[0165] - Communication Considerations: Avoid piling on overly technical jargon; clearly state the pros and cons for key decision points.

[0166] < / attention_points>

[0167] After sending to the planning agent, proceed to step 105.

[0168] In one implementation, the planning agent obtains information about the business sub-agents used for analyzing road network quality from the business agent according to a planning request, and plans the invocation decisions for each business sub-agent based on the business sub-agent information, specifically including:

[0169] The planning agent obtains the available business sub-agents that meet the planning request and adapts to the business agent. It queries the RAG knowledge base for the functional descriptions of the available business sub-agents that adapt to the business agent and requests the LLM to perform task planning based on the planning request and functional descriptions.

[0170] Based on the planning request and functional description, LLM selects the appropriate business sub-intelligent agents from the available business sub-intelligent agents that meet the planning request, arranges the execution order of each business sub-intelligent agent and formulates the calling requirements for each business sub-intelligent agent, so as to obtain the calling decision for each business sub-intelligent agent and return it to the planning agent.

[0171] In this embodiment, as Figure 4 As shown:

[0172] Step 105: Plan the intelligent agent to obtain business sub-intelligent agent information from the knowledge base.

[0173] Once the planning agent receives the prompt from the central planning agent, it understands the expanded content of the user's query. Based on the prompt, it needs to further understand the functions and roles of the business sub-agents to prepare for the next step of the LLM to call which business sub-agents and in what order. Therefore, it needs to retrieve information from the knowledge base about which business sub-agents are available and what their functions are to help the user solve the question. After this, it proceeds to step 106.

[0174] Step 106: The knowledge base returns the available business sub-agent information content.

[0175] After receiving a request from the planning agent to obtain information about business sub-agents, the knowledge base uses RAG technology to enhance the retrieval of user query extended content in the prompt words. The purpose is to build a dynamic cognitive environment for the LLM, retrieve relevant information about business sub-agents from the knowledge base, and provide it as context to the LLM, so that the LLM can more accurately select business sub-agents and arrange their order.

[0176] In one implementation, the planning agent obtains the available business sub-agents that meet the planning request and then queries the RAG knowledge base for the functional descriptions of the available business sub-agents that meet the planning request. Specifically, this includes:

[0177] The planning agent obtains the currently available service sub-agents, including the adapted service agent of all service sub-agents used to analyze road network quality. All service sub-agents used to analyze road network quality include: road problem type analysis sub-agent, road problem master cell analysis sub-agent, alarm analysis sub-agent, neighbor cell analysis sub-agent, and antenna and feeder optimization sub-agent.

[0178] The functional description of the sub-agent for querying road problem types from the RAG knowledge base includes: inputting the road segment ID and user query time to query key performance indicator data of the road network, analyzing the types of road network quality problems based on the key performance indicator data of the road network, and generating road network quality problem analysis results in a preset format.

[0179] The functional description of the sub-agent for querying the RAG knowledge base for the main control cell analysis of road problems includes: obtaining all coverage cells and sampling point data of the road based on the road segment ID, and identifying the TOPN main control cells of the road from all coverage cells based on the sampling point data;

[0180] The functional description of querying the RAG knowledge base for the alarm analysis sub-agent includes: obtaining alarm information of the road TOPN master cell at the time of user query, analyzing the impact of each alarm message, and identifying alarms that cause road network quality problems;

[0181] The functional description of the neighbor cell analysis sub-agent for querying the RAG knowledge base includes: obtaining neighbor cells that correspond to the ECI of the main control cell of the road TOPN, establishing cell pairs that match the main control cell and neighbor cells, and identifying abnormal cell pairs with missing neighbor cell matching.

[0182] The function description of the antenna and feeder optimization sub-agent retrieved from the RAG knowledge base includes: generating an antenna and feeder optimization and adjustment scheme for the road network based on the type of road network quality problem, the alarms that caused the road network quality problem, and the abnormal cells with missing neighboring cells.

[0183] In this embodiment, the business sub-agents in the knowledge base of step 106 mainly include the following five, and their main functions are as follows:

[0184] 1. Intelligent Agent for Analyzing Road Problem Types

[0185] Based on the road segment ID number of the road problem, query the network key indicators for the corresponding time period, and determine the type of problem based on the results, which are mainly coverage problems and poor quality problems. Then send the results to the planning agent.

[0186] The JSON format of the intelligent agent framework design is as follows:

[0187] {

[0188] "agent_name": "Road Problem Type Analysis Agent",

[0189] "description": "Based on the road segment ID number with the problem, query the network key indicators for the corresponding time period, determine the problem type (coverage problem or poor quality problem), and send the results to the planning agent."

[0190] "processing_steps": [

[0191] {

[0192] "step": 1,

[0193] "action": "Input reception",

[0194] "description": "Receives request data containing road segment ID and time period parameters",

[0195] "input": {

[0196] "road_segment_id": "String type, uniquely identifies a road segment",

[0197] "time_period": "Time range, in the format YYYY-MM-DD HH:MM:SS to YYYY-MM-DD HH:MM:SS"

[0198] }

[0199] },

[0200] {

[0201] "step":2,

[0202] "action": Invokes the "Indicator Query" tool.

[0203] "description": "Query network key performance indicator data based on road segment ID and time period",

[0204] "indicator query": [

[0205] "rsrp: Network coverage signal strength", "sinr: Downlink SINR quality",

[0206] ],

[0207] },

[0208] {

[0209] "step":3,

[0210] "action": Invokes the "Problem Type Analysis" tool.

[0211] "description": "Identifying the type of problem based on key network indicators found in the query".

[0212] "judgment logic": {

[0213] "characteristics of coverage problem": [

[0214] "Network signal strength threshold is below the threshold"

[0215] "characteristics of quality problem": [

[0216] "Downlink SINR quality threshold is below the threshold"

[0217] ],

[0218] "Judgment rule": "When the coverage metric is abnormal but the quality metric is normal, it is judged as a coverage problem; when the coverage metric is normal but the quality metric is abnormal, it is judged as a quality problem; when both the coverage metric and the quality metric are abnormal, it is judged as a coverage / quality problem."

[0219] }

[0220] },

[0221] {

[0222] "step": 4,

[0223] "action": "Result generation",

[0224] "description": "Generate an analysis report containing the problem type determination results",

[0225] "output": {

[0226] "problem_type": "Coverage problem, poor quality problem, or coverage quality problem",

[0227] "confidence_level": "Indicates confidence level (0-100%)",

[0228] "key_metrics": "Key metric values ​​and explanations of abnormal situations",

[0229] }

[0230] },

[0231] {

[0232] "step": 5,

[0233] "action": "Result sent",

[0234] "description": "Send the analysis results to the planning agent for further processing",

[0235] "transmission protocol": "RESTful API or message queue",

[0236] "data format ": "JSON",

[0237] "target endpoint": "The receiving interface of the planning agent"

[0238] }

[0239] 2. Intelligent agent for analyzing road-related issues in the main control area

[0240] Based on the ID number of the problematic road segment, the main control cell covering the problematic road segment is identified, and sorted according to rules. The top three cells are selected as the Top 3 main control cells based on the proportion of sampling points, and the results are sent to the planning agent.

[0241] {

[0242] "agent_name": "Road Problem Control Cell Analysis Agent",

[0243] "description": "Based on the road segment ID with the problem, identify the main controlling cells and sort them according to rules, then filter the top 3 main controlling cells".

[0244] "input": {

[0245] "road_segment_id": "string (ID of the problematic road segment)"

[0246] },

[0247] "processing_steps": [

[0248] {

[0249] "step": 1,

[0250] "action": "Data Acquisition",

[0251] "description": "Based on the road segment ID with the problem, retrieve all coverage cells and corresponding sampling point data for that road segment from the network measurement database."

[0252] },

[0253] {

[0254] "step": 2,

[0255] "action": "Master control cell identification",

[0256] "description": "The number of sampling points in each community is counted, and the coverage percentage of each community in that road segment is calculated."

[0257] },

[0258] {

[0259] "step": 3,

[0260] "action": "Application of sorting rules",

[0261] "description": "Sorted by the percentage of sampling points from highest to lowest; if the percentages are the same, sorted in descending order by the average signal strength."

[0262] },

[0263] {

[0264] "step": 4,

[0265] "action": "Top3 filter",

[0266] "description": "Select the top three communities as the main control communities, requiring that the sampling points of each community account for no less than a threshold (e.g., 5%)."

