Question answering method and device for communication network optimization, electronic equipment and storage medium

Abstract

The invention provides a question answering method and device for communication network optimization, electronic equipment and a computer readable storage medium, and relates to the technical field of network optimization. The question and answer method comprises the following steps: acquiring a communication network optimization problem, and vectorizing the communication network optimization problem to obtain a communication network optimization problem vector; based on the communication network optimization problem vector, matching similar segmentation text vectors of communication network optimization knowledge in a communication network optimization knowledge vector library; combining the similar segmented text vector and the communication network optimization problem to generate an intention prompt; and inputting the intention prompt into a large language model, and outputting a communication network optimization solution. The problems that the scope of knowledge in the wireless network optimization field is narrow and the answer effect of a large language model in the communication field is poor in the prior art are at least solved. The method is suitable for a communication network intelligent optimization scene.

Description

Technical Field

[0001] The present invention relates to the technical field of network optimization, and in particular to a question-and-answer method, device, electronic device and computer-readable storage medium for communication network optimization. Background Art

[0002] Due to the high complexity of the wireless network environment and the diversity of maintenance personnel, a large amount of manual work is required to carry out daily optimization work. The business requirements that the network needs to support are very high, and there is great pressure to ensure stable network operation and excellent network service quality, which brings huge challenges and pressure to the daily production and operation of the network.

[0003] In order to reduce costs, operators are also actively exploring how to apply large language models to production practices. At present, some operators have successively launched large language models in the communications field. The large language models in the communications field mainly empower intelligent customer service and smart government affairs. Among them, intelligent customer service introduces data and knowledge of the communications industry on the basis of the general large language model, and forms a large language model in the communications field with strong knowledge understanding and generation capabilities through retraining, tuning, and knowledge enhancement.

[0004] The large language model in the field of communications can automatically answer common network professional questions, automatically identify and repair common network faults, and can interact with humans in natural language. At the same time, the large language model in the field of communications can also realize the ability of knowledge precipitation, knowledge management, knowledge extraction and knowledge production, provide employees with services for automatic storage, retrieval and generation of network professional knowledge, greatly improve the quality and efficiency of intellectual intensive work with high complexity in network business lines, and form a human-machine collaborative work mode. Compared with traditional knowledge question-and-answer systems, the large language model in the field of communications can combine contextual information and scenario analysis to improve the rationality of answers, and has the ability to answer more naturally, fluently and accurately. Therefore, the large language model in the field of communications can currently replace knowledge encyclopedias and expert experience bases to improve the accuracy and efficiency of knowledge questions and answers in the field of communications. Users only need to describe the problem in simple language to get accurate, natural and fluent answers.

[0005] However, as network structures become increasingly complex, the difficulty and knowledge required for network operation, maintenance and optimization have increased exponentially. For specific domain knowledge in wireless network optimization, the large language model in the communications field is difficult to give expected results, resulting in the answer effect of the large language model in the communications field being unable to meet the production requirements of the front line. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a question-and-answer method, device, electronic device and computer-readable storage medium for communication network optimization in response to the above-mentioned deficiencies in the prior art. The method can realize accurate and rapid question-and-answer of communication network optimization problems, expand the knowledge of communication network optimization, expand the applicable scenarios of communication network optimization question-and-answer models, reduce the difficulty of network operation and maintenance and optimization for complex network structures, significantly improve the accuracy and efficiency of communication network optimization questions and answers, thereby improving user satisfaction and trust.

[0007] In a first aspect, the present invention provides a question-answering method for communication network optimization, comprising: obtaining a communication network optimization problem, and vectorizing the communication network optimization problem to obtain a communication network optimization problem vector; based on the communication network optimization problem vector, matching similar segmentation text vectors of communication network optimization knowledge in a communication network optimization knowledge vector library; combining the similar segmentation text vectors and the communication network optimization problem to generate an intent prompt; inputting the intent prompt into a large language model, and outputting a communication network optimization answer.

[0008] In a second aspect, the present invention also provides a question-and-answer device for communication network optimization, comprising: an acquisition module, a matching module, a combination module and a generation module, the acquisition module being used to acquire a communication network optimization problem and vectorize the communication network optimization problem to obtain a communication network optimization problem vector, the matching module being connected to the acquisition module and being used to match similar segmented text vectors of communication network optimization knowledge in a communication network optimization knowledge vector library based on the communication network optimization problem vector, the combination module being connected to the matching module and being used to combine similar segmented text vectors and communication network optimization problems to generate intent prompts, the generation module being connected to the combination module and inputting the intent prompts into a large language model to output a generated communication network optimization answer.

[0009] In a third aspect, the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the question-and-answer method for communication network optimization provided in the first aspect above.

[0010] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the question-and-answer method for communication network optimization provided in the first aspect above is implemented.

[0011] The present invention provides a communication network optimization question-and-answer method, device, electronic device, and computer-readable storage medium. Through various communication network structures, communication network problems, and problem solving situations in the existing communication network optimization field, a communication network optimization knowledge vector library is constructed and updated in a timely manner, and then through the collaborative large language model and the communication network optimization knowledge vector library, more accurate and fast communication network optimization answers are provided for users with communication network optimization problems. Therefore, the present invention can realize accurate and fast question-and-answer of communication network optimization problems, expand the knowledge of communication network optimization, expand the applicable scenarios of communication network optimization question-and-answer models, reduce the difficulty of complex network structures for network operation and optimization, and significantly improve the accuracy and efficiency of communication network optimization questions and answers, thereby improving user satisfaction and trust. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A flowchart of a question-and-answer method for communication network optimization according to Embodiment 1 of the present invention;

[0013] Figure 2 This is an example diagram of performing communication network optimization question-answering based on a communication large language model in Embodiment 1 of the present invention;

[0014] Figure 3 This is an example diagram of determining a poor-quality object in Embodiment 1 of the present invention;

[0015] Figure 4 This is an example diagram of determining the root cause of a poor quality object in Embodiment 1 of the present invention;

[0016] Figure 5 This is an example diagram of a virtual azimuth angle of a master cell in a calculation scenario in Embodiment 1 of the present invention;

[0017] Figure 6 This is an example diagram of a virtual azimuth angle of a master cell in another calculation scenario in Embodiment 1 of the present invention;

[0018] Figure 7 This is an example diagram of determining the root cause of a poor-quality surface scene in Embodiment 1 of the present invention;

[0019] Figure 8 This is an example diagram of a PCI collision of a primary neighboring cell in Embodiment 1 of the present invention;

[0020] Fig. 9 This is an example diagram of PCI confusion between neighboring areas in Embodiment 1 of the present invention;

[0021] Fig.10 This is another example diagram of determining the root cause of a poor-quality surface scene in Embodiment 1 of the present invention;

[0022] Fig.11 This is an example diagram of calculating the protection range of a macro base station in Embodiment 1 of the present invention;

[0023] Fig.12 This is an example diagram of the level distribution of indoor users and outdoor users in Embodiment 1 of the present invention;

[0024] Fig.13 This is an example diagram of the level distribution of users near a building in Embodiment 1 of the present invention;

[0025] Fig.14 This is an example diagram of the level distribution of indoor users in embodiment 1 of the invention;

[0026] Fig.15 This is an example diagram of the level distribution of outdoor users in embodiment 1 of the invention;

[0027] Fig.16 This is an example diagram of a decision-making scheme for determining poor-quality objects in Embodiment 1 of the invention;

[0028] Fig.17 This is an example diagram of the processing effect of evaluating a poor-quality object in Embodiment 1 of the invention;

[0029] Fig.18 A flowchart of a question-and-answer method for communication network optimization according to Embodiment 2 of the present invention;

[0030] Fig.19 A flowchart of a question-and-answer method for communication network optimization according to Embodiment 3 of the present invention;

[0031] Fig. 20 This is a structural diagram of a communication network optimized question-answering device according to Embodiment 4 of the present invention. DETAILED DESCRIPTION

[0032] In order 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.

[0033] It should be understood that the specific embodiments and drawings described herein are only used to explain the present invention rather than to limit the present invention.

[0034] It can be understood that, in the absence of conflict, the various embodiments of the present invention and the various features in the embodiments can be combined with each other.

[0035] It can be understood that, for the convenience of description, the drawings of the present invention only show the parts related to the present invention, while the parts irrelevant to the present invention are not shown in the drawings.

[0036] It can be understood that each unit and module involved in the embodiments of the present invention may correspond to only one physical structure, or may be composed of multiple physical structures, or multiple units and modules may be integrated into one physical structure.

[0037] It can be understood that, in the absence of conflict, the functions and steps marked in the flowcharts and block diagrams of the present invention may occur in an order different from that marked in the drawings.

[0038] It is understood that the flowcharts and block diagrams of the present invention illustrate the possible architectures, functions, and operations of the systems, devices, equipment, and methods according to the embodiments of the present invention. Each box in the flowchart or block diagram may represent a unit, module, program segment, or code, which contains executable instructions for implementing the specified functions. Moreover, each box or combination of boxes in the block diagram and flowchart may be implemented by a hardware-based system that implements the specified functions, or by a combination of hardware and computer instructions.

[0039] It can be understood that the units and modules involved in the embodiments of the present invention can be implemented by software or hardware. For example, the units and modules can be located in a processor.

[0040] Embodiment 1:

[0041] like Figure 1 As shown, this embodiment provides a question-answering method for communication network optimization. The question-answering method for communication network optimization includes:

[0042] S101, obtaining a communication network optimization problem, and vectorizing the communication network optimization problem to obtain a communication network optimization problem vector.

[0043] In this embodiment, the communication network includes but is not limited to wireless networks and mobile networks. Communication network optimization problems include but are not limited to: coverage optimization problems, spectrum allocation optimization problems, traffic management and load balancing problems, QoS (quality of service) optimization problems, and user experience optimization problems. The wireless network optimization problem or query of the wireless network optimization engineer or user is obtained, and the wireless network optimization problem or query is vectorized, and the invalid information of the wireless network optimization problem or query is removed to obtain the wireless network optimization problem vector. For example, if the wireless network optimization problem is "How is the wireless network of Changning Road in July?", the communication network optimization problem vector can be obtained as "Wireless indicators of Changning Road in July".

[0044] S102, matching similar segmented text vectors of communication network optimization knowledge in a communication network optimization knowledge vector library based on the communication network optimization problem vector.

[0045] In this embodiment, Figure 2 As shown, the communication large language model in the prior art usually completes the communication network question answering by collaborating with the enterprise professional knowledge file and the LLM (Large Language Model), wherein the enterprise professional knowledge file is Figure 2However, as the communication network structure becomes increasingly complex, the difficulty and knowledge of network operation and maintenance and optimization will increase exponentially. Enterprise professional knowledge documents are not enough to provide professional and complex knowledge in wireless network optimization, which results in the answer effect of the communication large language model not being able to meet the production requirements of the front line.

[0046] Therefore, this embodiment collects various main work contents of wireless network optimization in the existing communication network optimization field, namely various communication network structures, communication network problems and problem solving situations, and focuses various main work contents of wireless network optimization in the existing communication network optimization field on three objects: cells, scenarios and users. A communication network optimization knowledge vector library based on the five aspects of perception, cognition, decision-making, execution and evaluation of the self-intelligent network is constructed and updated in a timely manner around these three objects. Then, combined with the intention understanding, summary generation and text retrieval capabilities of the large language model, the interactively enhanced human-computer collaborative form is provided to the front-line employees of all lines and professions of the network with network indicators, network element failures, user complaint problem inquiries, network problem root cause location, and problem optimization solutions. The interactive answering capabilities make it easy for front-line personnel to quickly and conveniently understand and grasp the network situation at any time, solve the work pain points of slow query of network element indicators and complex and difficult network problem location in daily network operation and maintenance work, and improve the work efficiency of various professional networks.

