Visual network optimization AI assistant implementation method and device

Abstract

The invention relates to the technical field of artificial intelligence, and particularly provides a visual network optimization AI assistant implementation method and device, based on Transform and LLM, the method comprises the following steps: S1, a database modeling module carries out dimension definition, index definition and term library construction; s2, the voice interaction module selects a basic model to carry out data fine tuning and keyword enhancement; s3, the LLM-SQL generation module uses a GPT-4 or an open source LLM as a base; s4, the SQL correction module carries out logic verification and term replacement; s5, establishing a mapping table of aliases and actual fields by a field mapping module, and expanding dynamic fields; and S6, performing chart engine and geographical rendering by a multi-modal presentation module. Compared with the prior art, the method has the advantages that the development cost can be remarkably reduced, the user query flexibility and the data presentation intuition are improved, and the network optimization decision is effectively supported.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and specifically provides a method and device for implementing a visual network optimization AI assistant. Background Art

[0002] Large language models are deep learning models trained using large amounts of text data, enabling them to generate natural language text or understand the meaning of text. These models can be trained on massive datasets to provide in-depth knowledge and language production across a wide range of topics. Their core concept is to learn the patterns and structure of natural language through large-scale unsupervised training, simulating, to a certain extent, the human language cognition and generation process.

[0003] The Transformer model architecture, proposed by Google in their 2017 paper "Attention is All You Need," replaces the RNN architecture commonly used in NLP tasks with a self-attention structure. Its biggest advantage over the RNN architecture is its parallel computing capabilities.

[0004] The current communication network optimization field has the following pain points:

[0005] (1) A query interface needs to be developed independently for each indicator (such as cell traffic and base station load), resulting in high system maintenance costs. Uncommon indicators (such as sudden fault correlation analysis) require manual SQL writing, resulting in a long response cycle. Results are usually presented in tables, lacking the integration of geographic information and trend analysis.

[0006] (2) LLM cannot obtain metadata such as table structure and field constraints, and is prone to generating invalid SQL (such as referencing undefined fields). Natural language ambiguity causes different SQL to be generated multiple times for the same question, and specific terms in the communications field are easily disassembled or incorrectly associated. Summary of the Invention

[0007] The present invention aims to address the deficiencies of the above-mentioned prior art and provides a practical method for implementing a visual network optimization AI assistant.

[0008] A further technical task of the present invention is to provide a reasonably designed, safe and applicable visual network optimization AI assistant implementation device.

[0009] The technical solution adopted by the present invention to solve its technical problem is:

[0010] A method for implementing a visual network optimization AI assistant, based on Transformer and LLM, has the following steps:

[0011] S1, the database modeling module defines dimensions, indicators and constructs a term base;

[0012] S2. The voice interaction module selects the basic model and performs data fine-tuning and keyword enhancement;

[0013] S3 and LLM-SQL generation modules use GPT-4 or open source LLM as the base;

[0014] S4, SQL correction module performs logic verification and term replacement;

[0015] S5. The field mapping module establishes a mapping table between aliases and actual fields to expand dynamic fields.

[0016] S6, the multimodal presentation module performs chart engine and geographic rendering.

[0017] Furthermore, in step S1, when dimension definition is performed, the database table fields are mapped to business terms;

[0018] When defining the above indicators, cross-table aggregation calculation is performed;

[0019] The term base construction is to abstract the filtering conditions into natural language descriptions.

[0020] Furthermore, in step S2, the model SenseVoiceSmall is selected, the attention mechanism is used to weight the base station and cell nouns, and the communication field speech dataset is used in the fine-tuning stage.

[0021] Furthermore, in step S4, the term replacement replaces high load with traffic>100;

[0022] Contains a library of prompt word constraints that enforce LLM output to conform to at least one of the following requirements:

[0023] (1) Use only the fields declared in the Schema;

[0024] (2) Aliases must be enclosed in underscores;

[0025] (3) Prefer using CTE instead of nested subqueries.

[0026] Furthermore, in step S5, when the dynamic field is expanded, when the new field 5G_number of users is added, it is registered in the modeling module without modifying the code.

[0027] Furthermore, in step S6, a line chart and a heat map are generated based on ECharts as a chart engine, and the base station layer is superimposed through OpenLayers. The base station icon is clicked to view the real-time indicators, the coverage radius and signal strength hotspots are drawn, and the KML file is exported for use by third-party tools.

[0028] Furthermore, in the multimodal presentation module, if the result contains latitude and longitude fields, OpenLayers is called to render a geographic distribution map; if it is a pure numerical value, a histogram is generated and the top few abnormal cells are marked.

[0029] A visual network optimization AI assistant implementation device, comprising: at least one memory and at least one processor;

[0030] The at least one memory is configured to store a machine-readable program;

[0031] The at least one processor is used to call the machine-readable program to execute a method for implementing a visual network optimization AI assistant.