[0267] },

[0268] {

[0269] "step": 5,

[0270] "action": "Result sent",

[0271] "description": Sends the analysis results to the planning agent for further processing.

[0272] "transmission protocol": "RESTful API or message queue",

[0273] "data format ": "JSON",

[0274] "target endpoint": "the receiving interface of the planning agent"}

[0275] 3. Alarm Analysis Intelligent Agent

[0276] Based on the top 3 controlling cells identified by the road problem control agent, the alarm information is queried according to the query time range involved in the user's problem. It is determined whether there are alarms affecting the road problem in the top 3 controlling cells, and the results are sent to the planning agent.

[0277] {

[0278] "agent_name": "Alarm Analysis Agent",

[0279] "description": "Based on the top 3 controlling cells provided by the road problem controlling cell analysis agent and the user query time range, query alarm information, determine whether there are alarms affecting road problems, and send the results to the planning agent."

[0280] "inputs": {

[0281] "top3_cells": {

[0282] "source": "Road Problem Master Control Cell Analysis Agent",

[0283] "description": "List of TOP3 controlling cells, each cell includes cell ID and other relevant information",

[0284] "example": ["cell_id_1", "cell_id_2", "cell_id_3"]

[0285] },

[0286] "time_range": {

[0287] "source": "user query",

[0288] "description": "Query time range, including start and end times",

[0289] "example": {"start_time": "2023-10-01 00:00:00", "end_time": "2023-10-01 23:59:59"}

[0290] }

[0291] },

[0292] "processing_steps": [

[0293] {

[0294] "step": 1,

[0295] "action": "Receive input",

[0296] "description": "From the analysis agent of the main control cell of road problems, obtain the list of the top 3 main control cells, and obtain the time range from user queries."

[0297] },

[0298] {

[0299] "step": 2,

[0300] "action": "Query alarm information",

[0301] "description": "For each of the top 3 controlling cells, use the cell ID and time range to query relevant alarm records from the alarm database or API. Query criteria include cell identifier and time filtering."

[0302] },

[0303] {

[0304] "step": 3,

[0305] "action": "Analyze the impact of alarms",

[0306] "description": "Filter and analyze the retrieved alarm information to determine whether it affects road conditions. The impact criteria are based on predefined rules, such as the alarm type (e.g., signal interruption, coverage problem, hardware failure) or severity (e.g., severe, emergency) being related to the road problem."

[0307] },

[0308] {

[0309] "step": 4,

[0310] "action": "Summarize results",

[0311] "description": "Generates analysis results for each of the top 3 controlling cells, including whether there are alarms affecting road issues, and alarm details (such as alarm type, occurrence time, and description)."

[0312] },

[0313] {

[0314] "step": 5,

[0315] "action": "Send result",

[0316] "description": Sends the analysis results to the planning agent for further processing.

[0317] "transmission protocol": "RESTful API or message queue",

[0318] "data format ": "JSON",

[0319] "target endpoint": "the receiving interface of the planning agent"}

[0320] 4. Neighborhood Analysis Agent

[0321] The system matches the data of the main control cell and the neighbor cell analysis results for the problematic road segment, identifies cell pairs with neighbor cell issues, generates a correspondence between the serving cell ECI and the neighbor cell ECI in the neighbor cell pair relationship table, and associates the "main cell ECI" and "neighbor cell ECI" fields in the neighbor cell analysis result table. When the "judgment result" is "neighbor cell mismatch", an abnormal result is output and sent to the planning agent.

[0322] {

[0323] "agent_name": "Neighborhood Analysis Agent",

[0324] "description": "This agent is responsible for analyzing the neighbor cell pair data and neighbor cell analysis results of the main control cell in the problem road segment, matching the cell pairs with neighbor cell problems, and outputting the abnormal results to the planning agent."

[0325] "inputs": {

[0326] "problem_road_sector_data": {

[0327] "description": "Data table of neighboring cells in the main control cell of the problem road section, including the ECI correspondence between the serving cell and the neighboring cells."

[0328] "required_fields": ["service_cell_eci", "neighbor_cell_eci"]

[0329] },

[0330] "neighbor_analysis_results": {

[0331] "description": "Neighbor cell analysis results table, including the primary cell's ECI, neighboring cell's ECI, and judgment results."

[0332] "required_fields": ["main_cell_eci", "neighbor_cell_eci", "judgment_result"]

[0333] }

[0334] },

[0335] "process": [

[0336] {

[0337] "step": 1,

[0338] "action": "data association",

[0339] "details": "Based on the mapping between the serving cell's ECI and the neighboring cell's ECI, the service_cell_eci and neighbor_cell_eci in problem_road_sector_data are matched with the main_cell_eci and neighbor_cell_eci in neighbor_analysis_results. The matching method is an inner join, ensuring that the ECI pairs in the two tables are completely consistent."

[0340] },

[0341] {

[0342] "step": 2,

[0343] "action": "Condition check",

[0344] "details": "For each successfully matched cell pair, check the `judgment_result` field in `neighbor_analysis_results`. If the value of `judgment_result` is 'missing neighbor cell match', then mark it as an abnormal cell pair."

[0345] },

[0346] {

[0347] "step": 3,

[0348] "action": "Summarize results",

[0349] "details": "Collect all cell pairs marked as abnormal, including the serving cell ECI, neighboring cell ECI, and judgment results, to form an abnormal result list."

[0350] }

[0351] ,

[0352] {

[0353] "step": 4,

[0354] "action": "Send result",

[0355] "description": Sends the analysis results to the planning agent for further processing.

[0356] "transmission protocol": "RESTful API or message queue",

[0357] "data format ": "JSON",

[0358] "target endpoint": "the receiving interface of the planning agent"}

[0359] 5. Antenna Feed Optimization Intelligent Agent

[0360] After confirming that the alarm analysis agent and neighbor cell analysis agent outputs no issues, the system then judges the road problem type based on the output of the road problem type analysis agent. If the road problem type is a coverage issue, the system uses the output of the master cell analysis agent to analyze the azimuth, electronic downtilt, mechanical downtilt, and antenna station height of the first three master cells, providing specific analysis results and antenna optimization adjustment schemes. If the road problem type is a quality issue, the system uses the output of the master cell analysis agent to analyze the azimuth, electronic downtilt, mechanical downtilt, and antenna station height of the first three master cells, providing specific analysis results and antenna optimization adjustment schemes. The results are then sent to the planning agent.

[0361] {

[0362] "agent_name": "Tianwei Tuning Intelligent Agent",

[0363] "description": "After confirming that the alarm analysis and neighbor cell analysis results are correct, the operating parameters of the main control cell are analyzed according to the road problem type (coverage problem / quality problem), an antenna and feeder optimization and adjustment scheme is generated, and the results are sent to the planning agent."

[0364] "modules": [

[0365] {

[0366] "module_name": "Input validation module",

[0367] "description": "Verify that the outputs from the alarm analysis agent, neighbor cell analysis agent, and road problem type analysis agent are complete and error-free."

[0368] "inputs": [

[0369] "Alarm Analysis Results"

[0370] "Neighborhood Analysis Results",

[0371] "Road Problem Types (Coverage / Quality)"

[0372] ],

[0373] "outputs": [

[0374] "Verification passed"

[0375] Error log (if any) ]

[0377] },

[0378] {

[0379] "module_name": "Master Cell Acquisition Module",

[0380] "description": "Invokes the main control cell analysis agent to obtain the operating parameters of the top three main control cells related to road problems."

[0381] "inputs": [

[0382] "Road Problem Signs"

[0383] "Verification passed" indicator

[0384] ],

[0385] "outputs": [

[0386] "List of Controlled Cells (Maximum 3)"

[0387] "Engine parameters for each cell: azimuth, electronic downtilt angle, mechanical downtilt angle, and antenna station height." ]

[0389] },

[0390] {

[0391] "module_name": "Coverage Problem Analysis Module",

[0392] "description": "When the road problem type is a coverage problem, perform coverage analysis on the acquired engineering parameters and generate an optimization scheme."

[0393] "condition": "Road problem type == "coverage"",

[0394] "inputs": [

[0395] "Master Control Community Engineering Parameters"

[0396] ],

[0397] "process": [

[0398] ① Calculate the antenna coverage radius for each cell (based on azimuth, downtilt, and station height)

[0399] "② Compare with actual road requirements (road width, road direction) to determine coverage gaps,"

[0400] "③ If a coverage gap exists, parameter adjustment suggestions are generated: adjust the azimuth angle, electronic downtilt angle, mechanical downtilt angle, or increase the antenna feeder station height."

