Road network quality optimization method and device and electronic equipment

Abstract

The invention discloses a road network quality optimization method and device and electronic equipment. The method comprises the following steps: acquiring poor-quality data in a road grid network; processing the poor-quality data through a drive test agent according to a ReAct mechanism, and generating a problem solution corresponding to the poor-quality data; whether the problem solutions can meet preset requirements or not is judged, if the problem solutions cannot meet the preset requirements, defects existing in the reflection problem solutions are simulated, and according to the defects, the problem solutions corresponding to the poor-quality data are iteratively generated; if the maximum number of iterations is reached or the problem solution can meet a preset requirement, confirming the problem solution as a target solution; and optimizing the poor-quality data according to the target solution, thereby realizing road network quality optimization based on the road test agent ReAct mechanism. The method can improve the efficiency and accuracy of road network quality optimization.

Description

Technical Field

[0001] This invention belongs to the field of network communication technology, and specifically relates to a method, apparatus and electronic equipment for optimizing road network quality. Background Technology

[0002] Traditional DT (drive testing) and CQT (call quality testing) methods have significant limitations: they are not only time-consuming, cumbersome, and costly, but they can only cover a limited number of roads and locations, making it difficult to accurately locate network problems. This results in some potential issues not being detected in time, directly affecting network quality. At the same time, these methods cannot detect potential problems in advance, nor can they fully reflect user experience. After user complaints, it is difficult to quickly locate and resolve the problems, leading to an increase in the complaint rate and further weakening user perception. Obviously, they cannot meet the need for rapid and accurate improvement of road network quality.

[0003] To improve this situation, a virtual road test solution based on MDT (Minimum Drive Testing, also known as "Crowdfunded Drive Testing") has emerged. This solution, from the user's perspective, uses MR (Measurement Report) data reported by User Equipment (UE) to achieve intelligent and precise location of network problems. It has played a role in reducing user complaint rates and improving user experience. Its advantages include flexible testing methods, low cost, and comprehensive and timely analysis. The specific process includes multi-dimensional data cleaning, road rasterization (using GIS technology to create road grates of different sizes), and raster aggregation. It combines indicators such as RSRP (Reference Signal Received Power) to assess road coverage quality and can also use AI mini-models to classify problem grates according to multiple dimensions such as coverage and interference, forming a problem management table. However, existing MDT-based virtual road testing systems still have significant shortcomings: they are cumbersome to operate and have a high barrier to entry, requiring frontline personnel to undergo professional training to master the operation steps; querying specific road indicators requires multiple clicks (e.g., querying road indicators for a certain scenario requires 7 clicks), and analyzing a single problem takes about 20 minutes; the system adopts a traditional menu-based design, which is hierarchical and lacks personalization and adaptability; cross-platform efficiency is low, with indicator extraction and problem analysis often relying on multiple systems and heavily depending on the experience of network optimization engineers, making it difficult to respond promptly to sudden network problems; in addition, the system development cycle is long (usually measured in months), requiring collaboration among multiple roles such as architecture, front-end, and testing, and the adjustment and adaptation capabilities are insufficient, which also fails to meet the actual needs of quickly and accurately improving road network quality.

[0004] In summary, both traditional DT and CQT methods, as well as existing MDT virtual road testing solutions, have their own limitations and cannot meet the needs of rapidly and accurately improving road network quality. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to address the above-mentioned shortcomings of the prior art by proposing a method, apparatus and electronic device for optimizing road network quality, which can improve the efficiency and accuracy of road network quality optimization.

[0006] In a first aspect, the present invention provides a road network quality optimization method based on the ReAct mechanism of a road test agent, the method comprising the following steps:

[0007] Step S1: Obtain the quality difference data in the road grid network;

[0008] Step S2: The road test agent processes the poor-quality data according to the ReAct mechanism and generates solutions to the problems corresponding to the poor-quality data.

[0009] Step S3: Determine whether the solution to the problem meets the preset requirements:

[0010] If the solution to the problem cannot meet the preset requirements, then the shortcomings of the solution are simulated and reflected upon, and steps S2 to S3 are repeated based on the shortcomings; if the maximum number of iterations is reached or the solution to the problem can meet the preset requirements, then the solution to the problem is confirmed as the target solution and the process proceeds to step S4.

[0011] Step S4: Optimize the poor quality data according to the target solution, thereby realizing road network quality optimization based on the ReAct mechanism of the road test agent.

[0012] Furthermore, in step S1, the quality difference data in the road grid network is obtained through virtual road testing.

[0013] Further, step S2 specifically includes:

[0014] By using a large language model of the road test agent to understand the poor-quality data, the task objective can be obtained;

[0015] The task objective is transformed into a road network problem to be processed and input into the large language model dialog box;

[0016] By querying and calling the target knowledge base through a large language model, and performing intent recognition and understanding on the road network problem to be addressed, the problem objective is obtained.

[0017] The tool module is invoked to process the problem objective and obtain an optimization solution;

[0018] Based on the optimization plan, generate solutions to the problems corresponding to the poor-quality data.

[0019] Furthermore, before querying and invoking the target knowledge base through the large language model, the method also includes: constructing the target knowledge base;

[0020] Building the target knowledge base involves the following steps:

[0021] Identify application scenarios, including road network quality;

[0022] Collect guidance manuals, analysis processes, coverage optimization processes, alarm analysis processes, coverage optimization knowledge cases, and alarm analysis knowledge cases related to application scenarios;

[0023] Transform the guidance manual, analysis process, coverage optimization process, alarm analysis process, coverage optimization knowledge cases, and alarm analysis knowledge cases into a structured or pre-structured knowledge form to form a target knowledge base.

[0024] Furthermore, before invoking the tool module, the method also includes: constructing the target knowledge base;

[0025] Initiate interaction with the knowledge base through RAG to obtain a list of tool capabilities.

[0026] Furthermore, the tool module is invoked to process the problem objective and obtain an optimization solution, specifically including the following steps:

[0027] Step A1: Identify the problem target, and when the problem target is the main source of poor quality data, call the TOP master cell identification tool to identify one or more master cells that have the most significant impact on road network quality, and obtain the target master cell;

[0028] Step A2: Call the alarm verification tool for the main control cell to confirm the equipment failure or abnormal parameters of the target main control cell and obtain the cause of the alarm;

[0029] Step A3: Based on the cause of the alarm, call the optimal master cell judgment tool to filter out the cells to be optimized;

[0030] Step A4: Call the multi-cell joint antenna and feeder optimization tool to design an antenna and feeder adjustment scheme for the cell to be optimized, and obtain the optimization scheme.

[0031] Secondly, the present invention provides a road network quality optimization device based on the ReAct mechanism of a roadside intelligent agent, the device comprising:

[0032] The acquisition unit is used to acquire poor quality data in the road grid network.

[0033] The generation unit, connected to the acquisition unit, is used to process the poor-quality data through the road test agent according to the ReAct mechanism and generate the corresponding problem solutions for the poor-quality data.

[0034] The judgment unit, connected to the generation unit, is used to determine whether the solution to the problem meets the preset requirements;

[0035] The reflection unit, connected to the judgment unit and the generation unit respectively, is used to simulate the defects in the problem solution when the judgment unit determines that the problem solution cannot meet the preset requirements, and trigger the generation unit to regenerate the problem solution corresponding to the poor quality data.