[0032] Compared with the prior art, the method and device for implementing a visual network optimization AI assistant of the present invention have the following outstanding beneficial effects:

[0033] The present invention significantly reduces development costs, improves user query flexibility and data presentation intuitiveness, and effectively supports network optimization decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0035] Attachment Figure 1 This is an architectural diagram of a visual network optimization AI assistant implementation method;

[0036] Attachment Figure 2 This is a schematic diagram of the SQL generation and correction process in the implementation method of a visual network optimization AI assistant. DETAILED DESCRIPTION

[0037] In order to enable those skilled in the art to better understand the solutions of the present invention, the present invention will be further described in detail below in conjunction with specific embodiments. Obviously, the embodiments described are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0038] A best embodiment is given below:

[0039] like Figure 1 、 2As shown in FIG, a method for implementing a visual network optimization AI assistant in this embodiment, based on Transformer and LLM, has the following steps:

[0040] S1, the database modeling module defines dimensions, indicators and constructs a term base;

[0041] When defining dimensions, map database table fields (such as cell_id, signal_strength) to business terms (such as "cell ID" and "signal strength").

[0042] When defining the indicator, perform cross-table aggregation calculations (e.g., "average daily traffic" = SUM(traffic) / COUNT(DISTINCT date));

[0043] The term base construction is to abstract the filtering conditions into natural language descriptions (eg, "high load" = traffic > 100 Gbps).

[0044] S2. The voice interaction module selects the basic model and performs data fine-tuning and keyword enhancement;

[0045] Transformer-based speech recognition model, using domain adaptive training:

[0046] Basic model: SenseVoiceSmall;

[0047] Fine-tuning data: Contains 100,000 speech-text pairs of communication-related conversations;

[0048] Keyword enhancement: Increase the attention weight of terms such as "base station", "PCI", and "switching success rate".

[0049] S3 and LLM-SQL generation modules use GPT-4 or open source LLM as the base;

[0050] Input format:

[0051] {

[0052] "query": "Count the number of high-load communities in Chaoyang District in the past week",

[0053] "schema": {

[0054] "dimensions":["cell_name","district"],

[0055] "metrics": ["high_load_count=COUNT_IF(traffic>100)"],

[0056] "terms":{"high load":"traffic>100"}.

[0057] Example output:

[0058] WITH cte AS(

[0059] SELECT_Cell name_,_Area_

[0060] FROM tb_cell

[0061] WHERE_traffic_>100AND_time_BETWEEN'2023-10-01'AND'2023-10-07'

[0062] SELECT_region_,COUNT(_cell name_)AS_number of high-load cells_

[0063] FROM cte

[0064] WHERE_Area_='Chaoyang District'

[0065] GROUP BY _region_

[0066] S4, SQL correction module performs logic verification and term replacement;

[0067] Logical check:

[0068] Complete the missing time range (if "last week" is not specified, the default is the last 7 days);

[0069] Refactor nested subqueries into CTEs to improve readability.

[0070] Term replacement: Replace "high load" with traffic>100.

[0071] Contains a library of prompt word constraints that enforce LLM output to conform to at least one of the following requirements:

[0072] (1) Use only the fields declared in the Schema;

[0073] (2) Aliases must be enclosed in underscores;

[0074] (3) Prefer using CTE instead of nested subqueries.

[0075] S5. The field mapping module establishes a mapping table between aliases and actual fields to expand dynamic fields.

[0076] In dynamic field expansion, when adding the new field 5G_Number of Users, you only need to register it in the modeling module without modifying the code.

[0077] S6, multimodal presentation module performs chart engine and geographic rendering;

[0078] The chart engine generates line charts and heat maps based on ECharts, overlays base station layers through OpenLayers, and allows users to click on base station icons to view real-time indicators, draw coverage radius and signal strength hotspots, and export KML files for use by third-party tools.

[0079] In the multimodal presentation module, if the result contains latitude and longitude fields, OpenLayers is called to render a geographic distribution map; if it is a pure numerical value, a histogram is generated and the top few abnormal cells are marked.

[0080] For example:

[0081] Administrators can complete the following operations through the visual interface:

[0082] Import database table structure and automatically parse field types;

[0083] Define indicator formulas (e.g., "user penetration rate" = number of 5G users / total number of users * 100);

[0084] Configure term mapping (e.g., "late busy hours" = 18:00-22:00).

[0085] User voice input: "Help me find the PCI conflicting cells in Haidian District during last night's rush hour."