[0401] ④ Assess the coverage improvement after parameter adjustment and prepare a feasibility report.

[0402] ],

[0403] "outputs": [

[0404] "Coverage Issue Analysis Report"

[0405] Antenna and Feeder Optimization and Adjustment Plan (Addressing Coverage Gaps) ]

[0407] },

[0408] {

[0409] "module_name": "Quality Problem Analysis Module",

[0410] "description": "When the road problem type is a quality problem, perform quality analysis on the acquired engineering parameters and generate an optimization solution."

[0411] "condition": "Road Problem Type == "quality"",

[0412] "inputs": [

[0413] "Master Control Community Engineering Parameters"

[0414] ],

[0415] "process": [

[0416] ① Calculate the signal quality metrics (such as RSRP and SINR) for each cell based on the current operating parameters.

[0417] "② Compare with the actual mass threshold measured on the road to pinpoint the direction of insufficient mass."

[0418] "③ Based on simulation or empirical models, parameter adjustment suggestions are proposed: fine-tune the azimuth angle, electronic downtilt angle, mechanical downtilt angle, or appropriately change the antenna feeder station height."

[0419] ④ Evaluate the quality improvement after parameter adjustment and prepare a feasibility report.

[0420] ],

[0421] "outputs": [

[0422] "Quality Problem Analysis Report"

[0423] Antenna and Feeder Optimization and Adjustment Plan (Addressing Quality Insufficiency) ]

[0425] },

[0426] {

[0427] "module_name": "Result Integration and Sending Module",

[0428] "description": "After formatting the analysis report and optimization plan in a unified way, send them to the planning agent."

[0429] "inputs": [

[0430] "Coverage Issue Analysis Report / Quality Issue Analysis Report",

[0431] "Corresponding antenna and feeder optimization adjustment scheme"

[0432] ],

[0433] "process": [

[0434] ① Merge reports and plans to generate a unified JSON structure.

[0435] "② Call the planning agent's interface (such as a REST API) to send the results",

[0436] ③ Record sending logs and return status.

[0437] ],

[0438] "outputs": [

[0439] "Successful delivery indicator",

[0440] Send Log ]

[0442] }

[0443] ],

[0444] "workflow": [

[0445] {

[0446] "step_id": 1,

[0447] "name": "Input Validation",

[0448] "module": "Input validation module",

[0449] "next_step": 2

[0450] },

[0451] {

[0452] "step_id": 2,

[0453] "name": "Getting the main control cell's operating parameters",

[0454] "module": "Master Cell Acquisition Module",

[0455] "next_step": 3

[0456] },

[0457] {

[0458] "step_id": 3,

[0459] "name": "Problem Type Branch",

[0460] "condition": "Road Problem Type",

[0461] "branches": {

[0462] "coverage": 4,

[0463] "quality": 5

[0464] }

[0465] },

[0466] {

[0467] "step_id": 4,

[0468] "name": "Coverage Problem Analysis",

[0469] "module": "Coverage Problem Analysis Module",

[0470] "next_step": 6

[0471] },

[0472] {

[0473] "step_id": 5,

[0474] "name": "Quality Problem Analysis",

[0475] "module": "Quality Problem Analysis Module",

[0476] "next_step": 6

[0477] },

[0478] {

[0479] "step_id": 6,

[0480] "name": "Results integrated and sent",

[0481] "module": "Result Integration and Sending Module",

[0482] "next_step": null

[0483] } ]

[0485] After completion, proceed to step 107.

[0486] In one implementation, the invocation requirements for each adapted business sub-agent are defined, specifically including:

[0487] Based on the road type being one of National Highway / Provincial Highway / County Road / Township Road / Expressway / High-speed Railway / Elevated Road, the calling requirements for the road problem type analysis sub-agent include: querying key performance indicator data of road networks of different road segment lengths, and locating road segments with road network quality problems;

[0488] The calling requirements for the main control cell analysis sub-agent for road problems include: obtaining the percentage of sampling points for road segments with road network quality problems, and obtaining the top 8 and top 3 main control cells with the largest percentage of sampling points;

[0489] The requirements for calling the alarm analysis sub-agent include: obtaining alarm information from the top 3 master cells within a preset time period, and diagnosing whether the alarm information affects the services of the master cells;

[0490] The calling requirements for the neighbor cell analysis sub-agent include: obtaining whether any of the top 8 main control cells have been missing from the neighbor cell configuration of the top 3 main control cells;

[0491] The requirements for calling the antenna feeder optimization sub-agent include: formulating an antenna feeder optimization and adjustment scheme based on the sampling contribution of associated cells of road segments with road network quality problems, the distance of the problematic road segments, and the voltage level.

[0492] In this embodiment, as Figure 4 As shown:

[0493] Step 107: Based on the functional description of the business sub-agent returned by the knowledge base, request detailed task planning from the LLM.

[0494] The business sub-agent function description obtained from the knowledge base in step 106 is updated in conjunction with the prompt words written by the central control agent in step 104, and then a request is sent to the LLM. The specific content is as follows:

[0495] ## Role Definition

[0496] <role_definition>

[0497] You are a professional wireless network automated drive test consultant with the following expertise:

[0498] - In-depth analysis of road network problems

[0499] - Design an approach to analyze the root causes of problem sections in automated road testing.

[0500] - Invoking different sub-agents and specialized tools for system diagnosis

[0501] - Provide coherent suggestions based on historical interactions

[0502] - Refer to highly relevant expert knowledge

[0503] < / role_definition>

[0504] ## Execution Logic

[0505] <execution_logic>

[0506] 1. Problem Understanding Phase

[0507] - Analyze the user's true intent in the current query, which can be expanded as follows:

[0508] 1) Please help me find the index for the problematic section of Changning Road, Shanghai, with ID=21789-S1, in the second week of August 2025.

[0509] 2) Determine whether the road section meets the criteria for poor road quality based on the query results.

[0510] 3) If the conditions are met, please continue to analyze the root causes and solutions to the problem.

[0511] 4) If the conditions are not met, no analysis is required; simply provide the indicator results and end the analysis.

[0512] 2. Sub-agent selection phase

[0513] - Select one or more sub-agents based on problem complexity.

[0514] The specific functions of the business sub-agent are as follows:

[0515] 1) Intelligent agent for analyzing road problem types

[0516] Based on the road segment ID number of the road problem, query the network key indicators for the corresponding time period, and determine the type of problem based on the results, which are mainly coverage problems and poor quality problems. Then send the results to the planning agent.

[0517] JSON format: ... (Please see step 106 for details)

[0518] 2) Intelligent agent for analyzing road problems in the main control cell

[0519] Based on the ID number of the problematic road segment, the main control cell covering the problematic road segment is identified, and sorted according to rules. The top three cells are selected as the Top 3 main control cells based on the proportion of sampling points, and the results are sent to the planning agent.

[0520] JSON format: ... (Please see step 106 for details)

[0521] 3) Alarm Analysis Agent

[0522] Based on the top 3 controlling cells identified by the road problem control agent, the alarm information is queried according to the query time range involved in the user's problem. It is determined whether there are alarms affecting the road problem in the top 3 controlling cells, and the results are sent to the planning agent.

[0523] JSON format: ... (Please see step 106 for details)

[0524] 4) Neighborhood Analysis Agent

[0525] The system matches the data of the main control cell and the neighbor cell analysis results for the problematic road segment, identifies cell pairs with neighbor cell issues, generates a correspondence between the serving cell ECI and the neighbor cell ECI in the neighbor cell pair relationship table, and associates the "main cell ECI" and "neighbor cell ECI" fields in the neighbor cell analysis result table. When the "judgment result" is "neighbor cell mismatch", an abnormal result is output and sent to the planning agent.

[0526] JSON format: ... (Please see step 106 for details)

[0527] 5) Antenna feed optimization intelligent agent

[0528] After confirming that the alarm analysis agent and neighbor cell analysis agent outputs no issues, the system then judges the road problem type based on the output of the road problem type analysis agent. If the road problem type is a coverage issue, the system uses the output of the master cell analysis agent to analyze the azimuth, electronic downtilt, mechanical downtilt, and antenna station height of the first three master cells, providing specific analysis results and antenna optimization adjustment schemes. If the road problem type is a quality issue, the system uses the output of the master cell analysis agent to analyze the azimuth, electronic downtilt, mechanical downtilt, and antenna station height of the first three master cells, providing specific analysis results and antenna optimization adjustment schemes. The results are then sent to the planning agent.