[0036] The determination unit, connected to the judgment unit and the generation unit respectively, is used to confirm the problem solution as the target solution when the judgment unit determines that the problem solution meets the preset requirements or when the maximum number of generation times of the generation unit reaches the maximum number of iterations.

[0037] The optimization unit, connected to the determination unit, is used to optimize the poor-quality data according to the target solution, thereby realizing road network quality optimization based on the ReAct mechanism of the road test agent.

[0038] Furthermore, the generation unit includes:

[0039] The first processing module is used to understand the poor-quality data through the large language model of the road test agent to obtain the task objective;

[0040] The second processing module, connected to the first processing module, is used to transform the task objective into a road network problem to be processed and input it into the large language model dialog box.

[0041] The third processing module, connected to the second processing module, is used to query and call the target knowledge base through the large language model, and to perform intent recognition and understanding on the road network problem to be processed, so as to obtain the problem target;

[0042] The fourth processing module, connected to the third processing module, is used to call the tool module to process the problem target and obtain the optimization solution;

[0043] The generation module, connected to the fourth processing module, is used to generate solutions to problems corresponding to poor-quality data based on the optimization scheme.

[0044] Furthermore, the fourth processing module includes:

[0045] The first processing submodule is used to identify the problem target, and when the problem target is the main source of the impact of poor quality data, it calls the TOP master cell identification tool to identify one or more master cells that have the most significant impact on road network quality, and obtain the target master cell;

[0046] The second processing submodule, connected to the first processing submodule, is used to call the alarm verification tool of the main control cell to confirm the equipment failure or parameter abnormality of the target main control cell and obtain the alarm cause.

[0047] The third processing submodule, connected to the second processing submodule, is used to call the optimal master cell judgment tool based on the alarm cause to filter out the cells to be optimized.

[0048] The fourth processing submodule, connected to the third processing submodule, is used to call the multi-cell joint antenna feeder optimization tool to design antenna feeder adjustment schemes for the cells to be optimized and obtain optimization schemes.

[0049] Thirdly, the present invention provides an electronic 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 road network quality optimization method based on the ReAct mechanism of the road test agent as described in the first aspect.

[0050] This invention introduces a road test agent based on the ReAct mechanism, combined with multi-round reasoning, action, and observation-reflection processes, to achieve intelligent analysis and optimization of poor-quality road network data. Specific beneficial effects include:

[0051] (1) This invention utilizes intelligent agents to automatically reason and generate targeted solutions, thereby improving the speed and accuracy of locating road network problems.

[0052] (2) Through repeated simulation and evaluation, this invention can ensure the scientific nature and effectiveness of the solution, thereby achieving rapid problem repair.

[0053] (3) The present invention adopts a dynamic feedback mechanism to adjust and optimize strategies in real time, which significantly improves the overall service quality and operational efficiency of the road network.

[0054] In summary, this invention can significantly improve the response speed and operational efficiency of road network quality improvement while ensuring high accuracy, thus meeting the needs for rapid, accurate, and intelligent road network optimization. Attached Figure Description

[0055] Figure 1 This is a schematic diagram of a road network quality optimization method based on the ReAct mechanism of a road test agent in an embodiment of the present invention;

[0056] Figure 2 This is a schematic diagram of a road test in an embodiment of the present invention;

[0057] Figure 3 This is a flowchart of the MDT test in an embodiment of the present invention;

[0058] Figure 4 This is a flowchart of the MDT localization algorithm in an embodiment of the present invention;

[0059] Figure 5 This is a flowchart of the highway gridding process in an embodiment of the present invention;

[0060] Figure 6 This is a schematic diagram of weak coverage in an embodiment of the present invention;

[0061] Figure 7 This is a diagram of the road network quality optimization framework based on the ReAct mechanism of the road test agent in this embodiment of the invention.

[0062] Figure 8 This is a flowchart illustrating the ReAct process in an embodiment of the present invention.

[0063] Figure 9 This is a schematic diagram of a road network quality optimization device based on the ReAct mechanism of a road test agent in an embodiment of the present invention;

[0064] Figure 10 This is an architectural diagram of an electronic device according to an embodiment of the present invention.

[0065] In the attached figures, the reference numerals are as follows: 10, acquisition unit; 20, generation unit; 30, judgment unit; 40, reflection unit; 50, determination unit; 60, optimization unit; 100, processor; and 200, memory. Detailed Implementation

[0066] To enable those skilled in the art to better understand the technical solution of the present invention, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0067] It is understood that the specific embodiments and accompanying drawings described herein are merely for explaining the invention and are not intended to limit the invention.

[0068] It is understood that, without conflict, the various embodiments and features in the embodiments of the present invention can be combined with each other.

[0069] It is understood that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, while the parts unrelated to the present invention are not shown in the drawings.

[0070] It is understood that each unit or module involved in the embodiments of the present invention may correspond to only one entity structure, or may be composed of multiple entity structures, or multiple units or modules may be integrated into one entity structure.

[0071] It is understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of this invention may occur in a different order than that marked in the accompanying drawings.

[0072] It is understood that the flowcharts and block diagrams of this invention illustrate the possible architecture, functions, and operations of systems, apparatuses, devices, and methods according to various embodiments of this invention. Each block in the flowchart or block diagram may represent a unit, module, program segment, or code, containing executable instructions for implementing the specified function. Furthermore, each block or combination of blocks in the block diagram and flowchart can be implemented using a hardware-based system to achieve the specified function, or using a combination of hardware and computer instructions.

[0073] It is understood that the units and modules involved in the embodiments of the present invention can be implemented by software or by hardware. For example, the units and modules can be located in a processor.

[0074] Example 1:

[0075] This embodiment provides a road network quality optimization method based on the ReAct mechanism of road test agents. This method can be applied to urban road maintenance and optimization, rapid response to network faults, intelligent traffic scheduling, speed measurement and road condition analysis, new route planning, automated detection and system integration, etc., effectively improving traffic capacity, communication stability and road safety, and providing strong technical support for the intelligent development of urban infrastructure and intelligent transportation systems.

[0076] like Figure 1 As shown, the road network quality optimization method based on the ReAct mechanism of the road test agent in this embodiment includes steps S1 to S4.

[0077] Step S1: Obtain the poor quality data in the road grid network.

[0078] As a preferred implementation method, the quality difference data in the road grid network is obtained through virtual road testing.