[0047] Specifically, S102: Based on the communication network optimization problem vector, matching similar segmented text vectors of communication network optimization knowledge in the communication network optimization knowledge vector library includes steps S1021 to S1023:

[0048] S1021, matching the segmented text vector of the communication network optimization knowledge corresponding to the communication network optimization index vector from the communication network optimization knowledge vector library.

[0049] S1022, calculating the similarity between the communication network optimization problem vector and the segmented text vector.

[0050] S1023, sorting similarities, determining that the segmented text vectors corresponding to the first k similarities in the sorting results are similar segmented text vectors of the communication network optimization knowledge, wherein k refers to a preset number of similar segmented text vectors.

[0051] In this embodiment, after vectorizing the communication network optimization problem and obtaining the communication network optimization problem vector, a search is performed in the communication network optimization knowledge vector library based on the communication network optimization problem vector, and the similarity between each communication network optimization knowledge vector in the communication network optimization knowledge vector library and the communication network optimization problem vector is calculated, and then the similarities are sorted to determine the top k communication network optimization knowledge vectors of the sorting results as similar segmentation text vectors.

[0052] Optionally, before S102: matching similar segmented text vectors of communication network optimization knowledge in a communication network optimization knowledge vector library based on the communication network optimization question vector, the communication network optimization question-answering method further includes S105-step S108:

[0053] S105, constructing a communication network expert knowledge base.

[0054] In this embodiment, before constructing the communication network optimization knowledge vector library, it is necessary to first combine the existing communication network optimization functions of the existing communication system to obtain the communication network expert knowledge in the five aspects of perception, cognition, decision-making, execution, and evaluation based on the self-intelligent network, that is, the knowledge of poor quality objects, the root causes of poor quality objects, the decision-making plans for poor quality objects, the processing results of poor quality objects, and the processing effects of poor quality objects, so as to form a communication network expert knowledge base in the five aspects of perception, cognition, decision-making, execution, and evaluation based on the self-intelligent network.

[0055] It should be noted that the existing communication network optimization functions of the existing communication system include but are not limited to: communication network quality perception of cells, network element alarms and user indicator queries, root cause location and analysis of communication network quality cognition, intelligent diagnosis of communication network quality decisions and optimization solution recommendations, automatic adjustment of cell parameters and MassiveMIMO automatic optimization of communication network decision execution, and application generation of result tables for communication network decision evaluation. The communication network expert knowledge base can be a structured document or a database containing an expert system, supporting search and recommendation functions.

[0056] Specifically, S105: constructing a communication network expert knowledge base, including steps S1051 and S1052:

[0057] S1051, constructing knowledge of poor quality objects, root causes of poor quality objects, decision plans for poor quality objects, and processing effects of poor quality objects, wherein the objects include one of the following: a cell, a scene, and a user.

[0058] Specifically, S1051: constructing knowledge of poor quality objects, root causes of poor quality objects, decision plans for poor quality objects, and processing effects of poor quality objects, including: monitoring business performance data, communication coverage data, and user perception data to obtain communication network data, and dividing the communication network data into communication network data of each object; determining the poor quality object according to the communication network data of each object and a preset poor quality threshold; locating the root cause of the communication network data of the poor quality object to determine the root cause of the poor quality object; determining the decision plan for the poor quality object according to a preset mapping relationship between the root cause and the decision plan; processing the root cause of the poor quality object based on the decision plan of the poor quality object, and evaluating the processing effect of the poor quality object, so as to obtain knowledge of the poor quality object, the root cause of the poor quality object, the decision plan for the poor quality object, and the processing effect of the poor quality object.

[0059] In this embodiment, the quality detection and problem identification function of the existing communication system, that is, the communication network quality perception function, is combined to obtain the poor quality object. The quality problem location function of the existing communication system, that is, the communication network quality recognition function, is combined to obtain the root cause of the poor quality object. The decision-making plan of the poor quality object is combined with the optimization plan generation and decision-making function of the existing communication system, that is, the communication network decision-making function, to obtain the decision-making plan of the poor quality object. The processing result of the poor quality object and the processing effect of the poor quality object are obtained in combination with the effect evaluation and verification function of the existing communication system, that is, the communication network decision evaluation. According to the poor quality object, the root cause of the poor quality object, the decision-making plan of the poor quality object, and the processing effect of the poor quality object, the knowledge of the poor quality object, the root cause of the poor quality object, the decision-making plan of the poor quality object, and the processing effect of the poor quality object is formed. This embodiment records and analyzes the information of the poor quality cells, scenes, and users in detail in the existing communication system to build a communication network expert knowledge base, which can improve the identification accuracy of network problems and shorten the fault location time.

[0060] Specifically, the preset poor quality threshold includes a preset service poor quality threshold, a preset coverage poor quality threshold and a preset poor quality user attribute.

[0061] Specifically, based on the communication network data of each object and a preset quality difference threshold, a poor quality object is determined, including: determining whether the object is one of a cell, a scene, and a user; in response to the object being a cell, determining a poor quality cell based on the service performance data of each cell and a preset service quality difference threshold, wherein the service performance data includes access success rate, drop rate, packet loss rate, and average perceived rate; in response to the object being a scene, determining a poor quality scene based on the communication coverage data of each scene and a preset coverage quality difference threshold, wherein the communication coverage data includes a reference signal receiving power of a sampling point; in response to the object being a user, determining a poor quality user based on the user perception data of each user and preset quality difference user attributes, wherein the poor quality user attributes include one of the following: off-network predicted users, complaining users, and off-network mined users.

[0062] In this embodiment, Figure 3 As shown, combined with the existing quality detection and problem identification functions of the existing communication system, poor quality cells, poor quality scenes and poor quality users are obtained. This embodiment constructs a communication network expert knowledge base by locating and analyzing the information of poor quality problems at the three levels of cells, scenes and users, which can quickly identify the level corresponding to the communication optimization problems raised by users, and then quickly realize the matching of communication network optimization knowledge, thereby improving the accuracy and efficiency of communication network optimization.

[0063] To obtain a poor quality cell, specifically: for each cell, the preset service poor quality threshold is used to determine whether the service performance data of the cell meets one of the preset service poor quality thresholds. When the service performance data of the cell meets one of the preset service poor quality thresholds, the cell is determined to be a poor quality cell corresponding to the preset service poor quality threshold. For example, if the service performance data meets the VoLTE (Voice over Long-Term Evolution, Voice over LTE, high-definition call or long-term evolution voice bearer) poor quality threshold, the cell is determined to be a VoLTE poor quality cell. Cell services can be divided into voice services and data services, wherein voice services include 4G (The 4th generation mobile communication technology, fourth generation mobile communication and its technology) voice services and 5G (The 5th generation mobile communication technology, fifth generation mobile communication and its technology) voice services, and data services include 4G data services and 5G data services. Therefore, the poor quality cells include at least one of the following: 4G voice service low perception cells, i.e. VoLTE poor quality cells, 5G voice service low perception cells, i.e. VONR (Voice over New Radio, new air interface voice bearer) poor quality cells, 4G data service poor perception cells, 5G data service poor perception cells and 5G high fallback cells. Correspondingly, the preset service poor quality thresholds include but are not limited to: VoLTE poor quality threshold, VONR poor quality threshold, 4G data poor quality threshold, 5G data poor quality threshold.

[0064] It should be noted that, in this embodiment, the VoLTE poor quality cell, the VONR poor quality cell, the 4G data service perception poor cell, and the 5G data service perception poor cell may be further subdivided into the following situations according to the specific situation of the preset service quality poor threshold:

[0065] ① Definition of a 4G cell with poor voice access quality: The E-RAB (Evolved Radio Access Bearer, Evolved Radio Access Bearer) establishment success rate (QCI1, QoS Class Identifier) ​​and E-RAB establishment success rate (QCI5) in the service performance data are both less than 98%. Definition of a 4G cell with poor voice drop quality: The LTE service drop rate (QCI1) in the service performance data is >1%. Definition of a 4G cell with poor voice uplink packet loss rate: The uplink PDCP SDU packet loss rate (QCI1) in the service performance data is >1%. ④ Definition of a 4G cell with poor voice downlink packet loss rate: The downlink PDCP SDU packet loss rate (QCI1) in the service performance data is >1%.

[0066] ② Definition of a cell with low VoNR access quality: VoNR wireless access success rate in the service performance data is less than 98%. Definition of a cell with high VoNR call drop rate: VoNR voice QoS Flow drop rate in the service performance data is greater than 1%. Definition of a cell with high VoNR packet loss quality: Any of the indicators of VoNR voice uplink PDCP (Packet Data Convergence Protocol) layer user plane packet loss rate and VoNR voice uplink PDCP layer user plane packet loss rate in the service performance data is greater than 1%.

[0067] ③ Definition of 4G low-rate and poor-quality cells: Calculated according to the average perceived rate in the service performance data, the average perceived rate of 20M bandwidth cells is lower than 10Mbps, the average perceived rate of 15M bandwidth cells is lower than 7.5Mbps, the average perceived rate of 10M bandwidth cells is lower than 5Mbps, and the average perceived rate of 5M bandwidth cells is lower than 2Mbps. Definition of 4G low-access and poor-quality cells: Regardless of bandwidth, if the data service wireless access success rate in the service performance data is less than 98%, it is a low-access cell. Among them, the data service wireless access success rate = RRC (Radio Resource Control) establishment success rate × [eRAB (QCI = 6) establishment success times + eRAB (QCI = 8) establishment success times + eRAB (QCI = 9) establishment success times] / [eRAB (QCI = 6) establishment attempts + eRAB (QCI = 8) establishment attempts + eRAB (QCI = 9) establishment attempts]. Definition of 4G high dropped call quality cell: Regardless of bandwidth, a dropped call rate in the service performance data > 2% is a high dropped call cell. Dropped call rate = (total number of UE context releases - normal number of UE context releases) / total number of UE context releases.

[0068] ④5G low-rate and poor-quality cells are defined as follows: A. 100M bandwidth TDD (Time Division Duplexing) cells: the proportion of cells where the average downlink perceived rate of 5G users is <50Mbps. B. FDD (Frequency Division Duplexing) cells: Taking a 40M bandwidth, 4-channel configuration cell as an example, the proportion of cells where the average downlink perceived rate of 5G users is <30Mbps. 5G low-access and poor-quality cells can be divided into two categories: NSA (Non-Standalone) single-mode cells and SA (Standalone) single-mode cells: A. NSA single-mode cells: SgNB addition success rate <98% is a low-access cell; B. SA single-mode cells: wireless access success rate <98% is a low-access cell. 5G cells with high dropped call quality can be divided into two categories: NSA single-mode cells and SA single-mode cells. A. NSA single-mode cells: If the SgNB abnormal release rate is >2%, it is a cell with high dropped call quality. B. SA single-mode cells: If the UE context dropped call rate is >2%, it is a cell with high dropped call quality.

[0069] The scope of investigation for VoLTE poor quality cells is the cells with a total E-RAB establishment request number (QCI1) of ≥100 for 30 consecutive days. The scope of investigation for VONR poor quality cells is the cells with a total VoNR voice QosFlow access request number of ≥100 for 30 consecutive days. The scope of investigation for 4G data service poor perception cells is the cells with a total RRC connection request number of ≥100 for 30 consecutive days, a total UE context release number of ≥100 per month, and a monthly downlink cell average traffic volume of >0 during busy hours. The scope of investigation for 5G data service poor perception cells is the cells with a total access number of ≥100 for 30 consecutive days, a total release number of ≥100 per month, and a monthly downlink cell average traffic volume of >0 during busy hours. The total number of 5G cells is the number of 5G cells that meet the scope of investigation. The statistics of service performance data are the average of the voice busy-hour data of the cell every day for 30 consecutive days.