[0529] JSON format: ... (Please see step 106 for details)

[0530] - Prioritize diagnostic sub-agents to identify the root cause of the problem.

[0531] Based on the description of the user's question and the specific functional description of the business sub-agent, accurately select the business sub-agent and arrange its order, with the aim of answering the user's question more precisely.

[0532] - Use a toolchain invocation strategy for complex problems.

[0533] 3. Solution Generation Phase

[0534] - Provide actionable step-by-step implementation suggestions based on the output schemes of multiple sub-agents.

[0535] - Clearly define expected results and risk warnings

[0536] 4. Follow-up Action Plan

[0537] - Define clear follow-up steps

[0538] - Provide methods for verifying the effectiveness of the solution.

[0539] - Establish a feedback mechanism for continuous optimization.

[0540] < / execution_logic>

[0541] ## Output Format

[0542] <output_format>

[0543] - Problem Analysis Summary

[0544] [The process of understanding and analyzing the problem]

[0545] {JSON format for intelligent agent or tool calls}

[0546] < / output_format>

[0547] ## Specific Requirements

[0548] <requirements>

[0549] - Response quality requirements: All recommendations must be based on expert knowledge or tool analysis results, and technical solutions must be specific to the feasibility level.

[0550] - Requirements for using intelligent agents or tools: Each invocation must clearly state the reason for the invocation, including the functionality of the intelligent agent and that the parameters for invoking the tool must be reasonable and complete.

[0551] - User experience requirements: Technical explanations should balance professionalism and comprehensibility, maintaining consistency with historical context.

[0552] < / requirements>

[0553] ## Precautions

[0554] <attention_points>

[0555] - Technical Risk Warning: Architectural changes may introduce new complexities, and performance optimizations may affect system stability.

[0556] - Implementation constraints: Consider the team's technical capabilities and resource limitations, and balance short-term results with long-term maintainability.

[0557] - Communication Considerations: Avoid piling on overly technical jargon; clearly state the pros and cons for key decision points.

[0558] < / attention_points>

[0559] After completion, proceed to step 108.

[0560] Step 108: LLM returns the execution plan content.

[0561] Based on the updated prompts, combined with the expanded content of the user's question and the specific functional description of the business sub-agents, LLM provides a specific execution plan, including which business sub-agents to select and their execution order.

[0562] Based on the prompts in step 107, the LLM, after understanding the expanded content of the user's question and the functions of different business sub-agents, gives different decision results, as follows:

[0563] Decision 1:

[0564] 1) Step 1: First, call the "Road Problem Type Analysis Agent" to obtain the key indicators of road problems, then complete the road problem type judgment, and send the results to the planning agent.

[0565] 2) Step 2: After determining the type of road problem through Step 1, call the "Road Problem Master Control Cell Analysis Agent" to obtain the TOP 3 master control cell information of the road problem, and send the results to the planning agent.

[0566] 3) Step 3: After obtaining the information of the top 3 main control cells of road problems through Step 2, call the "alarm analysis agent" to obtain the alarm information of the top 3 main control cells and determine whether there are alarms affecting road problems. If there are, stop the subsequent analysis and send the results to the planning agent.

[0567] 4) Step 4: The planning agent forwards the call results of all business sub-agents to the central control agent for status confirmation, to see if there is any call result feedback.

[0568] 5) Step 5: After checking the feedback results of all business sub-intelligent agents, the central control intelligent agent assembles the content and sends it to the summarizing intelligent agent for summarization.

[0569] 6) Step 6: The agent expands the content assembled in Step 5, extends the root cause analysis and solution request into the prompt words, and sends it to the LLM to provide specific root cause analysis and solutions, then ends.

[0570] Decision Two:

[0571] 1) Step 1: First, call the "Road Problem Type Analysis Agent" to obtain the key indicators of road problems, then complete the road problem type judgment, and send the results to the planning agent.

[0572] 2) Step 2: After determining the type of road problem through Step 1, call the "Road Problem Master Control Cell Analysis Agent" to obtain the TOP 3 master control cell information of the road problem, and send the results to the planning agent.

[0573] 3) Step 3: After obtaining the information of the top 3 main control cells of road problems through Step 2, call the "alarm analysis agent" to obtain the alarm information of the top 3 main control cells and determine whether there are alarms affecting road problems. If there are no problems, send the results to the planning agent.

[0574] 4) Step 4: Call the "Neighbor Analysis Agent" to obtain whether the main control cell of the road problem is missing key neighbor cells. If it exists, stop the subsequent analysis and send the results to the planning agent.

[0575] 5) Step 5: The planning agent forwards the call results of all business sub-agents to the central control agent for status confirmation, to see if there is any call result feedback.

[0576] 6) Step 6: After checking the feedback results of all business sub-intelligent agents, the central control intelligent agent assembles the content and sends it to the summarizing intelligent agent for summarization.

[0577] 7) Step 7: The agent expands the content assembled in Step 5, extends the root cause analysis and solution request into the prompt words, and sends it to the LLM to provide specific root cause analysis and solutions, then ends.

[0578] Decision Three:

[0579] 1) Step 1: First, call the "Road Problem Type Analysis Agent" to obtain the key indicators of road problems, then complete the road problem type judgment, and send the results to the planning agent.

[0580] 2) Step 2: After determining the type of road problem through Step 1, call the "Road Problem Master Control Cell Analysis Agent" to obtain the TOP 3 master control cell information of the road problem, and send the results to the planning agent.

[0581] 3) Step 3: After obtaining the information of the top 3 main control cells of road problems through Step 2, call the "alarm analysis agent" to obtain the alarm information of the top 3 main control cells and determine whether there are alarms affecting road problems. If there are no problems, send the results to the planning agent.

[0582] 4) Step 4: Call the "Neighbor Analysis Agent" to obtain whether the main control cell of the road problem is missing key neighbor cells. If it exists, stop the subsequent analysis and send the results to the planning agent.

[0583] 5) Step 5: The planning agent forwards the call results of all business sub-agents to the central control agent for status confirmation, to see if there is any call result feedback.

[0584] 6) Step 6: After checking the feedback results of all business sub-intelligent agents, the central control intelligent agent assembles the content and sends it to the summarizing intelligent agent for summarization.

[0585] 7) Step 7: The agent expands the content assembled in Step 5, extends the root cause analysis and solution request into the prompt words, and sends it to the LLM to provide specific root cause analysis and solutions, then ends.

[0586] Decision Four:

[0587] 1) Step 1: First, call the "Road Problem Type Analysis Agent" to obtain the key indicators of road problems, then complete the road problem type judgment, and send the results to the planning agent.

[0588] 2) Step 2: After determining the type of road problem through Step 1, call the "Road Problem Master Control Cell Analysis Agent" to obtain the TOP 3 master control cell information of the road problem, and send the results to the planning agent.

[0589] 3) Step 3: After obtaining the information of the top 3 main control cells of road problems through Step 2, call the "alarm analysis agent" to obtain the alarm information of the top 3 main control cells and determine whether there are alarms affecting road problems. If there are no problems, send the results to the planning agent.

[0590] 4) Step 4: Call the "Neighbor Analysis Agent" to obtain whether the main control cell of the road problem is missing key neighbor cells. If no key neighbor cells are missing, send the result to the planning agent.

[0591] 5) Step 5: Call the "Antenna and Feeder Optimization Intelligent Agent" to perform coverage or quality problem analysis on the azimuth angle, electronic downtilt angle, mechanical downtilt angle and antenna and feeder station height in the top 3 main control cell of road problems, provide specific analysis results and antenna and feeder optimization adjustment schemes, and send the results to the planning intelligent agent.

[0592] 6) Step 6: The planning agent forwards the call results of all business sub-agents to the central control agent for status confirmation, to see if there is any call result feedback.

[0593] 7) Step 7: After the central control agent checks the feedback results of all business sub-agents, it assembles the content and sends it to the summarizing agent for summarization.

[0594] 8) Step 8: The agent expands the content assembled in Step 5, extends the root cause analysis and solution request into the prompt words, and sends it to the LLM to provide specific root cause analysis and solutions, then ends.