[0079] Drive testing is an indispensable part of network optimization. Traditional drive testing methods (DT&CQT testing) rely on professional testers driving specialized vehicles to conduct on-site tests on target routes to obtain network coverage data. This method is inefficient, accounting for approximately 45% of the total cost of the network optimization project, and requires highly skilled personnel, specialized equipment, and vehicles. Data collection has certain limitations, testing efficiency is low, and it is difficult to meet the needs of real-time and large-area coverage. This embodiment adopts virtual drive testing technology, fundamentally changing the traditional drive testing method and bringing operators both cost and efficiency improvements. Virtual drive testing is based on massive MR (Measurement Report) data from real users, combined with AGPS positioning information, and associates this data with CDT (Call Detail Trace) call records, performs road grid matching, and combines GIS visualization to achieve network coverage analysis results similar to on-site drive testing. Virtual drive testing has the advantages of wider coverage and closer resemblance to the actual situation of end users, becoming an important tool for wireless network maintenance and optimization. The virtual road test implementation steps in this embodiment include three parts: First, route definition, which involves importing a map and combining manually defined and historical road test data to rasterize urban roads. Second, data extraction, which involves extracting MR data uploaded by pedestrians and vehicles carrying mobile phones that continuously appear on urban roads. This data, containing AGPS information, is then linked to call detail records (CDRs) and matched to the road raster to accurately reflect network quality. Third, results display, which involves statistically analyzing the filtered data and using GIS to mark coverage, anomalies, and other information with different colors to aid in network performance analysis. A comparison between traditional and virtual road tests is shown in Table 1.

[0080] Table 1: Comparison of Traditional Road Testing and Virtual Road Testing

[0081]

[0082] Virtual drive testing aims to improve the efficiency of wireless network operation and maintenance and optimization, partially replacing traditional drive testing and becoming an innovative coverage analysis and RF (radio frequency) optimization solution. The main differences between the two include: in terms of tools, traditional drive testing relies on drive testing and analysis software, while virtual drive testing is based on a system development platform, offering higher stability; in terms of data sources, traditional methods use a single test terminal to collect data, while virtual drive testing utilizes a large amount of MR data carrying latitude and longitude coordinates; in terms of analysis scope, traditional drive testing mainly targets vehicular access roads, while virtual drive testing can cover a wider range of outdoor scenarios; in terms of functionality, traditional drive testing focuses on coverage, event, and throughput analysis, while virtual drive testing primarily performs coverage analysis (including uplink data). Furthermore, virtual drive testing requires no additional manpower, tools, or vehicle resources, saving costs.

[0083] Figure 2This demonstrates the evolution from traditional drive testing to virtual drive testing. The traditional drive testing (DT / CQT) shown on the left relies on manual testing with a test terminal, actively collecting base station signals through vehicle-based road testing or fixed-point call testing. Because it requires manual on-site testing of each point and road segment, it suffers from limited coverage (only a small number of roads can be tested), is time-consuming, costly, and lacks complete coverage, reflecting the drawbacks of traditional testing methods. The virtual drive testing shown on the right, however, eliminates the need for manual on-site participation. It passively reports data through user terminals (UEs), automatically collecting and reporting network signal data (such as signal strength and latency) via MR (Measurement Report) / CDT (Call Detail Tracking), combined with AGPS. (Assisted by the Global Positioning System) to achieve more accurate positioning, and then through the cloud to aggregate data from multiple terminals, to achieve full coverage perception of the entire road segment, with advantages such as intelligence and wide coverage, corresponding to the characteristics of the MDT-based virtual road test solution; the arrows in the figure show that the road test mode has been upgraded from the human-driven "human + equipment to the site" to the data-driven "automatic terminal reporting + cloud aggregation and analysis", solving the problems of incomplete coverage and high cost of the traditional mode, echoing the design logic of virtual road test to replace traditional road test to improve efficiency and coverage. The overall diagram uses a very simple illustration to clearly explain the core logic of virtual road test to replace traditional manual road test by terminal data reporting, intuitively showing the transformation of the technical solution from human-driven to intelligent data-driven.

[0084] This embodiment employs a virtual drive test scheme based on Multi-Dimensional Testing (MDT). From the user's perspective, it utilizes UE (User Equipment) mobile phone measurement reports (MR) to intelligently and accurately locate network problems, effectively reducing user complaint rates and improving user experience. Its core processes include multi-dimensional data cleaning, road test analysis, road problem output, and automatic closed-loop evaluation. In terms of innovation, the team researched MR localization algorithms for different road scenarios and proposed a virtual drive test scheme that includes road rasterization, data cleaning, raster aggregation, and road coverage GIS visualization. Comparative verification with traditional manual drive testing demonstrates that virtual drive testing has good feasibility and high accuracy, thus exhibiting significant advantages in testing methods, cost investment, personalized analysis, comprehensiveness, and timeliness. Utilizing big data analytics for in-depth analysis of MR data reported by mobile phones helps reduce the workload of manual testing, achieving cost reduction and efficiency improvement, and providing a new technical path for network optimization.

[0085] The virtual drive test solution is primarily based on MDT (Minimum Drive Test) technology, which optimizes the network by collecting measurement reports (including GPS, RSRP, latency, UL_SINR, etc.) from commercial terminals. This technology relies on user terminal-reported data, collecting data through different measurement events such as ImmediateMDT and LoggedMDT, mainly using measurement reports in both non-real-time and real-time states. To ensure the effectiveness of virtual drive testing, an innovative high-precision positioning algorithm based on AGPS has been optimized, abandoning the traditional fingerprint recognition algorithm. This significantly reduces hardware performance and storage requirements, achieving a positioning accuracy of approximately 20 meters in urban areas, densely populated communities, and highway scenarios. This allows for accurate reflection of network coverage quality on GIS cloud maps, supporting network planning and optimization.

[0086] like Figure 3 As shown, this embodiment uses MDT drive testing. The core of MDT (Minimum Drive Testing, also known as "Crowdfunding Drive Testing") is a closed-loop automated drive testing system of "network management configuration → terminal measurement → data reporting". It aims to use commercial terminals (such as mobile phones and other UEs) to replace traditional manual drive testing and achieve network automation and lightweight evaluation. The specific process is as follows: First, the network side (such as the operator's network management system) generates MDT measurement tasks based on network optimization needs such as regional coverage and interference investigation. Measurement indicators and trigger conditions are then sent to the terminal via base stations, core network equipment, etc. Next, after receiving the configuration, the terminal (UE) automatically starts measurement when conditions such as specific areas or signal changes are met. It collects network and terminal data including GPS (terminal location), RSRP (signal strength), latency (transmission delay), PHR (remaining transmit power), UL_SINR (uplink signal-to-interference-plus-noise ratio), and neighboring serving cell information (surrounding base station conditions). At this time, the mobile phone acts like a "mini road tester," silently collecting comprehensive network signal information. Then, the terminal selects either "Logged MDT" or "Immediate MDT" reporting methods according to its needs. "Logged MDT" stores data in idle state, and after the terminal reconnects and the storage and network idle conditions are met, the cached data is sent back to the base station (such as ENodeB / RNC). This method does not affect user experience and reduces the occupation of real-time bandwidth. "Immediate MDT," on the other hand, stores data in idle state. "MDT" (Multi-Demand Testing) involves testing and transmitting data while connected, enabling rapid response to unexpected issues. However, it consumes real-time bandwidth, necessitating a balance between real-time performance and network load. MDT utilizes "crowdsourced" data from terminals, offering wider coverage and lower costs. It can replace traditional manual road testing, which is limited in manpower and coverage. By using GPS and RSRP data from numerous terminals, it accurately locates weak coverage and interference sections (such as sections where multiple terminals report low RSRP), enabling targeted network optimization.