[0070] The poor quality scene is obtained, specifically: for each scene, the reference signal receiving power of the sampling point of the scene is judged according to the preset coverage poor quality threshold to see whether it meets one of the preset coverage poor quality thresholds. When the preset coverage poor quality threshold of the scene meets one of the preset coverage poor quality thresholds, the scene is determined to be the poor quality scene corresponding to the preset coverage poor quality threshold. The scenes can be divided into road scenes, surface scenes and building scenes. Road scenes refer to traffic arteries. Surface scenes include but are not limited to: scenic spots, campuses, hospitals, residences, transportation hubs, party and government agencies, commercial shopping areas, hotels and hotel scenes. Building scenes include but are not limited to: business office scenes. Traffic arteries include any of the following: high-speed rail, expressway, elevated road, subway, national highway, provincial highway, scenic spot connecting line.

[0071] It should be noted that whether the reference signal received power of the sampling point of the scene satisfies one of the preset coverage quality difference thresholds is determined as follows:

[0072] A. Determine the 4G sampling points in each road section (20 meters) of each road scene: ① Calculate the number of sections with empty sampling points, that is, the number of sections with 0 sampling points in each section every month. ② Calculate the number of sections with sampling points, that is, the number of sections with sampling points > 0 in each section every month. ③ Calculate the number of sections with good coverage in sections with sampling points, that is, for each section with sampling points, determine the number of sections with [RSRP (Reference Signal Receiving Power) >= -112dBm sampling point ratio] that meets the standard. ④ Calculate the number of sections with 4G cells within the range of sections with empty sampling points, that is, the number of sections with cells within the preset range of sections with empty sampling points, where the preset range includes the following: urban area, county town, township, rural area (i.e. 500 meters, 500 meters, 1500 meters, 1500 meters). ⑤ Calculate the proportion of sections with good coverage, which is (the number of sections with good coverage in sections with sampling points + the number of sections with 4G cells in the range of sections with no sampling points) / the total number of sections in the scene. ⑥ Determine whether the road scene meets the preset poor coverage threshold based on the proportion of sections with good coverage in the road scene, that is, determine whether the proportion of sections with good coverage is ≥90%. If it is <90%, the road scene is determined to be a poor quality road scene.

[0073] B. Determine the 5G sampling points in each road section (20 meters) of each road scene: ① Calculate the number of sections with empty sampling points. ② Calculate the number of sections with sampling points. ③ Calculate the number of good and bad sections with sampling points. cover The number of road sections, that is, for each road section with sampling points, the number of road sections that meet the standard [RSRP>=-105dBm sampling point ratio]. ④ Calculate the number of road sections with 5G cells within the range of empty sampling points, that is, the number of road sections with cells within the preset range of empty sampling points. ⑤ Calculate the proportion of sections with good coverage, the proportion of sections with good coverage = (the number of sections with good coverage in the sections with sampling points + the number of sections with 5G cells within the range of empty sampling points) / the total number of sections in the scene. ⑥ Determine whether the road scene meets the preset coverage quality threshold based on the proportion of sections with good coverage in the road scene.

[0074] C. Determine the 4G sampling points in each sub-scenario of each face-type scene: ① Calculate the proportion of sampling points with RSRP ≥ -112 of 4G sampling points in different sub-scenarios, that is, the proportion of 4G sampling points. ② Determine whether the proportion of sampling points with RSRP ≥ -112 of the sub-scenario meets the corresponding coverage quality difference threshold, where the proportion of sampling points with RSRP ≥ -112 of different sub-scenarios corresponds to different coverage quality difference thresholds, as shown in Table 1. The sub-scenario can be divided into 1st, 2nd and 3rd grade scenes. The coverage quality difference threshold corresponding to the 1st grade scene is 99%, the coverage quality difference threshold corresponding to the 2nd grade scene is 98.5%, and the coverage quality difference threshold corresponding to the 3rd grade scene is 98%. When the RSRP of the 4G sampling points of the sub-scenario meets the corresponding coverage quality difference threshold, the sub-scenario is determined to be a poor-quality sub-scenario. ③ If the proportion of 4G sampling points of all sub-scenarios in the face-type scene does not meet the corresponding coverage quality difference threshold proportion, that is, 100%, then the face-type scene is determined not to be a poor-quality scene.

[0075] Table 1 4G coverage quality thresholds for each sub-scenario in the surface scenario

[0076]

[0077]

[0078] D. Judge the 5G sampling points in each sub-scenario of each face-type scene: ① Calculate the proportion of sampling points with RSRP ≥ -105 of 5G sampling points in different sub-scenarios, that is, the proportion of 5G sampling points. ② Determine whether the proportion of sampling points with RSRP ≥ -105 of the sub-scenario meets the corresponding coverage quality difference threshold, where the proportion of sampling points with RSRP ≥ -105 of different sub-scenarios corresponds to different coverage quality difference thresholds. The coverage quality difference threshold corresponding to the 1-level scene is 97%, the coverage quality difference threshold corresponding to the 2-level scene is 96%, and the coverage quality difference threshold corresponding to the 3-level scene is 95%. When the RSRP of the 5G sampling points of the sub-scenario meets the corresponding coverage quality difference threshold, the sub-scenario is determined to be a poor-quality sub-scenario. ③ If the proportion of 5G sampling points of all sub-scenarios in the face-type scene does not meet the corresponding coverage quality difference threshold proportion, that is, 100%, then the face-type scene is determined not to be a poor-quality scene.

[0079] To obtain poor-quality users, specifically: analyze the user perception data of each user to determine whether the user has poor-quality user attributes, among which the preset poor-quality user attributes include off-network predicted users, complaining users, off-network mining users, derogatory users, potential 5G users, and 5G unopened users. User perception data includes user behavior analysis data, user complaint data, user source analysis data, user network perception data, user portrait data, and user package data. Specifically: ① Based on user behavior analysis data and data analysis models, comprehensively judge whether the user's usage frequency, payment record, and service usage satisfaction meet the preset threshold, and predict users who may cancel the service or change operators in the future, that is, off-network predicted users. ② Complaining users refer to users who have made formal complaints about the operator's services or products. Usually, user complaint data is analyzed based on customer service records and complaint management systems to determine the number of complaints submitted by users within a preset time period or the severity of the complaints to determine whether there are complaining users. ③ Off-network mining users refer to users attracted from competitors through market strategies and marketing activities. Based on the data analysis model and user source analysis data, the user sources of other operators are analyzed. If the user source analysis data meets the preset threshold, the user is determined to be a different network mining user. ④ Derogatory users refer to users who have a negative view of the operator's services or products and may have a negative impact on the brand through word-of-mouth communication. Based on the user's network perception data and the thresholds of voice and data service perception indicators under different network standards, a comprehensive analysis is conducted to obtain the derogatory score of each user, and based on the user's derogatory score, it is determined whether the user is a derogatory user. ⑤ 5G potential users refer to those who have not yet used 5G services, but based on the analysis of user behavior patterns, consumption capacity, and technology acceptance factors, they may become 5G service users in the future. Mainly based on the user portrait model and user portrait data, the potential score of each user is analyzed as the basis for judging 5G potential users. ⑥ 5G unopened users refer to users who already have 5G mobile phones or 5G packages, but have not activated or used 5G services for various reasons, such as not understanding how to use them or thinking that the current 4G service is sufficient. Based on the user package data and data analysis model, it is determined whether the user is a 5G unopened user.

[0080] Specifically, the poor quality scene includes one of the following: a poor quality road scene, a poor quality surface scene, and a poor quality building scene.

[0081] Specifically, the communication network data of the poor-quality object is root-caused and the root cause of the poor-quality object is determined, including: for poor-quality cells, the poor-quality cell problem is located based on the service performance data, communication coverage data and user perception data of the poor-quality cell, and the root cause of the poor-quality cell is determined. The poor-quality cell problem includes at least one of the following: fault alarm problem, coverage problem, interference problem, capacity problem, access problem, retention problem, neighboring cell problem, network structure problem and perception problem; for poor-quality users, the root cause of the poor-quality user is determined based on the root cause analysis model and the user perception data and service performance data of the poor-quality user; for poor-quality road scenes, the root cause of the poor-quality road scenes is determined based on the communication The root cause of the poor quality road scene is determined based on the preset mapping relationship between the coverage data and the root cause and the road scene coverage index threshold. For the poor quality surface scene, the root cause of the wireless network anomaly is analyzed on the communication coverage data of the poor quality surface scene to obtain the root cause of the poor quality surface scene. For the poor quality building scene, the building coverage index data of the poor quality building scene is evaluated based on the indoor user recognition model and the communication coverage data of the poor quality building scene, and the root cause of the poor quality building scene is determined based on the building coverage index data and the preset deep coverage threshold. The building coverage index data includes average level, average quality, weak coverage rate, overlapping coverage rate, quality difference rate, signal-to-interference ratio and MOD3 interference rate.

[0082] In this embodiment, Figure 4 As shown, by combining the quality problem location function already available in the existing communication system, the root causes of poor quality cells, poor quality scenes, and poor quality users are obtained.

[0083] Obtain the root cause of poor quality cells, specifically: ① Locate the poor quality problem of the cell based on the service performance data, communication coverage data and user perception data of the poor quality cell. ② If a cell fails and an alarm is issued, indicating that there is a device or function that is not normal, then the poor quality problem of the cell is determined to be a fault alarm problem. By checking the alarm log, the type and time of the fault are determined, and the root cause of the fault alarm problem is output by checking the status of related hardware and software. ③ If the cell signal coverage is insufficient, resulting in users being unable to receive signals or poor signal quality in some areas, then the poor quality problem of the cell is determined to be a coverage problem. By analyzing the signal coverage map, checking the antenna configuration, position and angle, and considering the environmental factors of the terrain and buildings, the root cause of the coverage problem is output. ④ If the cell is subject to external or internal interference, affecting the signal quality and reducing communication performance, then the poor quality problem of the cell is determined to be an interference problem. By using interference detection tools, the interference source is identified, the frequency planning and filter performance are checked, and the root cause of the interference problem is output. ⑤ If the cell cannot carry the communication needs of the current number of users, resulting in congestion or service degradation, the poor cell quality problem is determined to be a capacity problem. By monitoring the cell traffic and the number of users, evaluating resource allocation, considering capacity expansion or load balancing, the root cause of the capacity problem is output. ⑥ If the user fails when trying to connect to the cell and cannot establish communication, the poor cell quality problem is determined to be an access problem. By checking the signal strength and quality of the user equipment, verifying the access control parameters and signaling process of the cell, the root cause of the access problem is output. ⑦ If the user has connected to the cell, but the connection is unstable and frequently disconnected, the poor cell quality problem is determined to be a retention problem. By analyzing the signal stability of the cell, checking the switching and retention mechanism, considering user mobility management, the root cause of the retention problem is output. ⑧ If there is a problem in the coordinated work between the cell and its neighboring cells, affecting the user's switching between cells, the poor cell quality problem is determined to be a neighboring cell problem. By checking the neighboring cell list and switching parameters, analyzing the handover failure case, optimizing the synchronization and cooperation between cells, the root cause of the neighboring cell problem is output. ⑨If the network planning or site construction location is improper, resulting in poor cell performance or uneven resource utilization, the poor cell quality problem is determined to be a network structure problem. By evaluating the network topology, checking the cell layout and frequency reuse plan, optimizing the network structure, and outputting the root cause of the network structure problem. ⑩If the user's subjective evaluation of the service quality is poor, which may involve delay, packet loss, and rate problems, the poor cell quality problem is determined to be a perception problem. Through the network-side performance indicators and perception indicators, the user experience is actively analyzed, and the root cause of the perception problem is output.