[0595] After completion, proceed to step 109.

[0596] In one implementation, the business intelligence agent interacts with the planning intelligence agent to invoke various business sub-intelligent agents based on invocation decisions, obtain the task execution results of each business sub-intelligent agent analyzing road network quality, and send the task execution results to the central control intelligence agent. Specifically, this includes:

[0597] The planning intelligent agent controls the business intelligent agent to call each adapted business sub-intelligent agent according to the calling order. The business intelligent agent obtains the task execution results of each business sub-intelligent agent in analyzing road network quality according to the calling requirements. The planning intelligent agent controls the acquisition of each task execution result and sends it to the central control intelligent agent.

[0598] In one implementation, the planning agent controls the business agent to invoke each adapted business sub-agent according to the invocation order. The business agent obtains the task execution results of each business sub-agent analyzing road network quality according to the invocation requirements. The planning agent controls the acquisition of each task execution result and sends it to the central control agent. Specifically, this includes:

[0599] The planning agent controls the business agent to select the appropriate business sub-agent for the current call based on the calling order;

[0600] The business intelligence agent requests the LLM to generate call parameters including the currently called adaptive business sub-intelligence based on the call requirements. The call parameters are sent to the post-processing module for auditing and calibration. The audited and calibrated call parameters are then used to input the currently called adaptive business sub-intelligence to obtain the task execution result of the currently called adaptive business sub-intelligence.

[0601] The planning agent determines whether to call the next adaptable business sub-agent based on the current task execution result. If yes, the control business agent calls the next adaptable business sub-agent; otherwise, it sends the currently obtained task execution results to the central control agent.

[0602] In this embodiment, as Figure 4 As shown:

[0603] Step 109: Invoke the business intelligence agent according to the LLM return result.

[0604] After receiving multiple decision options from the LLM, the planning agent combines the user query and sends them to the business agent according to the decision steps. Then, proceed to step 110.

[0605] Step 110: The business sub-agent obtains the relevant toolset capability descriptions from the knowledge base.

[0606] The knowledge base primarily collects documents such as road optimization guidelines, analysis processes, coverage optimization strategies, and alarm analysis. These documents are processed to form toolsets, each with detailed functional descriptions and tool call parameter explanations, facilitating tool selection for different business sub-agents within the LLM. The main toolsets are as follows:

[0607] 1. Problem Road Section Type Verification Tool:

[0608] By defining the problem types based on the different road segment lengths, we can determine the types of problems related to coverage, interference, and quality, as shown in Table 2 below.

[0609] Table 2: Examples of Problem Road Section Type Verification Results

[0610]

[0611] The tool usage instructions are as follows:

[0612] "tools": [

[0613] {

[0614] "type": "function",

[0615] "function": {

[0616] "name": "get_issue type_road",

[0617] "description":"Determine the type of road problem",

[0618] "parameters": {

[0619] "properties": {

[0620] "problem definition": {

[0621] "type": "string"

[0622] }

[0623] },

[0624] "required": [

[0625] "problem definition" ]

[0627] type: "object"

[0628] }

[0629] }

[0630] } ]

[0632] }

[0633] 2. Problematic road section TOP control cell identification tool:

[0634] The Top N controlling cells used to identify problematic road sections are shown in Table 3. The identification rules are as follows:

[0635] 1) Aggregate and statistically analyze all sampling points in the problem road section area.

[0636] 2) Statistical analysis was conducted on the sampling points gathered on the problem road section under different master control ECI (cell name). The top 8 master control cells were selected based on the percentage of sampling points, and the top three cells with the highest percentage of sampling points were selected as the Top 3 master control cells.

[0637] Table 3: Examples of Main Control Cell Identification Results

[0638]

[0639] The tool usage instructions are as follows:

[0640] "tools": [

[0641] {

[0642] "type": "function",

[0643] "function": {

[0644] "name": "get_TopN_cell",

[0645] "description": "Identify the main cell",

[0646] "parameters": {

[0647] "properties": {

[0648] "sampling ratio": {

[0649] "type": "string"

[0650] }

[0651] },

[0652] "required": [

[0653] "sampling ratio" ]

[0655] type: "object"

[0656] }

[0657] }

[0658] } ]

[0660] }

[0661] 3. Alarm verification tool for main control area of ​​problematic road section:

[0662] Associate alarm information with the main control cell to diagnose whether there are alarms affecting services in the main control cell. The alarm occurrence time must be alarm data from the most recent week or alarm data that has not been cleared for a long time and affects services. The alarm diagnosis results are shown in Table 4 below.

[0663] Table 4: Examples of Alarm Diagnostic Results

[0664]

[0665] The tool usage instructions are as follows:

[0666] "tools": [

[0667] {

[0668] "type": "function",

[0669] "function": {

[0670] "name": "get_alarms",

[0671] "description": "Check for alarms",

[0672] "parameters": {

[0673] "properties": {

[0674] "alarm name": {

[0675] "type": "string"

[0676] }

[0677] },

[0678] "required": [

[0679] "alarm name" ]

[0681] type: "object"

[0682] }

[0683] }

[0684] } ]

[0686] }

[0687] 4. Tool for checking for missing configurations in the top 3 controlled communities:

[0688] The neighbor cell relationships of the top 3 controlled cells were checked to see if there were any missing neighbor cell configurations. If it was found that the top 3 controlled cells and other controlled cells in the problem section had no neighbor cells added, the corresponding ECI cell information for the unconfigured neighbor cells was output. The results of the missing neighbor cell configuration check are shown in Table 5 below.

[0689] Table 5: Examples of Neighboring Cell Missing Data Verification Results

[0690]

[0691] The tool usage instructions are as follows:

[0692] "tools": [

[0693] {

[0694] "type": "function",

[0695] "function": {

[0696] "name": "get_neicell_miss",

[0697] "description": "Check whether the neighboring areas of the main control community are missing configurations",

[0698] "parameters": {

[0699] "properties": {

[0700] "missing identification": {

[0701] "type": "string"

[0702] }

[0703] },

[0704] "required": [

[0705] "missing identification" ]

[0707] type: "object"

[0708] }

[0709] }

[0710] } ]

[0712] }

[0713] 5. Multi-cell joint antenna and feeder optimization tool:

[0714] By analyzing the sampling contribution of related cells, the distance to problematic road sections, and the signal strength of the affected sections using the multi-cell joint antenna and feeder optimization tool, it was determined that there is room for antenna and feeder optimization. An attempt was made to adjust the antenna and feeder adjustment scheme to enhance the main control coverage. The antenna and feeder optimization scheme is shown in Table 6 below.

[0715] Table 6: Examples of Antenna Feeder Optimization Schemes

[0716]

[0717] The tool usage instructions are as follows:

[0718] "tools": [

[0719] {

[0720] "type": "function",

[0721] "function": {

[0722] "name": "get_antenna_adjust",

[0723] "description":"Multi-cell adjustment and coverageoptimization",

[0724] "parameters": {

[0725] "properties": {

[0726] "adjust azimuth": {

[0727] "type": "string"

[0728] },

[0729] "adjust mechanical angle": {

[0730] "type": "string"

[0731] },

[0732] "adjust electronic angle": {

[0733] "type": "string"

[0734] }

[0735] },

[0736] "required": [

[0737] "adjust azimuth",

[0738] "adjust mechanical angle"

[0739] "adjust electronic angle" ]

[0741] type: "object"

[0742] }

[0743] }

[0744] } ]

[0746] }

[0747] After interacting with the knowledge base through RAG, obtain the function description and tool call instructions for each of the above tools, and then proceed to step 111.

[0748] Step 111: Return information on available toolsets.

[0749] Return all the tool function descriptions and JSON format content obtained to the business sub-agent, and then proceed to step 112.

[0750] Step 112: Request tool selection and parameters from the LLM.

[0751] The retrieved tool function descriptions and JSON format content, along with the user's question and React prompt template, are combined and sent to the LLM. The purpose is to interact with the knowledge base through RAG before the LLM decides which toolset capabilities to invoke. This allows the knowledge base to obtain relevant professional knowledge and toolset functions involved in the reasoning task, confirm the toolset functions that need to be invoked, and complete the problem analysis through a cyclical process of "observation" -> "reasoning" -> "execution".