[0087] Figure 4This document demonstrates the detailed workflow of the MDT (Minimum Drive Test) positioning algorithm. Starting with the original MRO (Measurement Report Optimization) file, it progressively processes the data and ultimately visualizes it through a GIS platform. The MRO file contains MR (Measurement Report) data reported by the terminal (such as a mobile phone) when connecting to the network. This data is rich, including RSRP (Reference Signal Received Power), RSRQ (Signal Quality), TA (Time Advance), and the latitude and longitude information of the terminal's location. Although there may be accuracy or formatting issues, it forms the data foundation for the entire MDT process. Subsequently, measurement report data containing latitude and longitude information is filtered out. Using programming scripts or tools, it is parsed according to formats such as XML and JSON to extract key information related to location and network signal strength. To ensure positioning accuracy, the latitude and longitude information also needs to be processed, including standardizing decimal places, removing erroneous or out-of-range data, and performing coordinate transformations (such as from WGS84 to GCJ-02) to adapt to subsequent processing. Next, using GIS technology, the processed latitude and longitude information is divided into grids of different sizes. MR data falling into the same grid are aggregated and statistically analyzed, for example, calculating the average RSRP and RSRQ values ​​within each grid to reflect the network coverage quality of the area. Finally, through the GIS system, these coverage conditions are visually displayed on a map using different colors or symbols, such as red for weak coverage areas and green for good coverage. This helps network optimization engineers locate weak areas and develop targeted network optimization measures, such as adjusting base station transmit power or antenna pointing, thereby improving the overall network coverage and performance.

[0088] MR rasterization, such as Figure 5 As shown, this is the rasterization process for MR (Measurement Report) in high-speed scenarios. First, high-speed information is selected and read into memory. After JSON conversion, combined with latitude and longitude correction, it is converted into high-speed locations in Baidu's latitude and longitude format. Then, rasterization is performed to obtain tiled high-speed areas. Simultaneously, MR coverage information is associated with the tiled high-speed areas, entering a massive data association stage. Baidu Maps is used to perform high-speed scene processing on the MR coverage information, ultimately achieving raster visualization on the map. The entire process is based on GIS technology, using a high-speed electronic map as the foundation. The road raster size can be set to 10m x 10m, 20m x 20m, or 50m x 50m. For example, a 20m x 20m road raster is generated with the high-speed line as the center and 20m to the left and right. During the rasterization process, secondary data cleaning is performed simultaneously to remove MR sampling points that do not fall into the road raster. Typically, one week's worth of data is accumulated to meet the quality requirements of virtual road test analysis, thus completing the rasterization process for highways and providing strong support for network optimization.

[0089] After MR (Match Array) rasterization is completed, the sampling points within the MR need to be aggregated into network coverage metrics at the raster granularity. The RSRP (Reference Signal Received Power) metric is typically used to evaluate LTE wireless network coverage. The automatic weak coverage detection algorithm first identifies problematic raster cells suspected of having weak coverage, then assesses the coverage quality of the surrounding eight raster cells. If the number of weak coverage raster cells is ≥3, the central raster area is defined as a weak coverage area. Metrics include average RSRP and RSRP weak coverage ratio. For example, in rural areas, -105dBm is used as the weak coverage threshold, and average RSRP is used for evaluation. In urban areas, raster cells with an RSRP ≤ -105dBm ratio exceeding 20% ​​are considered weak coverage, and RSRP is used for weak coverage ratio determination. Furthermore, multiple raster levels such as 2020 and 5050 are implemented. Rural scenarios use 100×100 raster cells, and urban scenarios use 50×50 raster cells. For specific results of the automatic weak coverage identification, please refer to [link to relevant documentation]. Figure 6 From the perspective of MDT positioning algorithm process and implementation, the current system only establishes road grids based on high-precision positioning algorithms and associates the latitude and longitude of each grid with MR information to obtain the coverage status of a single grid, connecting them to form the coverage of the entire road. Subsequently, after MR rasterization, the sampling points will be aggregated to the corresponding index values, mainly using RSRP to evaluate 4G and 5G network coverage.

[0090] Step S2: The road test agent processes the poor-quality data (including reasoning and action) according to the ReAct mechanism to generate solutions to the problems corresponding to the poor-quality data.

[0091] As a specific implementation method, step S2 specifically includes steps S21 to S25.

[0092] Step S21: Use the large language model of the road test agent to understand the poor quality data and obtain the task objective.

[0093] Step S22: Transform the task objective into a road network problem to be processed and input it into the large language model dialog box.

[0094] Step S23: Query and call the target knowledge base through the large language model, and perform intent recognition and understanding on the road network problem to be processed to obtain the problem target.

[0095] As a specific implementation method, before querying and calling the target knowledge base through the large language model, the method also includes: constructing the target knowledge base;

[0096] Building the target knowledge base involves the following steps:

[0097] Identify application scenarios, including road network quality;

[0098] Collect guidance manuals, analysis processes, coverage optimization processes, alarm analysis processes, coverage optimization knowledge cases, and alarm analysis knowledge cases related to application scenarios;

[0099] Transform the guidance manual, analysis process, coverage optimization process, alarm analysis process, coverage optimization knowledge cases, and alarm analysis knowledge cases into a structured or pre-structured knowledge form to form a target knowledge base.

[0100] Step S24: Call the tool module to process the problem target and obtain the optimization solution.

[0101] As a specific implementation method, before invoking the tool module, the method further includes: constructing the target knowledge base;

[0102] As another specific implementation method, the tool module is invoked to process the problem objective and obtain an optimization solution, which specifically includes the following steps:

[0103] Step A1: Identify the problem target, and when the problem target is the main source of poor quality data, call the TOP master cell identification tool to identify one or more master cells that have the most significant impact on road network quality, and obtain the target master cell;

[0104] Step A2: Call the alarm verification tool for the main control cell to confirm the equipment failure or abnormal parameters of the target main control cell and obtain the cause of the alarm;

[0105] Step A3: Based on the cause of the alarm, call the optimal master cell judgment tool to filter out the cells to be optimized;

[0106] Step A4: Call the multi-cell joint antenna and feeder optimization tool to design an antenna and feeder adjustment scheme for the cell to be optimized, and obtain the optimization scheme.

[0107] Initiate interaction with the knowledge base through RAG to obtain a list of tool capabilities.

[0108] Step S25: Based on the optimization plan, generate solutions to the problems corresponding to the poor-quality data.

[0109] Step S3: Determine whether the solution to the problem meets the preset requirements:

[0110] If the solution to the problem does not meet the preset requirements, then the solution is simulated and the defects are reflected upon. Based on the defects, steps S2 to S3 are repeated. If the maximum number of iterations is reached or the solution to the problem meets the preset requirements, then the solution to the problem is confirmed as the target solution and the process proceeds to step S4.

[0111] Step S4: Optimize the poor quality data according to the target solution, thereby realizing road network quality optimization based on the ReAct mechanism of the road test agent.

[0112] This embodiment, based on the ReAct mechanism of the road test agent, constructs a complete process for optimizing wireless network road problems. Starting with user input of a high-speed, weak coverage problem, it relies on the collaborative operation of a large model, knowledge base, and toolset. Through multiple rounds of closed-loop "observation-reasoning-action-feedback," it achieves accurate problem diagnosis and continuous iterative optimization of the solution. The entire process is as follows: Figure 7 As shown, it covers steps one through eleven.