[0084] It should be noted that when obtaining the root cause of poor cell quality, the root cause should include the symptoms, possible causes, detection methods and solution strategies of each of the above poor cell quality problems. In addition, the root cause should also include optimization case studies, fault handling procedures and fault prevention measures to help quickly locate and solve the problem.

[0085] Based on the CEI (Consumer Experience Indicator) user analysis function of the existing communication system, the root causes of poor quality objects are obtained, specifically: ① Data collection and integration: Collect data generated by users when using mobile networks, including surfing the Internet, watching videos, instant messaging and voice calls, to obtain user behavior data; collect performance indicators on the network side, such as delay, rate, drop rate, connection establishment time, to obtain network performance data; collect user feedback data, including user satisfaction surveys, complaint records, and online feedback; collect performance data of different services, such as the startup delay and download rate of video playback, to obtain service performance data. ② Data processing and analysis: Remove invalid or erroneous data to ensure data quality; extract key features that affect user perception from raw data; analyze the correlation between different indicators to determine the key factors affecting user experience. ③ Root cause analysis model construction: Use statistical analysis, machine learning or deep learning technology to build a root cause analysis model; use historical data to train the model and optimize model parameters; verify the accuracy and generalization ability of the model through a test set. ④ Based on the root cause analysis model, locate the root causes of user perception data and business performance data of poor quality users to determine the root causes of poor quality users. The root causes of poor quality users include problem description, possible causes, impact scope, and solutions.

[0086] It should be noted that after determining the root cause of poor quality users and generating the knowledge of the root cause of poor quality users, this embodiment classifies the knowledge of the root cause of poor quality users according to the problem type, severity, impact range, etc. In addition, this embodiment also regularly updates the knowledge of the root cause of poor quality users in the communication network expert knowledge base to reflect new user behavior patterns and network changes.

[0087] The root causes of poor quality scenes can be specifically divided into the root causes of poor quality road scenes, the root causes of poor quality surface scenes, and the root causes of poor quality building scenes. Among them, the root causes of poor quality road scenes are mainly based on the free road measurement function of the existing communication system to identify the root causes of poor quality road scenes. The root causes of poor quality building scenes are mainly determined by performing weak coverage diagnosis, no coverage diagnosis, indoor leakage diagnosis, and macro station intrusion diagnosis on poor quality building scenes through the building three-dimensional coverage system. The root causes of poor quality surface scenes are mainly obtained by performing antenna and feed anomaly analysis, neighboring cell anomaly matching analysis, and network structured analysis on the scene master control cell of the poor quality surface scene through the multi-scenario analysis system to determine the root causes of the poor quality surface scene.

[0088] Based on the free road measurement function of the existing communication system, the root causes of poor road quality scenes are identified, specifically: ① Based on the communication coverage data of the poor road scene, the downlink SINR (Signal to Interference plus Noise) of the main service cell associated with the longitude and latitude MR (Measurement Report) is predicted Ratio, signal to interference plus noise ratio), primary service cell TA (timeadvanced, maximum time advance), primary service cell RSRP and neighboring cell RSRP constructed coverage index, and the inter-station spacing of the macro base station coverage range, calculate the proportion of RSRP (reference signal received power) <-100dBm of all poor quality grid MR sampling points that meet the definition of poor quality section, that is, the proportion of weak coverage, the proportion of TA (coverage distance) > 2km of all poor quality grid MR sampling points that meet the definition of poor quality section, that is, the proportion of over-coverage, the proportion of overlapping coverage ≥ 1 of all poor quality grid MR sampling points that meet the definition of poor quality section, that is, the proportion of overlapping coverage, the proportion of SINR (signal to interference and noise ratio) < 0 of all poor quality grid MR sampling points that meet the definition of poor quality section, that is, the proportion of downlink poor quality. ② According to the preset mapping relationship between the root cause and the road scene coverage index threshold as shown in Table 2, determine whether the weak coverage ratio, over-coverage ratio, and overlapping coverage ratio are greater than or equal to the road scene coverage index. ③ If the weak coverage ratio is greater than the weak coverage ratio threshold, the root cause of the poor quality road scene is determined to be poor downlink quality caused by weak coverage; if the overcoverage ratio is greater than or equal to the overcoverage ratio threshold, the root cause of the poor quality linear scene is determined to be poor downlink quality caused by overcoverage; if the overlapping coverage ratio is greater than or equal to the overlapping coverage ratio threshold, the root cause of the poor quality linear scene is determined to be poor downlink quality caused by overlapping coverage; if the downlink poor quality ratio is greater than or equal to the downlink poor quality ratio threshold, and the weak coverage ratio, overcoverage ratio and overlapping coverage are respectively less than the coverage ratio threshold, the overcoverage ratio threshold and the overlapping coverage ratio threshold, then the root cause of the poor quality linear scene is determined to be poor downlink quality. Among them, the preset mapping relationship between the root cause and the road scene coverage index threshold. This embodiment uses the free road measurement function of the existing communication system to monitor road changes in real time, and can quickly identify signal attenuation or packet loss caused by road construction, traffic congestion or high-speed vehicle movement, quickly identify the root cause of the poor quality road scene, improve the accuracy and efficiency of root cause analysis, and thus improve the accuracy of the communication network expert knowledge base.

[0089] Table 2 Preset mapping relationship between root causes and road scene coverage index thresholds

[0090] The root cause of poor road scene quality Road scene coverage index threshold Poor coverage leads to poor downlink quality Weak coverage percentage threshold = 30% Overcoverage leads to poor downlink quality Over coverage threshold = 70% Overlapping coverage leads to poor downlink quality Overlap coverage threshold = 70% Poor downlink quality Downlink poor quality percentage threshold = 70%

[0091] It should be noted that according to the poor-quality problem section, it refers to the proportion of poor-quality problem grids, such as the proportion of the number of poor-quality problem grids to the effective 10-meter section grids within the set section length, that is, the section where the proportion is greater than the corresponding proportion threshold of the set section length range. Among them, the proportion threshold for a 50-meter section = 100%, the proportion threshold for a 100-meter section = 100%, the proportion threshold for a 200-meter section ≥ 80%, and the proportion threshold for a 300-meter section ≥ 60%. In the poor-quality problem section, a poor-quality problem grid refers to a grid in the effective grids where the proportion of sampling points with a fitted predicted SINR < 0 is greater than 50%. Therefore, it is necessary to judge the magnitude of the serving cell level strength of the MR sampling point "sc_rsrp", the coverage distance of the serving cell TA of the MR sampling point "sc_tadv", the proportion of the number of neighboring cells with a level strength difference within 6 dB greater than or equal to 3 (i.e., overlapping coverage ≥ 1) of the MR sampling point, and the proportion of the fitted predicted SINR value of the MR sampling point for all poor-quality problem grids in the poor-quality problem section.

[0092] Calculate the sum of the number of sampling points that meet the condition of MR sampling point RSRP < -100 dBm in each poor-quality problem grid and the total number of MR sampling points of all poor-quality problem grids, the sum of the number of sampling points that meet the condition of MR sampling point TA > 2 km in each poor-quality problem grid and the total number of MR sampling points of all poor-quality problem grids, the sum of the number of sampling points that meet the condition of the number of neighboring cells with a level strength difference within 6 dB greater than or equal to 3 (i.e., overlapping coverage ≥ 1) of the MR sampling point in each poor-quality problem grid and the total number of MR sampling points of all poor-quality problem grids, and the sum of the number of sampling points that meet the condition of MR sampling point SINR < 0 in each poor-quality problem grid and the total number of MR sampling points of all poor-quality problem grids, and obtain the ratio of the proportion of RSRP (reference signal received power) < -100 dBm of the MR sampling points of all poor-quality problem grids that meet the definition of the poor-quality problem section. Among them, TA > 2 km corresponds to sc_tadv ≥ 19 (corresponding to the coverage distance range 2034 < TADV < 2190). Based on the conversion of the coverage distance of the MR sampling point TA, as shown in Table 3:

[0093] Table 3 Conversion of the coverage distance of the MR sampling point TA

[0094]

[0095] Since the depth coverage of buildings is limited by the outdoor reception of GPS (Global Positioning System) signals of 4G / 5G MDT (Minimization Drive Test) data and the lack of indoor wireless signal characteristics and positioning fingerprints, it is impossible to evaluate and analyze the depth coverage of 4G and 5G buildings. Therefore, in order to conduct weak coverage diagnosis, no coverage diagnosis, indoor leakage diagnosis and macro station intrusion diagnosis for poor quality building scenes through the building three-dimensional coverage system and determine the root cause of poor quality building scenes, it is necessary to first implement indoor user identification based on 4G / 5G MR level feature classification through 4G / 5G building coverage intelligent evaluation. Combined with user mobility analysis and user residence analysis, it can support the statistical and geographical presentation of the average level (AvgRsrp), average quality (AvgRsrq), weak coverage rate (PoorCoverageRate), overlapping coverage rate (OverlappingRate), poor quality rate (Rsrq_N12_Rate), signal-to-interference ratio (SINR), and MOD3 interference rate (Mod3_Rate) of building coverage at the application level.

[0096] The dual-peak recognition method of 4G / 5G building coverage intelligent evaluation can be used to realize indoor user identification based on 4G / 5GMR level feature classification. Specifically, indoor (outdoor) users near the edge of the building can be identified based on MDT, and the level feature classification of users in the nearby building cells can be used to realize the indoor user identification model based on 4G / 5G MR level feature classification. Since the signal will have a penetration loss of 6-15db when penetrating the building, there will be a certain difference in the signal strength of the macro cell signal outdoors and indoors. The indoor user identification model will cluster the sampling points within the building range into two peaks according to the signal strength. The low-level peak is used as the indoor sampling interval, and the high-level peak is used as the outdoor sampling point. The outdoor sampling point will be eliminated if it is mistakenly located within the indoor building range due to positioning problems, such as Fig.12 Summarize and analyze the user level and distribution near the building, such as Fig.13As shown, it can be seen that there is a trough in the middle and peaks on both sides. The width and height of the peaks and troughs of different cells are different. In order to eliminate the difference, a series of operations on the data are required. For example: Since the coverage level values ​​of different cells are not equal, in order to complete the data center alignment (the peak and trough levels are close), the original level needs to be translated horizontally, and the center of gravity of the data is taken as the center point, and the original data is translated to ensure that all troughs are located in the same position (zero point). This embodiment analyzes the signal strength and quality at different locations in the house through the building three-dimensional coverage system, and effectively identifies the signal attenuation and insufficient coverage problems caused by building materials, structural design or floor settings. After identifying the problem, a targeted indoor coverage solution is formulated to improve the network experience of indoor users, ensure the network quality of users in the building, improve user satisfaction, and reduce the maintenance cost caused by network quality problems. Improve the accuracy of root cause analysis of poor quality building scenes, thereby improving the accuracy of the communication network expert knowledge base.

[0097] User identification can also be completed through the K-Means clustering algorithm, specifically: a) data frequency aggregation, constructing a one-dimensional vector based on RSRP and summarizing it; b) setting the RSRP value of the center of gravity to zero by searching for the center of gravity; c) data normalization; d) clustering cells with similar characteristics through the K-Means model; c) performing bimodal determination on the clusters, and users that meet the bimodal model are indoor users.