[0086] System output enhanced text:

[0087] [Region = Haidian District][Time = 2023-10-10 18:00 to 22:00][Problem Type = PCI Conflict];

[0088] Input prompt word example:

[0089] You are a communications network expert and need to generate SQL based on the following schema:

[0090] -Table name: tb_pci_conflict

[0091] -Dimension: cell_name,district

[0092] -Indicator: conflict_count=COUNT(*)

[0093] -Term: PCI Conflict = pci_status = 'CONFLICT'

[0094] User requirement: Count the number of PCI conflicting cells in Haidian District during the evening rush hour last night.

[0095] LLM output:

[0096] SELECT community_name, COUNT(*) AS conflict_count

[0097] FROM tb_pci_conflict

[0098] WHERE area = 'Haidian District' AND time BETWEEN '2023-10-10 18:00' AND '2023-10-10 22:00'

[0099] AND pci_status = 'CONFLICT'

[0100] GROUP BY community_name

[0101] Correction operation: Add the WITH cte AS (...) structure to optimize performance, and replace area with the actual field district.

[0102] Final SQL:

[0103] WITH cte AS (

[0104] SELECT cell_name AS community_name, district

[0105] FROM tb_pci_conflict

[0106] WHERE district = 'Haidian District'

[0107] AND time BETWEEN '2023-10-10 18:00' AND '2023-10-10 22:00'

[0108] AND pci_status = 'CONFLICT'

[0109] s SELECT community_name, COUNT(*) AS conflict_count

[0110] FROM cte

[0111] GROUP BY community_name.

[0112] Result type judgment:

[0113] If the result contains longitude and latitude fields, call OpenLayers to render the geographical distribution map;

[0114] If it is pure numerical value, generate a bar chart and mark the top 5 abnormal communities.

[0115] Based on the above method, a visual network optimization AI assistant implementation device in this embodiment includes: at least one memory and at least one processor;

[0116] The at least one memory is configured to store a machine-readable program;

[0117] The at least one processor is used to call the machine-readable program to execute a method for implementing a visual network optimization AI assistant.

[0118] The above-mentioned specific implementation methods are only specific cases of the present invention. The patent protection scope of the present invention includes but is not limited to the above-mentioned specific implementation methods. Any technical solutions that conform to the above-mentioned specific implementation methods of the present invention and any appropriate changes or substitutions made thereto by ordinary technicians in the relevant technical field shall fall within the patent protection scope of the present invention.

[0119] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for implementing a visual network optimization AI assistant, characterized in that: Based on Transformer and LLM, the steps are as follows: S1, the database modeling module defines dimensions, indicators and constructs a term base; S2. The voice interaction module selects the basic model and performs data fine-tuning and keyword enhancement; S3 and LLM-SQL generation modules use GPT-4 or open source LLM as the base; S4, SQL correction module performs logic verification and term replacement; S5. The field mapping module establishes a mapping table between aliases and actual fields to expand dynamic fields. S6, the multimodal presentation module performs chart engine and geographic rendering.

2. A method for implementing a visual network optimization AI assistant according to claim 1, characterized in that: In step S1, when defining dimensions, the database table fields are mapped to business terms; When defining the above indicators, cross-table aggregation calculation is performed; The term base construction is to abstract the filtering conditions into natural language descriptions.

3. The method for implementing a visual network optimization AI assistant according to claim 2, wherein: In step S2, the SenseVoiceSmall model is selected, the attention mechanism is used to weight the base station and cell nouns, and the communication field speech dataset is used in the fine-tuning stage.

4. The method for implementing a visual network optimization AI assistant according to claim 3, wherein: In step S4, the term "high load" is replaced with "traffic>100" in the term replacement; Contains a library of prompt word constraints that enforce LLM output to conform to at least one of the following requirements: (1) Use only the fields declared in the Schema; (2) Aliases must be enclosed in underscores; (3) Prefer using CTE instead of nested subqueries.

5. A method for implementing a visual network optimization AI assistant according to claim 4, characterized in that: In step S5, when the dynamic field is expanded, when the new field 5G_number of users is added, it is registered in the modeling module without modifying the code.

6. A method for implementing a visual network optimization AI assistant according to claim 5, characterized in that: In step S6, a line chart and heat map are generated based on ECharts as a chart engine, and the base station layer is superimposed through OpenLayers. The real-time indicators are viewed by clicking the base station icon, the coverage radius and signal strength hotspots are drawn, and the KML file is exported for use by third-party tools.

7. The method for implementing a visual network optimization AI assistant according to claim 6, wherein: In the multimodal presentation module, if the result contains latitude and longitude fields, OpenLayers is called to render a geographic distribution map; if it is a pure numerical value, a histogram is generated and the top few abnormal cells are marked.

8. A visual network optimization AI assistant implementation device, characterized in that: include: at least one memory and at least one processor; The at least one memory is configured to store a machine-readable program; The at least one processor is configured to call the machine-readable program to execute the method according to any one of claims 1 to 7.

Images