[0113] Step 1: Use the large network model as the interaction entry point to input the user's wireless network road problem into the dialog box.

[0114] Users can input the wireless network issues they need to analyze and resolve into the dialog box using natural language interaction:

[0115] For example: A certain highway section in a certain city, ID 28745-S1, has a 100-meter coverage issue. The key indicators for this problem section are as follows:

[0116] 1) RSRP = -105.60 dBm;

[0117] 2) SINR = 4.89 dB;

[0118] 3) Weak coverage rate = 62.15%;

[0119] Please analyze how to solve this problem section of road.

[0120] After inputting into the large model dialog box, the LLM (Large Language Model) performs intent recognition and understanding of the problem. That is, after the first "observation" of the road problem information obtained from the user, the solution dynamic reasoning will be realized through the collaboration of the large model, knowledge base, toolset, and executor based on the agent ReAct process. After the process is completed, the second step will be performed.

[0121] Step Two: After receiving the user's question, the LLM expands the question using prompts. The goal is to enhance its intelligence by dynamically reasoning about road problem optimization solutions through the collaboration of a large model, knowledge base, toolset, and executor, based on the agent's ReAct process. Therefore, before deciding which toolset capabilities to invoke after each LLM consideration, it needs to interact with the knowledge base via RAG. It needs to obtain relevant professional knowledge and toolset functions related to the reasoning task from the knowledge base to confirm the required tool functions. The problem analysis is completed through a cyclical process of "observation" -> "reasoning" -> "execution".

[0122] 1) The specific system-level prompts are as follows:

[0123] You are a professional automated road test analysis AI assistant, and you must strictly follow the following logic to operate:

[0124] (1) Background:

[0125] The root cause analysis process for problematic road coverage issues is as follows: First, extract key indicators (RSRP, SINR, weak coverage rate, etc.) for the problematic road sections, and output a list of the top 3 main control cells based on the proportion of sampling points of the main control cells and the signal strength ranking.

[0126] Prioritize checking the alarm status of the master cell. If an alarm exists, directly analyze the impact of the alarm and propose a solution. If no alarm exists, check the optimal master cell for judgment. After confirmation, complete the antenna and feeder adjustment by executing the multi-cell joint antenna and feeder optimization tool.

[0127] 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.

[0128] (2) Core Mechanism

[0129] Employing a Thought→Action→Observation cyclical workflow:

[0130] -Thought: You should always think about what to do.

[0131] -Action: The action to be taken. The action can only be a tool in [{tool_names}], and no other content should be added.

[0132] -Observation: The result of an action.

[0133] (3) Tool calling specifications

[0134] Only the following tools can be used, and multiple tools can be invoked at a time:

[0135] {tools}

[0136] (4) Output rules

[0137] (4.1) Output JSON format when not completed:

[0138] {{

[0139] "thought": "The reasoning process (must include the analysis of the previous round of observations)"

[0140] "action":{{"name":"tool name","args":{{"parameter 1":"value 1"}}}},

[0141] "observation": "The tool returns the raw results"

[0142] }}

[0143] (4.2) Incomplete standards:

[0144] - No final answer has been returned.

[0145] (4.3) Completion Criteria: Output the final answer if and only if any of the following conditions are met:

[0146] -TOP master control cell identification

[0147] - One or more master cells have alarms

[0148] - Optimal controlling cell determination

[0149] - Multi-cell joint antenna optimization

[0150] - If the input is a greeting or goodbye, please respond directly in a friendly manner instead of using the Thought→Action→Observation loop workflow.

[0151] (4.4) Final answer in JSON format:

[0152] {{

[0153] "final_answer":{{"root_cause_type":"xxx","root_cause":"xxxx"}}

[0154] }}

[0155] -root_cause_type: The root cause type of the problem section, including two types: device alarm and unreasonable main control.

[0156] -root_cause: First, describe the current situation of the problematic road section, and then analyze the main causes of the problem.

[0157] (5) List parsing rules

[0158] - The returned master cell ECI must be strictly matched; it must not be modified or speculated.

[0159] - The return format is ["Master Control Cell ECI1","Master Control Cell ECI2","Master Control Cell ECI3"]

[0160] (6) Strict requirement for contextual coherence

[0161] - Each action must be directly based on the result of the previous observation; ignoring or distorting the result is prohibited.

[0162] - The observation content must be analyzed word by word, and the next step should be executed based on the actual returned results (not assumptions).

[0163] - The master cell ECI must be strictly extracted from the observation content; fabrication or modification of the format is prohibited.

[0164] """

[0165] 2) ReAct Reasoning-Action-Observation (Feedback) Process Design

[0166] The ReAct abstract class is designed and inherited by the Agent class. It includes an LLM (Limited Learning Model) and a Memory or Knowledge base, and has two phases: think and act. The think phase involves inference through calls to the LLM, while the act phase mainly involves calling tools to obtain additional information. Steps are chained together, with each step repeatedly called by the Agent. Each step uses LLM inference to determine whether a tool call is necessary. The specific implementation flowchart and core code are as follows:

[0167] Core code implementation process:

[0168]

[0169]

[0170] Figure 7Deeply integrated with the ReActAgent class code, it fully demonstrates the working mechanism of the agent in solving road coverage problems, constructing a closed-loop process of "think-action-observation". Starting with Start, when a user requests "analyze road coverage problems", the core loop is entered: In the "Think-Act" step of "Step", the agent calls the step method to initiate a single-step process. First, it executes think (corresponding code should_act = await self.think()), relying on the Large Language Model (LLM) combined with system_prompt (system instruction), next_step_prompt (next step prompt), and memory (knowledge base / historical data) to analyze the coverage indicators of the current road segment, the effectiveness of historical tools, and other information. It outputs a boolean value should_act to determine whether to call a tool. For example, when the user needs to analyze weak coverage, LLM infers "the TOP controlling cell needs to be identified first", so think returns True, entering the "Act" step (corresponding code return await). `self.act()` calls the toolset to perform actions such as `get_TopN_cell` to identify the top controlling cells. The tool's execution results are fed back as observations for the next round of decision-making. In the loop, after each tool returns data (e.g., a list of the top 3 controlling cells), the agent "observes" the results and stores them in memory (corresponding to "Memory or Knowledge base" in the flowchart), providing input for the next round of `Think`. After each Step is completed, the loop is controlled based on whether the "Max Step" (maximum number of loop steps) has been reached. If the maximum number of steps has not been reached and the tool is not a "terminating tool," the loop continues back to `Think`. If the maximum number of steps has been reached or the tool is a "terminating tool," the process enters End. From the mapping between the code and the flowchart, in the `ReActAgent` class attributes, `name` and `description` identify the agent's identity; `system_prompt` and `next_step_prompt` correspond to the LLM "thinking basis" in the flowchart; and `llm` and `memory` correspond to the "inference core" and "data support," respectively. In terms of methodology, "think" corresponds to "thinking" and decision-making, "act" corresponds to "action" and calling tools, and "step" connects single-step loops to fully map process nodes.In short, the flowchart visually presents the ReActAgent's operating mechanism. Starting with a user request, it goes through the Think function to decide whether to call the tool, the Act function to execute the tool and observe the results, and combined with the MaxStep control loop, to achieve automated analysis of road coverage issues. The think, act, step, and other methods in the code correspond one-to-one with the flowchart nodes, together constructing the ReAct closed loop of "think-action-observation". This allows the agent to "analyze the problem → call the tool → iteratively optimize" like a human expert, achieving automated solutions to complex tasks such as network drive testing.