[0098] It should be noted that the K-Means clustering algorithm requires the number of categories to be specified in advance, and then the indoor or outdoor user is judged based on the category results. The level distribution of the clustering of indoor users has a bimodal feature, such as Fig.14 As shown in the figure, the horizontal axis is the level value and the vertical axis is the number of MRs. Users with obvious bimodal characteristics are indoor users. The level distribution of non-indoor users does not have bimodal characteristics, such as Fig.15 shown.

[0099] In addition, this embodiment combines the existing optimization scheme generation and decision-making functions of the existing communication system to obtain the decision-making scheme for the poor quality object, such as Fig.16As shown, specifically: based on the existing cell intelligent monitoring and intelligent troubleshooting suggestions of the communication system, a maintenance plan is generated; based on the existing cell antenna optimization suggestions and cell parameter optimization suggestions of the free measurement system, multi-scenario analysis system and three-dimensional coverage system, an optimization plan is generated; based on the existing new base stations, base station upgrades and cell expansion suggestions of digital delivery applications, NR (New Radio) upgrades and intelligent capacity allocation applications, a construction plan is generated; the maintenance plan, optimization plan and construction plan are summarized to obtain the optimization suggestion plan for the decision-making link, that is, the decision-making plan for the quality difference object, so as to manually or automatically handle and implement the user, cell and scene quality difference problems according to the optimization suggestion plan of the decision-making link in the execution link. Among them, manual processing refers to arranging optimization personnel to make on-site adjustments to the physical downtilt angle and azimuth angle of the antenna feeder and manual modification of important level parameters according to the optimization suggestion plan of the decision-making link. Automatic processing refers to the automatic processing of remote electronic downtilt angle and azimuth angle, addition and deletion of neighboring cells, carrier liscens adjustment, and switching parameters according to the optimization suggestion plan of the decision-making link, without manual intervention.

[0100] This embodiment combines the existing effect evaluation and verification function of the existing communication system to obtain the processing results and processing effects of the poor quality objects, such as Fig.17 As shown, through the 4 / 5G poor quality cell dispatching system, the weekly dispatching of problem sections and the user-level evaluation list, in the evaluation phase, the effect evaluation and verification of the optimization suggestions after execution are carried out for poor quality users, poor quality cells and poor quality scenarios, and the processing results and effects of the poor quality objects are generated.

[0101] Specifically, the wireless network abnormality includes an antenna feed abnormality.

[0102] Specifically, a root cause analysis of wireless network anomalies is performed on the communication coverage data of the poor quality surface scenario to obtain the root cause of the poor quality surface scenario, including: calculating the virtual azimuth of the scene master cell of the poor quality surface scenario according to the communication coverage data of the poor quality surface scenario; calculating the difference between the virtual azimuth and the original azimuth of the scene master cell; the virtual azimuth of the scene master cell is the main lobe, the side lobe and the back lobe are determined, and the sampling grid ratio of the side lobe and the back lobe is calculated; the root cause of the poor quality surface scenario is determined according to the difference between the virtual azimuth and the original azimuth, the sampling grid ratio of the side lobe and the back lobe, the mapping relationship between the root cause and the preset azimuth condition, the preset side lobe ratio and the preset back lobe ratio, wherein the root cause includes forward and reverse connection, reverse connection between two, abnormal azimuth, excessive sidelobe ratio and excessive backlobe ratio.

[0103] In this embodiment, the existing network poor quality surface scenario has problems such as inaccurate engineering parameter data, inaccurate engineering parameter azimuth, abnormal antenna sidelobe and backlobe coverage, and wrong connection of engineering parameters, which brings many problems to network neighboring cell optimization, switching band sorting, and wireless coverage problem analysis, and even easily causes poor indicators and frequent abnormal events. The existing technology uses on-site testing and neighboring cell switching method analysis, which is not only inefficient but also easy to omit. Therefore, this embodiment uses a multi-scenario analysis system to perform antenna feed abnormality analysis on the scene master cell of the poor quality surface scenario to determine the root cause of the poor quality surface scenario, specifically: ① Calculate the virtual azimuth of the scene master cell of the poor quality surface scenario based on the communication coverage data of the poor quality surface scenario. Considering that the cell has sampling distribution in different coverage directions, in order to reduce the problem of sparse sampling in different sub-directions, the most likely direction angle of the cell is found. This embodiment can use the vector method to merge the radian values ​​of each small sector, and finally calculate the virtual direction angle, such as Figure 5 and Figure 6 As shown, specifically: θ is the arc value from the north direction to the small sector, and the sin(θ) and cos(θ) values ​​of each small sector are used as the decomposition vectors of the current small sector on the x and y axes, thereby obtaining X1 and Y2; then the decomposition vectors are accumulated respectively, and the direction angle from the point to (ΣXn', ΣYn') is calculated as the virtual azimuth of the scene master cell. ② After calculating the virtual azimuth of the scene master cell, the virtual azimuth is compared with the original azimuth, and the difference between the virtual azimuth and the original azimuth of the scene master cell is calculated, the ratio of the sampling grids that fall within 60° to 120° on both sides of the main lobe (virtual azimuth) to the total grid, that is, the ratio of the side lobe sampling grid, and the ratio of the sampling grids that fall outside 120° of the main lobe (virtual azimuth) to the total grid, that is, the ratio of the back lobe sampling grid. ③ According to the mapping relationship between the root cause and the preset azimuth angle conditions shown in Table 4 and the mapping relationship between the root cause and the sidelobe / backlobe ratio conditions shown in Table 5, determine the root cause of the poor quality surface scene.

[0104] Table 4 Mapping relationship between root causes and preset azimuth conditions

[0105]

[0106] Table 5 Mapping relationship between root cause and side lobe / back lobe ratio conditions

[0107] Side lobe / back lobe ratio condition Root Cause Sidelobe sampling grid ratio is greater than 15% Side lobe ratio is too high The back lobe sampling grid ratio is greater than 5% Back flap ratio is too high

[0108] Optionally, the communication coverage data also includes MR sampling data of the scene master cell, actual neighboring cell information and neighboring cell switching data.

[0109] Optionally, the wireless network anomaly also includes abnormal neighboring cell matching.

[0110] Optionally, performing wireless network anomaly root cause analysis on the communication coverage data of the poor-quality surface scenario to obtain the root cause of the poor-quality surface scenario also includes: determining the actual neighboring area information of the scene master cell of the poor-quality surface scenario based on the MR sampling data of the scene master cell of the poor-quality surface scenario; judging whether there is a main neighboring cell PCI collision and PCI confusion between neighbors in the scene master cell of the poor-quality surface scenario based on the actual neighboring area information, and determining the necessary and non-essential neighboring areas of the scene master cell of the poor-quality surface scenario; judging the actual neighboring area information based on the scene master cell of the poor-quality surface scenario The actual configured neighboring cell information and neighboring cell switching data of the control cell are used to count the number and performance of neighboring cell switching of the scene main control cell in the poor quality surface scenario; according to the necessary neighboring cells and non-essential neighboring cells, the number and performance of neighboring cell switching are used to determine whether there are missed necessary neighboring cells and redundant non-essential neighboring cells in the scene main control cell; in response to the existence of PCI collision of the main neighboring cell and PCI confusion between neighbors, it is determined that the root cause of the poor quality surface scenario is PCI conflict; in response to the existence of missed necessary neighboring cells and redundant non-essential neighboring cells, it is determined that the root cause of the poor quality surface scenario is missed and redundancy of neighboring cells.

[0111] In this embodiment, the problem of abnormal neighbor matching is one of the topics that scene problem analysis pays more attention to. Coverage and performance degradation problems caused by the lack of neighboring areas, redundancy, and inaccurate neighboring area information often occur. The solution mechanism based on problem triggering in the prior art is often compensated afterwards. Therefore, with the help of big data and AI (Artificial Intelligence) analysis capabilities, it is extremely beneficial for front-line work to discover neighboring area problems earlier and more comprehensively, and give optimization and adjustment suggestions. In this embodiment, the scene main control cell of the poor quality surface scene is analyzed by a multi-scenario analysis system to determine the root cause of the poor quality surface scene, specifically: ① Because the neighboring area reported by the MR is only the neighboring area information found according to the corresponding frequency point, and it is not necessarily the real neighboring area. Therefore, this embodiment comprehensively judges whether it is necessary to add neighboring areas based on the quality and frequency of neighboring areas, combined with whether switching has occurred, the distance between neighbors, and whether multiple places appear. ② If Figure 8As shown, if two or more co-frequency E-UTRAN (Evolved UMTS Terrestrial Radio Access Network) cells using the same PCI (Physical Cell Identifier) ​​are too geographically isolated, the UE (User Equipment) cannot achieve normal signal synchronization and decoding in the signal overlap area of ​​the two cells, that is, a PCI collision occurs. Therefore, if a cell has the same frequency and PCI as a neighboring cell, a PCI collision will occur. Although the LTE configuration rules restrict the configuration of neighboring cells and do not allow the source cell to have the same PCI as the co-frequency neighboring cell, the LTE cell will not have the same frequency and PCI as its neighboring cell. However, the LTE cell may have the same frequency and PCI as its external cell, or multiple local cells may have the same frequency and PCI under one eNodeB. This embodiment determines whether there is a PCI collision by checking whether the main service cell of the same grid has the same frequency and PCI for multiple times. ③ As Fig. 9As shown in the figure, if the RSRP of the serving cell and the measuring cell meet the switching threshold, and the neighboring cells of the measuring cell and the serving cell have the same frequency and the same PCI, the UE will fail to switch and the call will be dropped, that is, PCI confusion occurs. PCI confusion usually occurs in the following two situations: a. The measured cell is a neighboring cell of the serving cell. If the UE does not support the ANR (Automatic Neighbour Relation) capability, after the UE reports the measured Cell B to the eNodeB, the eNodeB cannot distinguish whether the neighboring cell object measured by the UE is Cell B or Cell C, resulting in a switching failure (CellA does not initiate a switching). If the ANR capability is available and the ANR function of the system event is turned on, the UE reports the ECGI (E-UTRAN Cell Global Identity) of Cell B, which is used by the eNodeB to distinguish the target neighboring cell measured by the UE, so as to perform a switching. If there are two or more cells with the same frequency and PCI in the neighboring cell relationship of the cell, the cells with the same frequency and PCI are PCI conflicting cells. b. The measured cell is not a configured neighboring cell of the serving cell. The eNodeB mistakenly believes that the UE has measured the neighboring cell (Cell C) of the serving cell, and thus initiates a handover to the neighboring cell (Cell C). If the current area is not covered by the neighboring cell (Cell C) signal, such as the location of the UE that is actually covered by Cell B, the handover will fail. Since the measured cell eci is backfilled based on the cell with the same frequency and PCI closest to the sampling point, it is impossible to distinguish whether the measured cell is a configured neighboring cell or an unconfigured neighboring cell. However, it is still possible to check the measured cell, that is, whether the eci is the union of the backfilled cells and the configured neighboring cells has the same frequency and PCI.

[0112] In general Figure 7 As shown, based on the MR sampling data of the scene master cell in the poor quality surface scenario, the configured neighboring cells and PM (Performance Management) switching data, the missed configuration, redundant neighboring cells and neighboring cell pairs with frequent PCI conflicts and confusion are output in turn to determine the root cause of the poor quality surface scenario.

[0113] Optionally, the communication coverage data also includes macro station data and MDT data of the scene master cell.

[0114] Optionally, the wireless network anomaly also includes a network structural anomaly.