[0752] The final prompt message sent to the LLM is as follows:

[0753] ############

[0754] As an automated road test analysis expert, you must strictly adhere to the following rules:

[0755] ## background

[0756] The root causes of coverage issues on problematic road sections can be categorized into the following types: equipment alarms, missing neighboring cell coverage, antenna / feeder obstruction, and malfunctioning main control system. The root cause analysis process for coverage problems on these roads is as follows:

[0757] 1. First, extract key indicators (RSRP, SINR, weak coverage, etc.) of the problem road segment to determine the type of problem road segment.

[0758] 2. Then query the main control cell of the problem section, and output the list of the top 3 main control cells based on the sampling point ratio and signal strength ranking of the main control cell.

[0759] 3. Then check the alarm status of the main control cell. If there is an alarm, analyze the impact of the alarm and propose a solution. If there is no alarm, check the missing configuration of neighboring cells between the main control cells.

[0760] 4. If there are any missing configurations, analyze the impact of neighbor cell relationships and output a solution. If the neighbor cell configurations are normal, then there is an issue with the master control unit being unsuitable.

[0761] 5. Further analysis of the main control cell's operating parameters is needed. After confirming the antenna azimuth, electronic downtilt angle, and mechanical downtilt angle, multi-cell joint antenna optimization should be performed to complete the antenna adjustment and ultimately solve the problem section.

[0762] If the antenna and feeder adjustment scheme is not approved by the optimization personnel, it is necessary to rethink and re-reason, confirm the new optimal master control cell, and then complete the antenna and feeder adjustment by executing the multi-cell joint antenna and feeder optimization tool.

[0763] ## Core Mechanism

[0764] Employing a Thought→Action→Observation cyclical workflow:

[0765] - Thought: You should always think about what to do.

[0766] - Action: The action to be taken. The action can only be a tool in [{tool_names}], and no other content should be added.

[0767] - Observation: The result of the action

[0768] ## Tool Calling Standards

[0769] Only the following tools can be used, and multiple tools can be invoked at a time:

[0770] {tools}

[0771] 1) Problem Road Section Type Verification Tool

[0772] 2) Problematic road section TOP control cell identification tool

[0773] 3) Problematic road section main control area alarm verification tool

[0774] 4) Tool for checking missing configurations in the top 3 controlled communities

[0775] 5) Multi-cell joint antenna and feeder optimization tool

[0776] ## Output Rules

[0777] 1. Output JSON format when incomplete:

[0778] {{

[0779] "thought": "The reasoning process (must include the analysis of the previous round of observations)",

[0780] "action": {{"name": "tool name", "args": {{"parameter1":"value1"}}}},

[0781] "observation": "The tool returns the raw results"

[0782] }}

[0783] 2. Standards not met:

[0784] - No final answer returned

[0785] 3. Completion Criteria: Output the final answer if and only if any of the following conditions are met:

[0786] - TOP master control cell identification

[0787] - One or more master cells have alarms

[0788] - Detection of missing configurations in neighboring cells of the main control cell

[0789] - Multi-cell joint antenna and feeder optimization

[0790] - If the input is a greeting or goodbye, please respond directly in a friendly manner instead of using the Thought→Action→Observation loop workflow.

[0791] 4. Final answer in JSON format:

[0792] {{

[0793] "final_answer": {{"root_cause_type": "xxx", "root_cause": "xxxx"}}

[0794] }}

[0795] - root_cause_type: The root cause type of the problem section, including two types: device alarm and unreasonable main control.

[0796] - root_cause: First, describe the current situation of the problematic road section, and then analyze the main causes of the problem.

[0797] ## List Parsing Rules

[0798] - The returned master cell ECI must be strictly matched; it must not be modified or speculated upon.

[0799] - The returned format is ["Master Control Cell ECI1", "Master Control Cell ECI2", "Master Control Cell ECI3"]

[0800] ## Strict Contextual Coherence Requirements

[0801] - Each action must be directly based on the result of the previous observation; ignoring or distorting it is prohibited.

[0802] - The observation content must be analyzed word by word, and the next step should be performed based on the actual returned results (not assumptions).

[0803] - The master cell ECI must be strictly extracted from the observation content; fabrication or modification of the format is prohibited.

[0804] ############

[0805] After completion, proceed to step 113.

[0806] Step 113: LLM returns the tool selection result based on the Functioncall method.

[0807] LLM uses the prompt word sent in step 112, the user query, and all tool descriptions and parameter descriptions in the knowledge base, combined with the four decision steps in step 108, to finally deduce the business sub-agent to be called and the corresponding tool call output results, as shown in Table 7 below.

[0808] Table 7: Examples of Business Sub-Agent Invocation

[0809]

[0810] The LLM is used to determine which business sub-agent and its corresponding tool should be invoked in each round, and then the result is returned to the business sub-agent that made the request.

[0811] After that, proceed to step 114.

[0812] Step 114: Audit the input parameter results of the sending tool.

[0813] Using the tool call result returned by LLM in step 113, the "Input schema parameter" field in the tool output is sent as input parameter to post-processing for auditing. After completion, proceed to step 115.

[0814] Step 115: Return the result of the call parameters after audit calibration.

[0815] The main focus is on determining the accuracy of the input parameters. If the input parameters are correct, no modification is needed; if the input parameters are incorrect, the results are corrected and then returned to the business intelligence agent. After completion, proceed to step 116.

[0816] In one implementation, the central control intelligent agent generates a summary report on road network quality analysis based on the results of all task executions, and provides the summary report to answer user queries, specifically including:

[0817] The central control agent checks the execution results of each task, assembles all task execution results, and sends them to the summarizing agent.

[0818] The agent requests the LLM to summarize and generalize the results of all task executions, generating a summary report on road network quality analysis.

[0819] The summary agent sends the summary report to the central control agent, which then outputs the summary report to the user in response to the user's query.

[0820] In this embodiment, as Figure 4 As shown:

[0821] Step 116: After the tool is invoked, the task execution result of the sub-agent is returned.

[0822] After obtaining the post-processed and corrected input parameters in step 115, the corresponding tool is invoked for execution, and the tool execution result is returned to the planning agent. After completion, the planning agent judges the tool invocation result to determine whether all business sub-agents and their corresponding tool invocations involved in the decision content of step 108 have been completed. If not, it continues to step 109, looping until all sub-agent task invocations are completed, and then proceeds to step 117.

[0823] Step 117: Return the task results completed by all business sub-agents.

[0824] Based on the business sub-intelligent agent returned in step 116 and the corresponding tool call results below, send them to the central control intelligent agent, and then proceed to step 118.

[0825] Step 118: Check the completed tasks of all business sub-agents, assemble the content, and send it to the summarizing agent for summarization.

[0826] The central control agent checks the task completion status of the business sub-agents, assembles the feedback from all the business sub-agents, and then sends it together to the summarizing agent. After completion, proceed to step 119.

[0827] Step 119: After receiving the request, the agent sends it to the LLM for summarization.

[0828] The LLM will be responsible for the final summary of the content, after which we will proceed to step 120.

[0829] Step 120: Return to the summary content.

[0830] After LLM completes the summary, it feeds back to the summarizing agent, and then proceeds to step 121.

[0831] Step 121: Finally, return to the summary report.

[0832] After receiving the summary content, the summary agent generates a summary report and returns it to the central control agent. After completion, proceed to step 122.

[0833] Step 122: Return the complete solution to the user.

[0834] Finish

[0835] Currently, the automatic road test system is an application function designed based on the production operation process. The operation process is relatively complex and has the following main problems: 1) The operation is cumbersome and the threshold for use is high; 2) It is function-oriented, menu-driven, and has a complex system hierarchy; 3) It lacks personalized customization and adaptability.

[0836] To address the above issues, this paper proposes a road network quality analysis and processing method based on multi-agent automatic road testing in wireless networks. The concept of wireless network road testing agents is introduced into 4G and 5G automatic road testing systems. Agents are incorporated into the self-intelligent business processes of perception, cognition, decision-making, execution, and evaluation within the virtual automatic road testing system. By leveraging the synergy of large models and toolsets, the proportion of manual operations during the operation and updating of self-intelligent business processes is reduced, thereby improving the level of self-intelligence. This method can be designed for automatic road testing systems in any network or even higher-evolutionary networks and can be applied to automatic road testing agent products.

[0837] Example 2:

[0838] like Figure 4 As shown, this application provides a multi-agent road network quality analysis and processing device, the device comprising:

[0839] The central control intelligent agent is used to generate a planning request based on the user query for querying the quality of the road network, and then send the planning request to the planning intelligent agent.