[0171] 3) Knowledge base construction

[0172] The knowledge base primarily collects documents such as road optimization guidelines, analysis processes, coverage optimization strategies, and alarm analysis, specifically including:

[0173] The knowledge case studies covering optimization are as follows:

[0174] ① AI Solution for Self-Optimization of Wireless Network Parameters.docx

[0175] ②Common routine inspection items.docx

[0176] ③ AI Toolbox Algorithm Flow Introduction.docx

[0177] ④ Centralized Analysis and Optimization Manual.docx

[0178] ⑤SWAT Analysis Workflow.drawio

[0179] ⑥ Casual Test Analysis Process Description.pdf

[0180] ⑦ Weak Coverage Analysis.pptx

[0181] ⑧ Problem Road Segments and Problem Clusters Rules - V20220711V2.xlsx

[0182] ⑨ Road Intelligent Diagnosis - Analysis Process v1.xmind

[0183] ⑩ Coverage optimization approach.xmind

[0184] The above document serves as a guide for the expert analysis steps and thought process of large-scale model generation for wireless coverage optimization.

[0185] Alarm analysis knowledge examples are shown in Table 2.

[0186] Table 2: Rules for analyzing the impact of wireless alarms on coverage:

[0187]

[0188]

[0189] After performing initial reasoning based on the user's intent in the question, the system initiates an interaction with the knowledge base via RAG to obtain a list of tool capabilities from the knowledge base.

[0190] The knowledge base contains accumulated expert experience related to road problem optimization analysis in automated road testing scenarios, mainly divided into four tool capabilities, and these tools are described in a standardized manner:

[0191] The toolset in the Automated Road Test Knowledge Base includes the following:

[0192] (1) TOP master cell identification tool:

[0193] The Top N controlling cells used to identify problematic road sections are defined by the following rules:

[0194] All sampling points in the problematic road section area were aggregated and statistically analyzed.

[0195] The sampling points converged on the problem road section under different master control ECI (cell name) were statistically analyzed. The top 8 master control cells were selected according to the sampling point percentage. The top three cells with the highest sampling point percentage were selected as the Top 3 master control cells, as shown in Table 3.

[0196] Table 3: Sorting table by percentage of sampling points:

[0197]

[0198]

[0199] The tool usage instructions are as follows:

[0200]

[0201]

[0202] (2) Alarm verification tool for main control cell

[0203] 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, as shown in Table 4.

[0204] Table 4: Alarm Information Table Related to the Main Control Cell:

[0205]

[0206] The tool usage instructions are as follows:

[0207]

[0208]

[0209] (3) Optimal controlling cell determination tool

[0210] Based on the results of the TOP master cell identification tool, the average voltage level, sampling point ratio, and distance from the problem road segment of the master cell are combined and calculated. Finally, the optimal master cell is determined based on the calculation result of the "optimal master cell judgment factor". The specific calculation formula is as follows:

[0211] Optimal master control judgment factor = (absolute value of "average level") * 0.3 + (sampling percentage) * 0.4 + (distance from problem section) * 0.3

[0212] The optimal master control judgment factors are sorted in descending order, and the one with the largest final numerical result is selected as the optimal master control judgment result. The table of optimal master control judgment factors is shown in Table 5.

[0213] Table 5: Optimal Master Control Judgment Factor Table:

[0214]

[0215] The tool usage instructions are as follows:

[0216]

[0217]

[0218] (4) Multi-cell joint antenna and feeder optimization tool

[0219] By analyzing the sampling contribution of related cells in the road segment, the distance and signal strength of the problematic road segment, and the results of calling the multi-cell joint antenna and feeder optimization tool, it was determined that there is room for antenna and feeder optimization, and an attempt was made to enhance the main control coverage. The final antenna and feeder adjustment scheme was determined by predicting the effect, as shown in Table 6.

[0220] Table 6: Multi-cell joint antenna and feeder optimization table:

[0221]

[0222]

[0223] The tool usage instructions are as follows:

[0224]

[0225]

[0226] After interacting with the knowledge base via RAG, the functional descriptions and usage instructions for each tool are obtained. This retrieved knowledge of tool functions and usage instructions, combined with knowledge base documentation and user questions, is then returned to the larger model.

[0227] User questions + knowledge base documentation + tool function descriptions + tool usage instructions.

[0228] The large model infers from the content description of "user question + knowledge base document + prompt word template + tool function description + tool usage instructions" to select the corresponding tool to obtain key data, based on the "background description" in the prompt words, i.e.

[0229] The root cause analysis process for problematic road coverage issues is as follows: First, extract key indicators (RSRP, SINR, weak coverage rate, etc.) for the problematic road sections, and output a list of the top 3 main control cells based on the proportion of sampling points of the main control cells and the signal strength ranking.

[0230] Prioritize checking the alarm status of the master cell. If an alarm exists, directly analyze the impact of the alarm and propose a solution. If no alarm exists, check the optimal master cell for judgment. After confirmation, complete the antenna and feeder adjustment by executing the multi-cell joint antenna and feeder optimization tool.

[0231] 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.

[0232] After completing the first reasoning result, you will select the following two tools:

[0233] TOP master cell identification tool;

[0234] Alarm verification tool for main control cell;

[0235] After the inference is completed, the large model sends the inference results to the executor, and then proceeds to step three.

[0236] Step 3: After receiving the calling tool address from the large model, the executor initiates calling requests to the TOP master cell identification tool and the master cell alarm verification tool based on the address information, and feeds back the key input parameters of the tool calls, i.e., the execution result (required), to the large model. After completion, proceed to Step 4.

[0237] Step 4: After the large model obtains the result data from the TOP master cell identification tool and the master cell alarm verification tool, observe the relevant results, such as:

[0238] The top controlling cell was found to be "[cell1_1, cell1_2, cell2_1]".

[0239] The alarm received from the main control cell is as follows:

[0240] cell1_1 alarm name: Cell PIC unavailable

[0241] cell1_2alarm name: None

[0242] cell2_1alarm name: None

[0243] Proceed to step five.

[0244] Step 5: Perform a second inference based on the observation results, and use RAG to interact with the knowledge base. Following the process in Step 2, obtain the list of tool capabilities in the knowledge base, alarm analysis knowledge cases and coverage optimization knowledge cases in the knowledge base construction, and combine them with the TOP master cell identification tool and master cell alarm verification tool call result data from Step 4 to finally arrive at the "master cell alarm analysis result":

[0245] The alarm information from the main control cell is unrelated to coverage issues (because the impact of "cell PIC unavailable" is low).

[0246] The antenna and feeder optimization tool needs to be enabled to generate an antenna and feeder adjustment plan.

[0247] After the inference is completed, the large model sends the inference results to the executor, and then proceeds to step six.