[0115] Optionally, performing wireless network anomaly root cause analysis on the communication coverage data of the poor quality surface scenario to obtain the root cause of the poor quality surface scenario also includes: calculating the protection range of the macro station based on the macro station data of the poor quality surface scenario; calculating the out-of-region grid ratio of the scene master cell based on the protection range of the macro station and the MDT data of the scene master cell of the poor quality surface scenario; judging whether the out-of-region grid ratio of the scene master cell is greater than a preset out-of-region grid ratio threshold; in response to the out-of-region grid ratio of the scene master cell being greater than the preset out-of-region grid ratio threshold, determining that the root cause of the poor quality surface scenario is out-of-region coverage.

[0116] In this embodiment, a multi-scenario analysis system is used to perform a network structure analysis on the scene master cell of the poor quality surface type scene to determine the root cause of the poor quality surface type scene, mainly analyzing various network structure-related problems, outputting clusters of weak coverage problems and overlapping coverage problems, and outputting a list of problem cells such as over-coverage cells, overlapping coverage cells, and over-close or over-distant coverage cells, so as to realize the analysis of the network structure and output of solutions, such as Fig.10 As shown, specifically: ① Calculate the effective protection range of macro station B by shadow method, such as Fig.11As shown, the effective protection range of the macro station B is the size of the shaded area in the figure, wherein the default configuration ∠abc is 120° and ∠bac is 30°. ②First, find out the macro stations involved within the maximum and minimum center longitude and latitude range of all coverage grids in the cell, that is, the three stations B, B', and B" in the figure. Secondly, determine whether the current grid is within the red dotted line restriction range based on whether ∠bac is less than 30°, and determine whether the current grid is within the blue solid line inclusion range based on whether ∠abc is greater than 120°. Finally, determine whether the current grid is in the shadow area, that is, the effective protection range of macro station B. ③According to the protection range of the macro station and the MDT data of the scene master cell of the poor-quality surface scene, calculate the proportion of out-of-area grids in the scene master cell. ④If the cell has a grid sampling ratio greater than 10% as an effective coverage grid, and the grid ratio exceeds 2 times the average station spacing, that is, the cell with an out-of-area grid ratio of more than 20%, then it is determined that the root cause is too far coverage. Too close coverage refers to the proportion of grids below 1 / 2 of the average station spacing greater than 80%. % of cells. This embodiment uses a multi-scenario analysis system to perform antenna feed anomaly analysis, neighbor cell anomaly matching analysis and network structured analysis on the scene master cell of the poor quality surface type scenario to determine the root cause of the poor quality surface type scenario. Through multiple analysis methods, the poor quality problem can be refined to avoid the limitations of traditional single analysis and improve the accuracy of root cause identification. Combining multiple analysis results, more effective optimization strategies can be formulated, including equipment maintenance, interference optimization, and resource allocation, so as to systematically improve network quality. By solving antenna feeder, neighbor cell and structural problems, the overall stability and service quality of the network are improved, especially in the surface type scenario, which will help users experience under different environmental conditions. Effective root cause analysis and optimization measures will ultimately be directly reflected in user experience, reduce complaints and churn, and improve user satisfaction and loyalty, thereby improving the accuracy of the communication network expert knowledge base.

[0117] It should be noted that the station spacing and station height assessment is to compare the current station with the average station spacing and station height of the 20 nearest stations. If it exceeds 20% of the average value, it is defined as the station spacing is too large and the station height is too high.

[0118] S1052, associating knowledge to form a communication network expert knowledge base.

[0119] In this embodiment, a unified communication network expert knowledge base is formed by acquiring and associating knowledge of poor quality objects, root causes of poor quality objects, decision-making schemes for poor quality objects, and treatment effects of poor quality objects. Considering that the partition architecture can make more reasonable use of storage resources, avoid waste of storage space and idle resources, ensure efficient operation of the storage system, and that the partition storage data backup and recovery can be faster and more targeted, which helps to achieve rapid recovery, this embodiment can also form knowledge bases for poor quality objects, root causes of poor quality objects, decision-making schemes for poor quality objects, and treatment effects of poor quality objects, respectively. The knowledge bases include but are not limited to: constructing a first knowledge base to realize situational awareness of cell-level voice problems, constructing a second knowledge base to realize situational awareness of cell-level data problems, and constructing a third knowledge base to realize situational awareness of scene-level problems.

[0120] S106, based on the Langchain-ChatGLM architecture, loads and reads the communication network optimization knowledge text from the communication network expert knowledge base.

[0121] In this embodiment, ChatGLM is an open source Chinese-English bilingual dialogue model with 6.2 billion parameters, which supports local deployment. Langchain is a framework for LLM application development, including model calling and data source connection functions. The combination of ChatGLM and Langchain, namely the Langchain-ChatGLM architecture, can realize a question-answering system based on a knowledge base, involving model fine-tuning and prompt word engineering technology, which is suitable for vertical field applications and document processing. The Langchain-ChatGLM architecture also provides WebUI and command line tools, supporting access to multiple data formats.

[0122] S107, segmenting the communication network optimization knowledge text to obtain segmented text of the communication network optimization knowledge.

[0123] S108, vectorizing the segmented text of the communication network optimization knowledge to obtain the segmented text vector of the communication network optimization knowledge to form a communication network optimization knowledge vector library.

[0124] In this embodiment, after the communication network optimization knowledge text loaded and read from the communication network expert knowledge base is segmented and vectorized, not only the segmented text vectors obtained after segmentation and vectorization are stored in the communication network optimization knowledge vector library; after obtaining and vectorizing the communication network optimization problem, the communication network optimization problem vector is also stored in the communication network optimization knowledge vector library.

[0125] S103, combining similar segmented text vectors and communication network optimization problems to generate intent prompts.

[0126] Specifically, S103: combining similar segmented text vectors and communication network optimization problems to generate intent prompts, including steps S1031 and S1032:

[0127] S1031, converting the similar segmented text vector into similar segmented text.

[0128] S1032, using similar segmented text as context information for the communication network optimization problem to generate intent prompts.

[0129] In this embodiment, the matched similar segmented text vectors are converted into similar segmented texts through a vector conversion model, and the similar segmented texts are used as the context of the communication network optimization problem to generate an intent prompt. The intent prompt prompt uses examples to allow the large language model to generate analysis conclusions based on the combination of the communication network optimization problem asked and the root cause analysis results, including but not limited to: inductive or segmented forms, for one of the scenarios, cells, and users.

[0130] It should be noted that this embodiment can also vectorize communication network optimization problems and segment text of vectorized communication network optimization knowledge through a vector conversion model, wherein the vector conversion model includes but is not limited to: word embedding models (WordEmbedding Models), sentence embedding models (Sentence Embedding Models), and contextual embedding models (Contextual Embedding Models).

[0131] S104, input the intention prompt into the large language model, and output the communication network optimization solution.

[0132] A communication network optimization question-and-answer method provided in this embodiment builds and timely updates a communication network optimization knowledge vector library through various communication network structures, communication network problems, and problem solving situations in the existing communication network optimization field, and then provides users with communication network optimization problems with more accurate and rapid communication network optimization answers through the collaborative large language model and the communication network optimization knowledge vector library, realizes accurate and rapid question-and-answer of communication network optimization problems, expands the knowledge of communication network optimization, expands the applicable scenarios of communication network optimization question-and-answer models, reduces the difficulty of network operation and maintenance and optimization of complex network structures, and significantly improves the accuracy and efficiency of communication network optimization questions and answers, thereby improving user satisfaction and trust. In addition, by recording and analyzing the information of poor quality cells, scenarios, and users in detail in the existing communication system to build a communication network expert knowledge base, the accuracy of identifying network problems can be improved and the fault location time can be shortened. By building a communication network expert knowledge base based on the information of positioning, analyzing, and processing poor quality problems at the three levels of cells, scenarios, and users, the level corresponding to the communication optimization problem raised by the user can be quickly identified, and then the matching of communication network optimization knowledge can be quickly realized, and the accuracy and efficiency of communication network optimization can be improved. Through the existing communication system's free road measurement function, road changes can be monitored in real time, and signal attenuation or packet loss caused by road construction, traffic congestion or high-speed vehicle movement can be quickly identified, and the root cause of poor quality road scenes can be quickly identified, improving the accuracy and efficiency of root cause analysis, thereby improving the accuracy of the communication network expert knowledge base. Through the building's three-dimensional coverage system, the signal strength and quality at different locations in the house can be analyzed to effectively identify signal attenuation and insufficient coverage caused by building materials, structural design or floor settings. After identifying the problem, a targeted indoor coverage solution is formulated to improve the network experience of indoor users, ensure the network quality of users in the building, improve user satisfaction, and reduce maintenance costs caused by network quality issues. Improve the accuracy of root cause analysis of poor quality building scenes, thereby improving the accuracy of the communication network expert knowledge base. Through the multi-scenario analysis system, the scene master cell of the poor quality surface scenario is analyzed for antenna feed anomaly, neighboring cell anomaly matching analysis and network structure analysis to determine the root cause of the poor quality surface scenario. Through multiple analysis methods, the poor quality problem can be refined to avoid the limitations of traditional single analysis and improve the accuracy of root cause identification. Combining multiple analysis results, more effective optimization strategies can be formulated, including equipment maintenance, interference optimization, and resource allocation, so as to systematically improve network quality. By solving antenna feeder, neighboring cell and structural problems, the overall stability and service quality of the network can be improved, especially in the surface scenario, which will help users experience in different environmental conditions. Effective root cause analysis and optimization measures will ultimately be directly reflected in the user experience, reduce complaints and churn, and improve user satisfaction and loyalty, thereby improving the accuracy of the communication network expert knowledge base.

[0133] Embodiment 2:

[0134] like Fig.18 As shown, this embodiment provides a question-answering method for communication network optimization, and the question-answering method for communication network optimization includes:

[0135] S201, construct a communication network expert knowledge base; based on the Langchain-ChatGLM architecture, load and read the communication network optimization knowledge text from the communication network expert knowledge base, segment the communication network optimization knowledge text, obtain the segmented text of the communication network optimization knowledge, and vectorize the segmented text of the communication network optimization knowledge to obtain the segmented text vector of the communication network optimization knowledge to form a communication network optimization knowledge vector library.

[0136] In this embodiment, the communication network expert knowledge base is Fig.18 The local knowledge base of wireless network experts in the communication network optimization knowledge text is Fig.18 The text in the communication network optimization knowledge segmentation text is Fig.18 The smaller documents in the communication network optimization knowledge vector library are Fig.18 Vector database in .

[0137] S202, obtaining a communication network optimization problem, and vectorizing the communication network optimization problem to obtain a communication network optimization problem vector.

[0138] In this embodiment, the communication network optimization problem is Fig.18 Wireless Network Optimization Engineers in the field.

[0139] S203, based on the communication network optimization problem vector, matching the similar segmentation text vector of the communication network optimization knowledge in the communication network optimization knowledge vector library, converting the similar segmentation text vector into similar segmentation text, using the similar segmentation text as the context information of the communication network optimization problem, and generating an intention prompt.

[0140] In this embodiment, the communication network optimization problem vector is Fig.18 The vectorized results and the communication network optimization knowledge vector library are Fig.18 The vectorized knowledge base of wireless network experts in the communication network optimization knowledge is similar to the segmented text vector Fig.18 The top k most similar to the user's question vector in the similar segmentation text are Fig.18 The matched text in the intention hint is Fig.18 prompt in the .

[0141] S204, input the intention prompt into the large language model, and output the communication network optimization solution.

[0142] In this embodiment, the large language model is Fig.18 In LLM, the communication network optimization solution is Fig.18 Get answers from wireless network experts at .