[0840] The planning agent connects with the central control agent and is used to obtain information on the business sub-agents used to analyze road network quality from the business agent according to the planning request. Based on the information of the business sub-agents, it plans the calling decisions for each business sub-agent and sends the calling decisions to the business agent.

[0841] The business intelligence agent connects with the planning intelligence agent and is used to interact with the planning intelligence agent to invoke each business sub-intelligence agent according to the invocation decision, obtain the task execution results of each business sub-intelligence agent in analyzing road network quality, and send the task execution results to the central control intelligence agent.

[0842] The central control intelligent agent is also used to control the generation of a summary report on road network quality analysis based on the results of all task executions, and to provide the summary report to answer user queries.

[0843] In one embodiment, the central control intelligent agent specifically includes:

[0844] The dialogue module is used to receive user queries for road network quality and send the user queries to the preprocessing module.

[0845] The preprocessing module, connected to the dialogue module, is used to identify the intent of the user query, identify the preset intent category corresponding to the intent, determine the key parameters to be extracted and their preset description requirements based on the preset intent category, obtain the key parameters that meet the preset description requirements by combining the user query and the road network database, and return the intent and key parameters to the central control intelligent agent.

[0846] The expansion module, connected to the preprocessing module, is used to expand the user query based on intent and key parameters. The expansion content includes at least one of the following: querying road network quality indicators, analyzing quality problems, locating root causes of problems, and providing solutions, in order to form a planning request.

[0847] In one implementation, planning the intelligent agent specifically includes:

[0848] The RAG knowledge base query module is used to obtain the available business sub-agents that meet the planning request and adapt the business agents. It queries the RAG knowledge base for the functional descriptions of the available business sub-agents that adapt the business agents and requests the LLM to perform task planning based on the planning request and functional descriptions.

[0849] The LLM interaction module, connected to the RAG knowledge base query module, is used by LLM to select the appropriate business sub-intelligent agents from the available business sub-intelligent agents that meet the planning requirements based on the planning request and functional description. It also arranges the execution order of each appropriate business sub-intelligent agent and formulates the calling requirements for each appropriate business sub-intelligent agent to obtain the calling decision for each appropriate business sub-intelligent agent and returns it to the planning agent.

[0850] In one embodiment, the RAG knowledge base query module specifically includes functions for:

[0851] The planning agent obtains the currently available service sub-agents, including the adapted service agent of all service sub-agents used to analyze road network quality. All service sub-agents used to analyze road network quality include: road problem type analysis sub-agent, road problem master cell analysis sub-agent, alarm analysis sub-agent, neighbor cell analysis sub-agent, and antenna and feeder optimization sub-agent.

[0852] The functional description of the sub-agent for querying road problem types from the RAG knowledge base includes: inputting the road segment ID and user query time to query key performance indicator data of the road network, analyzing the types of road network quality problems based on the key performance indicator data of the road network, and generating road network quality problem analysis results in a preset format.

[0853] The functional description of the sub-agent for querying the RAG knowledge base for the main control cell analysis of road problems includes: obtaining all coverage cells and sampling point data of the road based on the road segment ID, and identifying the TOPN main control cells of the road from all coverage cells based on the sampling point data;

[0854] The functional description of querying the RAG knowledge base for the alarm analysis sub-agent includes: obtaining alarm information of the road TOPN master cell at the time of user query, analyzing the impact of each alarm message, and identifying alarms that cause road network quality problems;

[0855] The functional description of the neighbor cell analysis sub-agent for querying the RAG knowledge base includes: obtaining neighbor cells that correspond to the ECI of the main control cell of the road TOPN, establishing cell pairs that match the main control cell and neighbor cells, and identifying abnormal cell pairs with missing neighbor cell matching.

[0856] The function description of the antenna and feeder optimization sub-agent retrieved from the RAG knowledge base includes: generating an antenna and feeder optimization and adjustment scheme for the road network based on the type of road network quality problem, the alarms that caused the road network quality problem, and the abnormal cells with missing neighboring cells.

[0857] In one embodiment, the LLM interaction module specifically includes components for LLM:

[0858] Based on the road type being one of National Highway / Provincial Highway / County Road / Township Road / Expressway / High-speed Railway / Elevated Road, the calling requirements for the road problem type analysis sub-agent include: querying key performance indicator data of road networks of different road segment lengths, and locating road segments with road network quality problems;

[0859] The calling requirements for the main control cell analysis sub-agent for road problems include: obtaining the percentage of sampling points for road segments with road network quality problems, and obtaining the top 8 and top 3 main control cells with the largest percentage of sampling points;

[0860] The requirements for calling the alarm analysis sub-agent include: obtaining alarm information from the top 3 master cells within a preset time period, and diagnosing whether the alarm information affects the services of the master cells;

[0861] The calling requirements for the neighbor cell analysis sub-agent include: obtaining whether any of the top 8 main control cells have been missing from the neighbor cell configuration of the top 3 main control cells;

[0862] The requirements for calling the antenna feeder optimization sub-agent include: formulating an antenna feeder optimization and adjustment scheme based on the sampling contribution of associated cells of road segments with road network quality problems, the distance of the problematic road segments, and the voltage level.

[0863] In one implementation, the business intelligence agent interacts with the planning intelligence agent, specifically including:

[0864] The planning intelligent agent controls the business intelligent agent to call each adapted business sub-intelligent agent according to the calling order. The business intelligent agent obtains the task execution results of each business sub-intelligent agent in analyzing road network quality according to the calling requirements. The planning intelligent agent controls the acquisition of each task execution result and sends it to the central control intelligent agent.

[0865] In one embodiment, the business intelligence agent specifically includes:

[0866] The first controlled module is used to plan the intelligent agent to control the business intelligent agent to select the appropriate business sub-intelligent agent for the current call according to the calling order;

[0867] The calling module, connected to the first controlled module, is used by the business intelligence agent to request the LLM to generate calling parameters including the currently called adaptive business sub-intelligence according to the calling requirements, send the calling parameters to the post-processing module for auditing and calibration, use the audited and calibrated calling parameters to input the currently called adaptive business sub-intelligence, and obtain the task execution result of the currently called adaptive business sub-intelligence.

[0868] The loop execution module, connected to the calling module, is used to plan whether the intelligent agent should call the next adaptable business sub-intelligent agent based on the current task execution result. If yes, the control business intelligent agent calls the next adaptable business sub-intelligent agent; otherwise, the currently obtained task execution results are sent to the central control intelligent agent.

[0869] In one embodiment, the device further includes a summarizing agent connected to the central control agent, specifically including:

[0870] The second controlled module is used to receive all task execution results obtained after the central control agent checks the execution results of each task;

[0871] The summary and induction module, connected to the second controlled module, is used to summarize the results of all task executions requested by the agent from the LLM, and generate a summary report of road network quality analysis.

[0872] The results output module, connected to the summary and induction module, is used to send the summary report from the summary agent to the central control agent, which then outputs the summary report to the user in response to the user's query.

[0873] Example 3:

[0874] like Figure 5 As shown, Embodiment 3 of this application provides a computer-readable storage medium storing a computer program. When the computer program is run by a processor, it implements the multi-agent road network quality analysis and processing method as described in Embodiment 1.

[0875] 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 units, 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 technologies, CD-ROM (Compact Disc Read-Only Memory), DVD or other optical disc storage, cartridges, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer.

[0876] like Figure 6As shown, this application can also provide a computer device, including a memory and a processor. The memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the multi-agent road network quality analysis and processing method as described in Embodiment 1. This computer device can be the multi-agent road network quality analysis and processing apparatus as described in Embodiment 2.

[0877] The memory is connected to the processor. The memory can be flash memory, read-only memory or other types of memory. The processor can be a central processing unit or a microcontroller.

[0878] Embodiments 1-3 of this application provide a multi-agent road network quality analysis and processing method, apparatus, and medium. At least a central control agent, a planning agent, and a business agent are set up. The central control agent solves the collaboration problem among multiple agents, decomposes the user's complex problems into the planning agent, solves the complex task execution path problem, and then, based on the multiple business agents, accurately calls multiple tools according to the assigned task objectives, realizing full-link intelligent decision-making for dynamic task planning and autonomous execution. Finally, the entire detailed analysis process and its results are summarized and provided to frontline network personnel in the form of a report to assist in guiding the optimization of road network problems and improve the efficiency of problem closure.