[0248] Step Six: After receiving the calling tool address from the large model, the executor initiates calling requests to the optimal master cell judgment tool and the multi-cell joint antenna feeder optimization tool based on the address information, and feeds back the key input parameters of the tool calls, i.e., the execution results (required), to the large model. After completion, proceed to Step Seven.

[0249] Step 7: After the large model obtains the result data from the optimal master cell judgment tool and the multi-cell joint antenna and feeder optimization tool, observe the relevant results, such as:

[0250] The optimal controlling cell determination result is "[cell1_1]".

[0251] The results of the joint antenna and feeder adjustment for multiple cells are as follows:

[0252] Azimuth angle of cell1_1: adjusted from 65 degrees to 120 degrees.

[0253] cell1_1 downtilt angle: adjusted from 8 degrees to 5 degrees.

[0254] The adjusted prediction results for the problem road sections are as follows:

[0255] RSRP = 92dBm (RSRP before adjustment = -105.60dBm)

[0256] SINR = 6dB (SINR before adjustment = 4.98dB)

[0257] After all tools have been used and the data results observed, the optimization plan is summarized and then fed back to the frontline optimization personnel. After this, proceed to step eight.

[0258] Step 8: After receiving the optimization plan from the large model, frontline optimization personnel will manually review the results. If frontline personnel have objections to the cell antenna optimization plan that the large model needs to adjust, and provide feedback to the large model, the main issues are as follows:

[0259] The main control cell antenna feeder being adjusted has an obstruction issue;

[0260] There are coordination issues with the property management of the main control community stations to be adjusted, therefore adjustments to the stations are prohibited.

[0261] The main control cell site being adjusted is a VIP site and is therefore prohibited from being adjusted.

[0262] When the large model receives any of the above objections from frontline personnel, such as "the adjusted main control cell site is a VIP site and adjustment is prohibited", then proceed to step nine.

[0263] If no objections or issues are raised by frontline staff, the process ends, and the output optimization plan serves as the final result to assist frontline staff in implementing the optimization.

[0264] Step Nine: Based on the objections to the antenna and feeder optimization plan reported by frontline personnel in Step Eight (the adjusted main control cell site is a VIP site and adjustment is prohibited), use RAG to initiate interaction with the knowledge base. Following the process in Step Two, obtain the tool capability list, alarm analysis knowledge cases, and coverage optimization knowledge cases from the knowledge base construction. The conclusion is that "the main control cell antenna and feeder optimization adjustment result is unreasonable" and requires re-reasoning. The result is as follows:

[0265] To avoid adjusting the VIP site, cell1_1 needs to be removed from the Top Master Cell Identification List.

[0266] The antenna feeder optimization tool needs to be activated again to generate an antenna feeder adjustment plan, including the use of the "optimal master control cell judgment" and "multi-cell joint antenna feeder optimization" tools.

[0267] After the inference is completed, the large model sends the inference results to the executor and proceeds to step ten.

[0268] Step 10: After receiving the tool call address from the large model, the executor re-initiates call requests to the optimal master cell judgment tool and the multi-cell joint antenna feeder optimization tool based on the address information, and feeds back the key input parameters of the tool call, i.e., the execution result (required), to the large model. After completion, proceed to Step 11.

[0269] Step 11: After the large model obtains the result data from the optimal master cell judgment tool and the multi-cell joint antenna and feeder optimization tool, observe the relevant results, such as:

[0270] The optimal controlling cell determination result is "[cell1_1]".

[0271] The results of the joint antenna and feeder adjustment for multiple cells are as follows:

[0272] Cell2_1 azimuth angle: adjusted from 240 degrees to 310 degrees.

[0273] Cell2_1 downtilt angle: adjusted from 12 degrees to 6 degrees

[0274] The adjusted prediction results for the problem road sections are as follows:

[0275] RSRP = 90dBm (RSRP before adjustment = -105.60dBm)

[0276] SINR = 7dB (SINR before adjustment = 4.98dB)

[0277] After all tools have been used and the data results observed, the optimization plan is summarized and fed back to the frontline optimization personnel. If the frontline personnel have no objections or issues, the process ends, and the output optimization plan serves as the final result to assist the frontline personnel in implementing the optimization. If the frontline personnel still have objections to the cell antenna and feeder optimization plan that needs to be adjusted in the large model, the process jumps to step eight to continue the loop of reasoning and thinking, guiding the frontline personnel to accept the output plan for the large model's antenna and feeder adjustments, until all the main control cells in the "Top Main Control Cell Identification List" have been removed, at which point the loop ends.

[0278] The road network quality optimization method based on the ReAct mechanism proposed in this embodiment uses a multi-round "think-action-observation" closed-loop process to achieve intelligent analysis and precise optimization of poor-quality data by utilizing large models, knowledge bases and toolsets. This effectively improves the speed of problem diagnosis and the scientific nature of the solution. Combined with a dynamic feedback mechanism, it significantly improves the response efficiency and overall service quality of the road network, meeting the needs of fast, accurate and intelligent road network optimization.

[0279] Example 2:

[0280] like Figure 9As shown, this embodiment provides a road network quality optimization device based on the ReAct mechanism of roadside intelligent agents. The device includes:

[0281] Acquisition unit 10 is used to acquire poor quality data in the road grid network;

[0282] The generation unit 20, connected to the acquisition unit 10, is used to process (including reasoning and action) the poor quality data through the road test agent according to the ReAct mechanism, and generate the problem solutions corresponding to the poor quality data.

[0283] The judgment unit 30, connected to the generation unit 20, is used to determine whether the problem solution can meet the preset requirements;

[0284] The reflection unit 40 is connected to the judgment unit 30 and the generation unit 20 respectively. It is used to simulate the defects of the problem solution when the judgment unit determines that the problem solution cannot meet the preset requirements, and trigger the generation unit to regenerate the problem solution corresponding to the poor quality data.

[0285] The determination unit 50 is connected to the judgment unit 30 and the generation unit 20 respectively, and is used to confirm the problem solution as the target solution when the judgment unit determines that the problem solution can meet the preset requirements or when the maximum number of generation times of the generation unit reaches the maximum number of iterations.

[0286] The optimization unit 60, connected to the determination unit 50, is used to optimize the poor quality data according to the target solution, thereby realizing road network quality optimization based on the ReAct mechanism of the road test agent.

[0287] As one specific implementation, the generation unit 20 includes:

[0288] The first processing module is used to understand the poor-quality data through the large language model of the road test agent to obtain the task objective;

[0289] The second processing module, connected to the first processing module, is used to transform the task objective into a road network problem to be processed and input it into the large language model dialog box.

[0290] The third processing module, connected to the second processing module, is used to query and call the target knowledge base through the large language model, and to perform intent recognition and understanding on the road network problem to be processed, so as to obtain the problem target;

[0291] The fourth processing module, connected to the third processing module, is used to call the tool module to process the problem target and obtain the optimization solution;

[0292] The generation module, connected to the fourth processing module, is used to generate solutions to problems corresponding to poor-quality data based on the optimization scheme.