[0143] The present embodiment provides a question-and-answer method for communication network optimization. Through various communication network structures, communication network problems and problem solving situations in the existing communication network optimization field, a communication network optimization knowledge vector library is constructed and updated in a timely manner. Furthermore, through the coordinated large language model and the communication network optimization knowledge vector library, more accurate and rapid communication network optimization answers are provided to users with communication network optimization problems, thereby achieving accurate and rapid question-and-answer of communication network optimization problems, expanding the knowledge of communication network optimization, expanding the applicable scenarios of communication network optimization question-and-answer models, reducing the difficulty of network operation and maintenance and optimization for complex network structures, and significantly improving the accuracy and efficiency of communication network optimization question-and-answer, thereby improving user satisfaction and trust.

[0144] Embodiment 3:

[0145] like Fig.19 As shown, this embodiment also provides a question-answering method for communication network optimization. The question-answering method for communication network optimization includes:

[0146] S301, from the five levels of perception, cognition, decision-making, execution and evaluation of the wireless network self-intelligent network, obtain the knowledge of poor quality objects, root causes of poor quality objects, decision-making plans for poor quality objects, and processing effects of poor quality objects, where the objects include one of the following: cells, scenarios and users, and associate the knowledge to form a communication network expert knowledge base.

[0147] In this embodiment, the communication network expert knowledge base is Fig.19 The self-intelligence network knowledge base in .

[0148] S302, loading, reading, segmenting and vectorizing the communication network optimization knowledge text in the communication network expert knowledge base to obtain segmented text vectors of the communication network optimization knowledge to form a communication network optimization knowledge vector base.

[0149] In this embodiment, the communication network optimization knowledge text is loaded, read, segmented and vectorized. Fig.19 The result analysis in the communication network optimization knowledge vector library is Fig.19 Vector database in .

[0150] S303, obtain the communication network optimization problem, vectorize the communication network optimization problem, obtain the communication network optimization problem vector, and match the segmented text vector of the communication network optimization knowledge from the communication network optimization knowledge vector library, calculate the similarity between the communication network optimization problem vector and the segmented text vector, sort the similarity, and determine that the segmented text vectors corresponding to the first k similarities in the sorting result are similar segmented text vectors of the communication network optimization knowledge.

[0151] In this embodiment, the communication network optimization problem is Fig.19 Query in vectorization is Fig.19 In the Embedding, the communication network optimization problem vector is Fig.19 The query vector in .

[0152] S304, combining similar segmented text vectors and communication network optimization problems, generating intent prompts, and inputting the intent prompts into the large language model, and outputting communication network optimization solutions.

[0153] In this embodiment, the intention prompt is Fig.19 The prompt word template in the large language model is Fig.19 The communication network optimization solution is Fig.19 The answer in .

[0154] A question-and-answer method for communication network optimization provided in this embodiment builds and timely updates a communication network optimization knowledge vector library through various communication network structures, communication network problems, and problem solving situations in the existing communication network optimization field, and then provides more accurate and fast communication network optimization answers for users with communication network optimization problems through the coordinated large language model and the communication network optimization knowledge vector library, thereby achieving accurate and fast question-and-answer of communication network optimization problems, expanding the knowledge of communication network optimization, expanding the applicable scenarios of communication network optimization question-and-answer models, reducing the difficulty of network operation and maintenance and optimization for complex network structures, and significantly improving the accuracy and efficiency of communication network optimization question-and-answer, thereby improving user satisfaction and trust.

[0155] Embodiment 4:

[0156] like Fig. 20As shown, this embodiment also provides a question-answering device for communication network optimization, including: an acquisition module 41, a matching module 42, a combination module 43 and a generation module 44, the acquisition module 41 is used to acquire a communication network optimization problem, and vectorize the communication network optimization problem to obtain a communication network optimization problem vector, the matching module 42 is connected to the acquisition module 41, and is used to match similar segmented text vectors of communication network optimization knowledge in a communication network optimization knowledge vector library based on the communication network optimization problem vector, the combination module 43 is connected to the matching module 42, and is used to combine similar segmented text vectors and communication network optimization problems to generate intention prompts, and the generation module 44 is connected to the combination module 43, and inputs the intention prompts into a large language model to output a generated communication network optimization answer.

[0157] Optionally, the question-and-answer device for communication network optimization also includes: a construction module 45, a reading module 46, a segmentation module 47 and a vectorization module 48, the construction module 45 is used to construct a communication network expert knowledge base, the reading module 46 is used to load and read the communication network optimization knowledge text from the communication network expert knowledge base based on the Langchain-ChatGLM architecture, the segmentation module 47 is used to segment the communication network optimization knowledge text to obtain the segmented text of the communication network optimization knowledge, and the vectorization module 48 is used to vectorize the segmented text of the communication network optimization knowledge to obtain the segmented text vector of the communication network optimization knowledge to form a communication network optimization knowledge vector library.

[0158] Specifically, the construction module 45 includes: a construction unit 451 and an association unit 452, the construction unit 451 is used to construct knowledge of poor quality objects, root causes of poor quality objects, decision plans for poor quality objects, and processing effects of poor quality objects, wherein the objects include one of the following: cells, scenes, and users, and the association unit 452 is used to associate the knowledge to form a communication network expert knowledge base.

[0159] Specifically, the construction unit 451 includes: a monitoring subunit, a first determining subunit, a second determining subunit, a third determining subunit and an evaluation subunit. The monitoring subunit is used to monitor business performance data, communication coverage data and user perception data, obtain communication network data, and divide the communication network data into communication network data of each object. The first determining subunit is used to determine the poor quality object according to the communication network data of each object and a preset poor quality threshold. The second determining subunit is used to locate the root cause of the communication network data of the poor quality object and determine the root cause of the poor quality object. The third determining subunit is used to determine the decision plan of the poor quality object according to the preset mapping relationship between the root cause and the decision plan. The evaluation subunit is used to process the root cause of the poor quality object based on the decision plan of the poor quality object, and evaluate the processing effect of the poor quality object, so as to obtain knowledge of the poor quality object, the root cause of the poor quality object, the decision plan of the poor quality object, and the processing effect of the poor quality object.

[0160] Specifically, the first determination subunit includes: a small and micro unit for judging, a first determination small and micro unit, a second determination small and micro unit and a third determination small and micro unit. The small and micro unit for judging is used to judge whether the object is one of a cell, a scene and a user. The first determination small and micro unit is used to respond to the object being a cell, and determine the poor quality cell according to the service performance data of each cell and a preset service quality difference threshold, wherein the service performance data includes access success rate, drop rate, packet loss rate, and average perceived rate. The second determination small and micro unit is used to respond to the object being a scene, and determine the poor quality scene according to the communication coverage data of each scene and a preset coverage quality difference threshold, wherein the communication coverage data includes the reference signal receiving power of the sampling point. The third determination small and micro unit is used to respond to the object being a user, and determine the poor quality user according to the user perception data of each user and a preset quality difference user attribute, wherein the quality difference user attribute includes one of the following: off-network predicted user, complaint user, and cross-network mined user.

[0161] Specifically, the second determination subunit includes: a fourth determination micro unit, a fifth determination micro unit, a sixth determination micro unit, a seventh determination micro unit and an eighth determination micro unit. The fourth determination micro unit is used to locate the poor quality problem of the poor quality cell based on the service performance data, communication coverage data and user perception data of the poor quality cell, and determine the root cause of the poor quality cell. The poor quality problem of the cell includes at least one of the following: fault alarm problem, coverage problem, interference problem, capacity problem, access problem, retention problem, neighboring area problem, network structure problem and perception problem. The fifth determination micro unit is used to determine the root cause of the poor quality user according to the root cause analysis model and the user perception data and service performance data of the poor quality user. The sixth determination micro unit is used to determine the root cause of the poor quality user for the poor quality user. For poor road scenarios, the root cause of the poor road scenarios is determined according to the communication coverage data of the poor road scenarios and the preset mapping relationship between the root cause and the road scenario coverage index threshold. Seventh, small and micro units are determined for the poor surface scenarios. The communication coverage data of the poor surface scenarios are analyzed for the root cause of wireless network anomalies to obtain the root cause of the poor surface scenarios. Eighth, small and micro units are determined for the poor building scenarios. Based on the indoor user recognition model and the communication coverage data of the poor building scenarios, the building coverage index data of the poor building scenarios are evaluated. The root cause of the poor building scenarios is determined according to the building coverage index data and the preset deep coverage threshold. The building coverage index data includes average level, average quality, weak coverage rate, overlapping coverage rate, quality difference rate, signal-to-interference ratio and MOD3 interference rate.

[0162] Specifically, the seventh determination micro unit includes: a first calculation minimum unit, a second calculation minimum unit, a first determination minimum unit and a second determination minimum unit. The first calculation minimum unit is used to calculate the virtual azimuth of the scene main control cell of the poor quality surface scene according to the communication coverage data of the poor quality surface scene. The second calculation minimum unit is used to calculate the difference between the virtual azimuth and the original azimuth of the scene main control cell. The first determination minimum unit is used to determine the side lobe and the back lobe with the virtual azimuth of the scene main control cell as the main lobe, and calculate the sampling grid ratio of the side lobe and the back lobe. The second determination minimum unit is used to determine the root cause of the poor quality surface scene according to the difference between the virtual azimuth and the original azimuth, the sampling grid ratio of the side lobe and the back lobe, the mapping relationship between the root cause and the preset azimuth condition, the preset side lobe ratio and the preset back lobe ratio, wherein the root cause includes forward and reverse connection, reverse connection between two, abnormal azimuth, too high side lobe ratio and too high back lobe ratio.

[0163] Optionally, the seventh determination micro unit also includes: a third determination minimum unit, a fourth determination minimum unit, a statistical minimum unit, a first judgment minimum unit, a fifth determination minimum unit and a sixth determination minimum unit, the third determination minimum unit being used to determine the actual neighboring area information of the scene master cell of the quality-poor surface scene according to the MR sampling data of the scene master cell of the quality-poor surface scene, the fourth determination minimum unit being used to judge whether there is a main neighboring cell PCI collision, PCI confusion between neighbors in the scene master cell of the quality-poor surface scene according to the actual neighboring area information, and to determine the necessary neighboring areas and non-essential neighboring areas of the scene master cell of the quality-poor surface scene, the statistical minimum unit being used to determine the actual neighboring area information of the scene master cell of the quality-poor surface scene according to the actual neighboring area information. The actual configured neighboring cell information and neighboring cell switching data of the scene master cell of the face-type scenario, the number of neighboring cell switching and the performance of the scene master cell of the poor quality face-type scenario are counted, the first minimum judgment unit is used to judge whether there are missed necessary neighboring cells and redundant non-essential neighboring cells in the scene master cell according to the necessary neighboring cells and non-essential neighboring cells, the number of neighboring cell switching and the performance, the fifth minimum determination unit is used to determine that the root cause of the poor quality face-type scenario is the PCI conflict in response to the existence of the main neighboring cell PCI collision and the PCI confusion between the neighbors, the sixth minimum determination unit is used to determine that the root cause of the poor quality face-type scenario is the missed and redundancy of the neighboring cells in response to the existence of the missed necessary neighboring cells and the redundant non-essential neighboring cells.

[0164] Optionally, the seventh determination micro unit also includes: a third calculation minimum unit, a fourth calculation minimum unit, a second judgment minimum unit and a seventh determination minimum unit, the third calculation minimum unit being used to calculate the protection range of the macro station according to the macro station data of the quality-poor surface scenario, the fourth calculation minimum unit being used to calculate the cross-zone grid ratio of the scene master cell according to the protection range of the macro station and the MDT data of the scene master cell of the quality-poor surface scenario, the second judgment minimum unit being used to judge whether the cross-zone grid ratio of the scene master cell is greater than a preset cross-zone grid ratio threshold, and the seventh determination minimum unit being used to determine that the root cause of the quality-poor surface scenario is cross-zone coverage in response to the cross-zone grid ratio of the scene master cell being greater than a preset cross-zone grid ratio threshold.