[0879] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of this application, and this application is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and substance of this application, and these modifications and improvements are also considered to be within the scope of protection of this application.

Claims

1. A multi-agent road network quality analysis processing method, characterized by, The method includes: The central control agent generates a planning request based on the user query used to inquire about the quality of the road network, and sends the planning request to the planning agent; The planning agent obtains information about the business sub-agents used to analyze road network quality from the business agent according to the planning request, plans the calling decisions for each business sub-agent according to the business sub-agent information, and sends the calling decisions to the business agent. The business intelligence agent interacts with the planning intelligence agent to invoke each business sub-intelligence agent according to the invocation decision, obtain the task execution results of each business sub-intelligence agent analyzing the road network quality, and send the task execution results to the central control intelligence agent. The central control intelligent agent generates a summary report on road network quality analysis based on the results of all task executions, and provides the summary report to answer user queries.

2. The method of claim 1, wherein, The central control intelligent agent generates a planning request based on the user query used to inquire about road network quality, specifically including: The central control intelligent agent receives user queries for inquiring about road network quality and sends the user queries to the pre-processing module; The preprocessing module identifies the intent of the user query, identifies the preset intent category corresponding to the intent, determines the key parameters to be extracted and their preset description requirements based on the preset intent category, and obtains the key parameters that meet the preset description requirements by combining the user query and the road network database. The intent and key parameters are then returned to the central control intelligent agent. The central control intelligent agent expands the user query based on intent and key parameters. The expanded content includes at least one of the following: querying road network quality indicators, analyzing quality problems, locating root causes of problems, and providing solutions, in order to form a planning request.

3. The method according to claim 1 or 2, characterized in that, The planning agent retrieves information about the business sub-agents used for analyzing road network quality from the business agent based on the planning request, and plans the invocation decisions for each business sub-agent based on the business sub-agent information, specifically including: The planning agent obtains the available business sub-agents that meet the planning request and adapts to the business agent. It queries the RAG knowledge base for the functional descriptions of the available business sub-agents that adapt to the business agent and requests the LLM to perform task planning based on the planning request and functional descriptions. Based on the planning request and functional description, LLM selects the appropriate business sub-intelligent agents from the available business sub-intelligent agents that meet the planning request, arranges the execution order of each business sub-intelligent agent and formulates the calling requirements for each business sub-intelligent agent, so as to obtain the calling decision for each business sub-intelligent agent and return it to the planning agent.

4. The method of claim 3, wherein, The planning agent obtains available business sub-agents that meet the planning requirements and then queries the RAG knowledge base for the functional descriptions of the available business sub-agents that meet the planning requirements. Specifically, this includes: The planning agent obtains the currently available service sub-agents, including the adapted service agent of all service sub-agents used to analyze road network quality. All service sub-agents used to analyze road network quality include: road problem type analysis sub-agent, road problem master cell analysis sub-agent, alarm analysis sub-agent, neighbor cell analysis sub-agent, and antenna and feeder optimization sub-agent. The functional description of the sub-agent for querying road problem types from the RAG knowledge base includes: inputting the road segment ID and user query time to query key performance indicator data of the road network, analyzing the types of road network quality problems based on the key performance indicator data of the road network, and generating road network quality problem analysis results in a preset format. The functional description of the sub-agent for querying the RAG knowledge base for the main control cell analysis of road problems includes: obtaining all coverage cells and sampling point data of the road based on the road segment ID, and identifying the TOPN main control cells of the road from all coverage cells based on the sampling point data; The functional description of querying the RAG knowledge base for the alarm analysis sub-agent includes: obtaining alarm information of the road TOPN master cell at the time of user query, analyzing the impact of each alarm message, and identifying alarms that cause road network quality problems; The functional description of the neighbor cell analysis sub-agent for querying the RAG knowledge base includes: obtaining neighbor cells that correspond to the ECI of the main control cell of the road TOPN, establishing cell pairs that match the main control cell and neighbor cells, and identifying abnormal cell pairs with missing neighbor cell matching. The function description of the antenna and feeder optimization sub-agent retrieved from the RAG knowledge base includes: generating an antenna and feeder optimization and adjustment scheme for the road network based on the type of road network quality problem, the alarms that caused the road network quality problem, and the abnormal cells with missing neighboring cells.

5. The method of claim 4, wherein, Define the invocation requirements for each adapted business sub-agent, specifically including: Based on the road type being one of National Highway / Provincial Highway / County Road / Township Road / Expressway / High-speed Railway / Elevated Road, the calling requirements for the road problem type analysis sub-agent include: querying key performance indicator data of road networks of different road segment lengths, and locating road segments with road network quality problems; The calling requirements for the main control cell analysis sub-agent for road problems include: obtaining the percentage of sampling points for road segments with road network quality problems, and obtaining the top 8 and top 3 main control cells with the largest percentage of sampling points; The requirements for calling the alarm analysis sub-agent include: obtaining alarm information from the top 3 master cells within a preset time period, and diagnosing whether the alarm information affects the services of the master cells; The calling requirements for the neighbor cell analysis sub-agent include: obtaining whether any of the top 8 main control cells have been missing from the neighbor cell configuration of the top 3 main control cells; The requirements for calling the antenna feeder optimization sub-agent include: formulating an antenna feeder optimization and adjustment scheme based on the sampling contribution of associated cells of road segments with road network quality problems, the distance of the problematic road segments, and the voltage level.

6. The method of claim 3, wherein, The business intelligence agent interacts with the planning intelligence agent to invoke various business sub-intelligent agents based on the invocation decision, obtain the task execution results of each business sub-intelligent agent analyzing road network quality, and send the task execution results to the central control intelligence agent. Specifically, this includes: The planning intelligent agent controls the business intelligent agent to call each adapted business sub-intelligent agent according to the calling order. The business intelligent agent obtains the task execution results of each business sub-intelligent agent in analyzing road network quality according to the calling requirements. The planning intelligent agent controls the acquisition of each task execution result and sends it to the central control intelligent agent.

7. The method of claim 6, wherein, The planning intelligent agent controls the business intelligent agent to invoke each adapted business sub-intelligent agent according to the invocation order. The business intelligent agent obtains the task execution results of each business sub-intelligent agent analyzing road network quality according to the invocation requirements. The planning intelligent agent controls and obtains the task execution results and sends them to the central control intelligent agent, specifically including: The planning agent controls the business agent to select the appropriate business sub-agent for the current call based on the calling order; The business intelligence agent requests the LLM to generate call parameters including the currently called adaptive business sub-intelligence based on the call requirements. The call parameters are sent to the post-processing module for auditing and calibration. The audited and calibrated call parameters are then used to input the currently called adaptive business sub-intelligence to obtain the task execution result of the currently called adaptive business sub-intelligence. The planning agent determines whether to call the next adaptable business sub-agent based on the current task execution result. If yes, the control business agent calls the next adaptable business sub-agent; otherwise, it sends the currently obtained task execution results to the central control agent.

8. The method of claim 1 or 2, wherein, The central control intelligent agent generates a summary report on road network quality analysis based on the results of all task executions, and provides the summary report to answer user queries, specifically including: The central control agent checks the execution results of each task, assembles all task execution results, and sends them to the summarizing agent. The agent requests the LLM to summarize and generalize the results of all task executions, generating a summary report on road network quality analysis. The summary agent sends the summary report to the central control agent, which then outputs the summary report to the user in response to the user's query.

9. A multi-agent road network quality analysis processing apparatus characterized by comprising: The device includes: The central control intelligent agent is used to generate a planning request based on the user query for querying the quality of the road network, and then send the planning request to the planning intelligent agent. The planning agent connects with the central control agent and is used to obtain information on the business sub-agents used to analyze road network quality from the business agent according to the planning request. Based on the information of the business sub-agents, it plans the calling decisions for each business sub-agent and sends the calling decisions to the business agent. The business intelligence agent connects with the planning intelligence agent and is used to interact with the planning intelligence agent to invoke each business sub-intelligence agent according to the invocation decision, obtain the task execution results of each business sub-intelligence agent in analyzing road network quality, and send the task execution results to the central control intelligence agent. The central control intelligent agent is also used to control the generation of a summary report on road network quality analysis based on the results of all task executions, and to provide the summary report to answer user queries.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the multi-agent road network quality analysis and processing method as described in any one of claims 1-8.

Images