[0293] As a more specific implementation, the fourth processing module includes:

[0294] The first processing submodule is used to identify the problem target, and when the problem target is the main source of the impact of poor quality data, it calls the TOP master cell identification tool to identify one or more master cells that have the most significant impact on road network quality, and obtain the target master cell;

[0295] The second processing submodule, connected to the first processing submodule, is used to call the alarm verification tool of the main control cell to confirm the equipment failure or parameter abnormality of the target main control cell and obtain the alarm cause.

[0296] The third processing submodule, connected to the second processing submodule, is used to call the optimal master cell judgment tool based on the alarm cause to filter out the cells to be optimized.

[0297] The fourth processing submodule, connected to the third processing submodule, is used to call the multi-cell joint antenna feeder optimization tool to design antenna feeder adjustment schemes for the cells to be optimized and obtain optimization schemes.

[0298] The apparatus in this embodiment is capable of performing the method in Embodiment 1.

[0299] Example 3:

[0300] like Figure 10 As shown, this embodiment provides an electronic device, which includes a memory 200 and a processor 100. The memory 200 stores a computer program. When the processor 100 runs the computer program stored in the memory 200, the processor 100 executes the road network quality optimization method based on the ReAct mechanism of the road test agent as described in Embodiment 1.

[0301] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for optimizing road network quality based on a road testing intelligent agent (ReAct) mechanism, characterized in that, The method comprises the following steps: Step S1: obtaining quality difference data in a road grid network; Step S2: processing the quality difference data according to a ReAct mechanism of a road test agent to generate a problem solution corresponding to the quality difference data; Step S3: determining whether the problem solution meets preset requirements: If the problem solution does not meet the preset requirements, simulating reflection on defects of the problem solution, and repeating steps S2-S3 according to the defects; If the maximum number of iterations is reached or the problem solution meets the preset requirements, confirming the problem solution as a target solution, and entering step S4; Step S4: optimizing the quality difference data according to the target solution, thereby realizing road network quality optimization based on the ReAct mechanism of the road test agent.

2. The road network quality optimization method based on the ReAct mechanism of the road test agent according to claim 1, wherein the quality difference data in the road grid network is obtained through virtual road testing.

3. The road network quality optimization method based on the ReAct mechanism of the road test agent according to claim 1, wherein the step S2 specifically comprises: understanding the quality difference data through a large language model of the road test agent to obtain a task target; converting the task target into a to-be-processed road network problem and inputting it into a large language model dialogue box; querying and calling a target knowledge base through the large language model, and performing intent recognition and understanding on the to-be-processed road network problem to obtain a problem target; calling a tool module to process the problem target to obtain an optimization scheme; generating a problem solution corresponding to the quality difference data according to the optimization scheme.

4. The road network quality optimization method based on the ReAct mechanism of the road test agent according to claim 3, wherein before the querying and calling of the target knowledge base through the large language model, the method further comprises constructing a target knowledge base; the constructing of the target knowledge base specifically comprises the following steps: determining an application scenario, wherein the application scenario comprises road network quality; collecting guide manuals, analysis processes, coverage optimization processes, alarm analysis processes, coverage optimization knowledge cases, and alarm analysis knowledge cases related to the application scenario; converting the guide manuals, the analysis processes, the coverage optimization processes, the alarm analysis processes, the coverage optimization knowledge cases, and the alarm analysis knowledge cases into structured or unstructured knowledge forms to form the target knowledge base.

5. The road network quality optimization method based on the ReAct mechanism of the road test agent according to claim 3, wherein before the calling of the tool module, the method further comprises constructing a target knowledge base; initiating interaction with the knowledge base through a RAG to obtain a tool capability list.

6. The road network quality optimization method based on the ReAct mechanism of the road test agent according to any one of claims 3-5, wherein ​ ​ ​ ​ The calling tool module processes the problem target to obtain an optimization scheme, and specifically includes the following steps: Step A1: confirming the problem target, and calling a TOP master cell identification tool to identify one or more master cells with the most significant impact on the road network quality when the problem target is the main source of the quality difference data, to obtain target master cells; Step A2: calling a master cell alarm checking tool to confirm device faults or parameter abnormalities of the target master cells, to obtain alarm causes; Step A3: based on the alarm causes, calling an optimal master cell judgment tool to screen to obtain cells to be optimized; Step A4: calling a multi-cell joint antenna and feeder optimization tool to design an antenna and feeder adjustment scheme for the cells to be optimized, to obtain an optimization scheme.

7. A device for optimizing road network quality based on a road testing intelligent agent (ReAct) mechanism, characterized in that, Comprise: An acquisition unit configured to acquire quality difference data in a road grid network; A generation unit connected with the acquisition unit and configured to process the quality difference data according to a ReAct mechanism of a road test agent to generate a problem solving scheme corresponding to the quality difference data; A determination unit connected with the generation unit and configured to determine whether the problem solving scheme meets a preset requirement; A reflection unit connected with the determination unit and the generation unit respectively, and configured to simulate defects of the problem solving scheme and trigger the generation unit to re-generate the problem solving scheme corresponding to the quality difference data when the determination unit determines that the problem solving scheme does not meet the preset requirement; A determination unit connected with the determination unit and the generation unit respectively, and configured to determine that the problem solving scheme is a target solving scheme when the determination unit determines that the problem solving scheme meets the preset requirement or when a maximum generation number of the generation unit reaches a maximum iteration number; An optimization unit connected with the determination unit and configured to optimize the quality difference data according to the target solving scheme, thereby realizing road network quality optimization based on the ReAct mechanism of the road test agent.

8. The road network quality optimization device based on the ReAct mechanism of the road test agent according to claim 7, wherein The generation unit comprises: A first processing module configured to understand the quality difference data by a large language model of a road test agent to obtain a task target; A second processing module connected with the first processing module and configured to convert the task target into a road network problem to be processed and input into a large language model dialogue box; A third processing module connected with the second processing module and configured to query and call a target knowledge base by the large language model and perform intent recognition and understanding on the road network problem to be processed to obtain a problem target; A fourth processing module connected with the third processing module and configured to call a tool module to process the problem target to obtain an optimization scheme; A generation module connected with the fourth processing module and configured to generate a problem solving scheme corresponding to the quality difference data according to the optimization scheme.

9. The road network quality optimization device based on the ReAct mechanism of the road test agent according to claim 8, wherein The fourth processing module comprises: A first processing submodule configured to confirm a problem target, and when the problem target is a main influence source of the quality difference data, call a TOP master cell identification tool to identify one or more master cells that have the most significant influence on road network quality, and obtain a target master cell; A second processing submodule connected with the first processing submodule and configured to call a master cell alarm checking tool to confirm a device fault or parameter anomaly of the target master cell, and obtain an alarm cause; A third processing submodule connected with the second processing submodule and configured to call an optimal master cell judgment tool based on the alarm cause, and screen to obtain a to-be-optimized cell; A fourth processing submodule connected with the third processing submodule and configured to call a multi-cell joint antenna feeder optimization tool to design an antenna feeder adjustment scheme for the to-be-optimized cell, and obtain an optimization scheme.

10. An electronic device, comprising: The computer program product comprises a memory and a processor, and the memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes the road test intelligent entity ReAct mechanism-based road network quality optimization method according to any one of claims 1 to 6.

Images