[0165] Specifically, the matching module 42 includes: a matching unit 421, a calculation unit 422 and a sorting unit 423, the matching unit 421 is used to match the segmented text vector of the communication network optimization knowledge corresponding to the communication network optimization index vector from the communication network optimization knowledge vector library, the calculation unit 422 is used to calculate the similarity between the communication network optimization problem vector and the segmented text vector, and the sorting unit 423 is used to sort the similarity and determine that the segmented text vectors corresponding to the first k similarities in the sorting result are similar segmented text vectors of the communication network optimization knowledge, wherein k refers to the preset number of the similar segmented text vectors.

[0166] Specifically, the combination module 43 includes: a conversion unit 431 and a generation unit 432. The conversion unit 431 is used to convert the similar segmented text vector into the similar segmented text, and the generation unit 432 is used to generate an intention prompt using the similar segmented text as the context information of the communication network optimization problem.

[0167] It can be understood that the above-mentioned question-and-answer device for communication network optimization executes the question-and-answer method for communication network optimization corresponding to the embodiment 1 provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects of the scheme corresponding to the question-and-answer method for communication network optimization in the above embodiment 1, which will not be repeated here.

[0168] Embodiment 5:

[0169] This embodiment also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the question-and-answer method for communication network optimization in the above-mentioned embodiment 1, embodiment 2, or embodiment 3.

[0170] Embodiment 6:

[0171] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the question-and-answer method for communication network optimization in the above-mentioned embodiment 1, embodiment 2 or embodiment 3 is implemented.

[0172] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and 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 question-answering method for communication network optimization, characterized in that: include: Obtain a communication network optimization problem, and vectorize the communication network optimization problem to obtain a communication network optimization problem vector; Based on the communication network optimization problem vector, similar segmented text vectors of the communication network optimization knowledge are matched in the communication network optimization knowledge vector library; Combine similar segmented text vectors and communication network optimization problems to generate intent prompts; The intention prompt is input into a large language model and a communication network optimization solution is output.

2. The question-answering method for communication network optimization according to claim 1, characterized in that: Before matching similar segmented text vectors of communication network optimization knowledge in a communication network optimization knowledge vector library based on the communication network optimization problem vector, the method further includes: Build a knowledge base of communication network experts; Based on the Langchain-ChatGLM architecture, the communication network optimization knowledge text is loaded and read from the communication network expert knowledge base; Segmenting the communication network optimization knowledge text to obtain segmented text of the communication network optimization knowledge; The segmented text of the communication network optimization knowledge is vectorized to obtain the segmented text vector of the communication network optimization knowledge to form a communication network optimization knowledge vector library.

3. The question-answering method for communication network optimization according to claim 2, characterized in that: The construction of the communication network expert knowledge base specifically includes: Constructing knowledge of poor quality objects, root causes of poor quality objects, decision-making plans for poor quality objects, and treatment effects of poor quality objects, wherein the objects include one of the following: cells, scenes, and users; The knowledge is associated to form a communication network expert knowledge base.

4. The question-answer method for communication network optimization according to claim 3, characterized in that: The knowledge of constructing poor quality objects, root causes of poor quality objects, decision-making plans for poor quality objects, and treatment effects of poor quality objects specifically includes: Monitor service performance data, communication coverage data and user perception data to obtain communication network data, and divide the communication network data into communication network data of each object; Determine the poor quality objects according to the communication network data of each object and the preset poor quality threshold; Locate the root cause of the communication network data of the poor quality object and determine the root cause of the poor quality object; Determine the decision-making plan for the poor quality object based on the preset mapping relationship between the root cause and the decision-making plan; Based on the decision-making plan for the poor quality object, the root cause of the poor quality object is processed, and the processing effect of the poor quality object is evaluated to obtain knowledge of the poor quality object, the root cause of the poor quality object, the decision-making plan for the poor quality object, and the processing effect of the poor quality object.

5. The question-answer method for communication network optimization according to claim 4, characterized in that: The preset poor quality threshold includes a preset service poor quality threshold, a preset coverage poor quality threshold and a preset poor quality user attribute. The determining of the poor quality objects according to the communication network data of each object and the preset poor quality threshold specifically includes: Determine whether the object is one of a community, a scene, or a user; In response to the object being a cell, determining a poor-quality cell according to service performance data of each cell and a preset poor-quality threshold of the service, wherein the service performance data includes an access success rate, a drop rate, a packet loss rate, and an average perceived rate; In response to the object being a scene, determining a poor quality scene according to communication coverage data of each scene and a preset coverage poor quality threshold, wherein the communication coverage data includes a reference signal received power of a sampling point; In response to the object being a user, poor-quality users are determined according to user perception data of each user and preset poor-quality user attributes, wherein the poor-quality user attributes include one of the following: off-network predicted users, complaining users, and off-network mined users.

6. The question-answer method for communication network optimization according to claim 5, characterized in that: Poor quality scenes include one of the following: poor quality road scenes, poor quality surface scenes, and poor quality building scenes. The root cause location of the communication network data of the poor quality object to determine the root cause of the poor quality object specifically includes: For cells with poor quality, locate the problem of poor cell quality based on the service performance data, communication coverage data and user perception data of the cells with poor quality, and determine the root cause of the poor cell quality. The problem of poor cell quality includes at least one of the following: fault alarm problem, coverage problem, interference problem, capacity problem, access problem, retention problem, neighboring cell problem, network structure problem and perception problem; For poor-quality users, determine the root cause of poor-quality users based on the root cause analysis model and the user perception data and service performance data of poor-quality users; For the poor quality road scene, determine the root cause of the poor quality road scene according to the communication coverage data of the poor quality road scene and the preset mapping relationship between the root cause and the road scene coverage index threshold; For poor quality surface scenarios, the root cause analysis of wireless network anomalies is performed on the communication coverage data of the poor quality surface scenarios to obtain the root cause of the poor quality surface scenarios; For poor quality building scenarios, based on the indoor user identification model and the communication coverage data of poor quality building scenarios, the building coverage index data of the poor quality building scenarios are evaluated, and the root cause of the poor quality building scenarios is determined according to the building coverage index data and the preset depth coverage threshold. The building coverage index data include average level, average quality, weak coverage rate, overlapping coverage rate, poor quality rate, signal-to-interference ratio and MOD3 interference rate.

7. The question-answer method for communication network optimization according to claim 6, characterized in that: Wireless network anomalies include antenna anomalies, The performing of wireless network abnormality root cause analysis on the communication coverage data of the poor quality surface scenario to obtain the root cause of the poor quality surface scenario specifically includes: Calculate the virtual azimuth angle of the scene master cell of the poor quality surface scene according to the communication coverage data of the poor quality surface scene; Calculating the difference between the virtual azimuth and the original azimuth of the scene master cell; Taking the virtual azimuth angle of the scene master cell as the main lobe, determining the side lobe and the back lobe, and calculating the sampling grid ratio of the side lobe and the back lobe; The root cause of the poor quality surface scene is determined based on the difference between the virtual azimuth and the original azimuth, the sampling grid ratio of the side lobe and the back lobe, the mapping relationship between the root cause and the preset azimuth condition, the preset side lobe ratio and the preset back lobe ratio, wherein the root cause includes forward and reverse connection, reverse connection between two, abnormal azimuth, too high side lobe ratio and too high back lobe ratio.

8. The question-answer method for communication network optimization according to claim 6, characterized in that: The communication coverage data also includes MR sampling data of the scene master cell, actual neighboring cell information and neighboring cell switching data. Wireless network anomalies also include abnormal matching of neighboring areas. The performing of wireless network abnormality root cause analysis on the communication coverage data of the poor quality surface type scenario to obtain the root cause of the poor quality surface type scenario also includes: Determine actual neighboring cell information of the scene master cell of the poor-quality surface scene according to the MR sampling data of the scene master cell of the poor-quality surface scene; According to the actual neighboring cell information, determine whether there is a PCI collision between the main neighboring cells and PCI confusion between the neighboring cells in the scene control cell of the poor quality surface type scene, and determine the necessary neighboring cells and non-essential neighboring cells of the scene control cell of the poor quality surface type scene; According to the actually configured neighboring cell information and neighboring cell switching data of the scene master cell of the poor quality surface type scene, the number and performance of neighboring cell switching of the scene master cell of the poor quality surface type scene are counted; According to the necessary and non-essential neighboring cells, the number of neighboring cell switching and the performance, it is determined whether there are any missed necessary neighboring cells and redundant non-essential neighboring cells in the scene master control cell; In response to the presence of a primary neighboring cell PCI collision and a neighboring cell PCI confusion, determining that the root cause of the poor quality surface scenario is a PCI conflict; In response to the existence of missing necessary neighboring areas and redundant non-essential neighboring areas, it is determined that the root cause of the poor quality surface scene is the missing and redundant neighboring areas.

9. The question-answer method for communication network optimization according to claim 6, characterized in that: The communication coverage data also includes macro base station data and MDT data of the scene master cell. Wireless network anomalies also include network structural anomalies. The performing of wireless network abnormality root cause analysis on the communication coverage data of the poor quality surface type scenario to obtain the root cause of the poor quality surface type scenario also includes: Calculate the protection range of the macro base station based on the macro base station data of the poor quality surface scene; According to the protection range of the macro base station and the MDT data of the scene master cell in the poor-quality surface scene, the cross-region grid ratio of the scene master cell is calculated; Determine whether the cross-region grid ratio of the scene master cell is greater than a preset cross-region grid ratio threshold; In response to the cross-region grid ratio of the scene master cell being greater than a preset cross-region grid ratio threshold, it is determined that the root cause of the poor quality surface scene is cross-region coverage.

10. The question-answer method for communication network optimization according to claim 2, characterized in that: The communication network optimization problem vector includes a communication network optimization index vector, The method of matching similar segmented text vectors of communication network optimization knowledge from a communication network optimization knowledge vector library based on the communication network optimization problem vector specifically includes: Matching the segmented text vector of the communication network optimization knowledge corresponding to the communication network optimization index vector from the communication network optimization knowledge vector library; Calculating the similarity between the communication network optimization problem vector and the segmented text vector; The similarities are sorted, and the segmented text vectors corresponding to the first k similarities in the sorted results are determined to be similar segmented text vectors of communication network optimization knowledge, wherein k refers to a preset number of the similar segmented text vectors.

11. The question-answer method for communication network optimization according to claim 1, characterized in that: The combination of similar segmentation text vectors and communication network optimization problems to generate intent prompts specifically includes: Convert similar segmented text vectors into similar segmented texts; Similar segmented texts are used as context information for communication network optimization problems to generate intent prompts.

12. A question-answering device for communication network optimization, characterized in that: include: Get modules, match modules, combine modules and generate modules, The acquisition module is used to acquire the communication network optimization problem and vectorize the communication network optimization problem to obtain the communication network optimization problem vector. The matching module is connected to the acquisition module and is used to match the similar segmented text vector of the communication network optimization knowledge in the communication network optimization knowledge vector library based on the communication network optimization problem vector. The combination module is connected to the matching module and is used to combine the similar segmented text vector and the communication network optimization problem to generate an intention prompt. The generation module is connected to the combination module, inputs the intention prompt into the large language model, and outputs the generated communication network optimization solution.

13. An electronic device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement a question-and-answer method for optimizing a communication network as described in any one of claims 1 to 11.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, a question-and-answer method for optimizing a communication network as described in any one of claims 1 to 11 is implemented.

Images