End-to-end radio monitoring data query method based on large model

Abstract

The invention discloses an end-to-end radio monitoring data query method based on a large model. A LoRA enhanced field large model module, a dynamic extension knowledge base module and a high-performance MCP data interface module are included. The LoRA enhanced field large model module adopts an LLaMA-Factory framework to be combined with a LoRA technology to carry out field adaptation on a 32B Deepseek-r1-distance-qwen large model, so that accurate understanding of radio professional semantics is realized; the dynamic extension knowledge base module constructs a three-level metadata management system and provides structured data semantic support for a large model; and the high-performance MCP data interface module designs a three-layer protocol architecture to realize efficient interaction between a large model and the HBase. The method has the advantages that the natural language query request of the user can be quickly and accurately processed, the efficiency and accuracy of data query are remarkably improved, efficient conversion from natural language input to accurate data query can be achieved, the technical threshold is lowered, and the query performance is improved.

Description

Technical Field

[0001] The present invention relates to the technical field related to radio monitoring data processing, and in particular to an end-to-end radio monitoring data query method based on a large model. Background Art

[0002] In the field of radio monitoring, traditional data query methods rely on manually written complex query statements, exposing multiple technical bottlenecks:

[0003] (1) Significant barriers to professional semantic understanding

[0004] Professional terms specific to the radio field, such as "time domain trend graph" and "spectrum occupancy," have a significant semantic gap with general natural language. Traditional rule engines, constrained by fixed grammatical matching patterns, struggle to accurately capture the domain-specific intent embedded in natural language queries. This results in low semantic parsing accuracy and an inability to effectively understand users' actual query needs.

[0005] (2) Data structure adaptation efficiency needs to be improved

[0006] Monitoring data is typically stored in distributed databases like HBase. The current model requires manually writing customized query logic for specific table structures. This reliance on manual adaptation requires significant manpower and time to adapt to dynamic changes in table structures, resulting in high adaptation costs and a significant impact on the flexibility and responsiveness of data queries.

[0007] (3) Insufficient adaptability to data query scenarios

[0008] As radio monitoring services continue to expand, new query requirements continue to emerge. However, the traditional query interface architecture is rigid and lacks the ability to adapt to diverse query scenarios. Each new query requirement often requires extensive repetitive development, which not only consumes resources but also leads to inefficient system iterations and makes it difficult to meet rapidly changing business needs.

[0009] While existing large-scale model technologies demonstrate strong capabilities in natural language processing, direct application to the specialized field of radio monitoring presents specific technical challenges. For one thing, the complexity of domain terminology prevents large-scale models from achieving sufficient recognition accuracy. Furthermore, the generated query logic often fails to meet the specifications of distributed databases and cannot be effectively executed. Therefore, an end-to-end solution integrating domain model fine-tuning, intelligent metadata management, and intelligent query interfaces is urgently needed to overcome traditional technical bottlenecks and enhance the intelligence and efficiency of radio monitoring data queries. Summary of the Invention

[0010] In order to overcome the above-mentioned deficiencies in the prior art, the present invention provides an end-to-end radio monitoring data query method based on a large model, which can realize efficient conversion of natural language input to accurate data query, reduce technical barriers and improve query performance.

[0011] In order to achieve the above object, the present invention adopts the following technical solutions:

[0012] A large-model-based end-to-end radio monitoring data query method includes a LoRA-enhanced domain large model module, a dynamically extended knowledge base module, and a high-performance MCP data interface module. The LoRA-enhanced domain large model module uses the LLaMA-Factory framework combined with LoRA technology to perform domain adaptation on the 32B Deepseek-r1-distill-qwen large model, achieving accurate understanding of radio professional semantics. The dynamically extended knowledge base module constructs a three-level metadata management system to provide structured data semantic support for the large model. The high-performance MCP data interface module designs a three-layer protocol architecture to achieve efficient interaction between the large model and HBase. The specific operation steps are as follows:

[0013] (1) User question input: The user submits a query in natural language;

[0014] (2) Intent classification and term analysis: The LoRA enhanced domain model module determines that the problem is a data query problem, and uses the mapping layer to parse the natural language terms into database fields and associate them with the corresponding data tables;

[0015] (3) Query logic generation: The LoRA enhanced domain model module combines the metadata of the dynamically extended knowledge base module to generate query parameters that comply with HBase specifications, including table name, target field, and filter conditions;

[0016] (4) Data query and result return: The high-performance MCP data interface module calls the HBase interface to execute the query, and returns the result after the optimization layer reduces invalid data scanning; the LoRA enhanced domain large model module organizes the data into natural language reports or structured tables and feeds them back to the user.

[0017] The present invention provides a method for intelligent query of radio monitoring data based on a LoRA enhanced large model, which realizes natural language intelligent query of radio monitoring data. The method first uses LoRA technology to perform lightweight fine-tuning on the large model so that it can understand the semantics of the radio professional field; then, a dynamically extended knowledge base is constructed to provide semantic support of dynamically updateable structured data for the large model to ensure the accuracy of the query logic; finally, a high-performance MCP data interface is designed to realize efficient interaction between the large model and the HBase database. Experimental verification shows that the method performs well in radio monitoring data query, can quickly and accurately process users' natural language query requests, significantly improves the efficiency and accuracy of data query, can realize efficient conversion from natural language input to precise data query, lowers the technical threshold and improves query performance. The present invention can be widely used in data query and analysis in the field of radio monitoring, and provide strong technical support for radio management and decision-making.

[0018] Preferably, the specific architecture of the LoRA enhanced domain large model module is as follows:

[0019] (1) Lightweight fine-tuning architecture: insert a low-rank adapter into the Transformer layer, freeze more than 98% of the original model parameters, and only train 2 million adapter parameters;

[0020] (2) Dual-task learning mechanism: Simultaneously train the "query intent classification" and "query logic generation" tasks. The former determines whether the user's question involves data query with an accuracy rate of ≥95%, and the latter generates query parameters that conform to the HBase structure based on the knowledge base metadata.

[0021] (3) Domain knowledge injection: Through the term mapping table and query logic template library, guide the model to generate standardized query logic.

[0022] Preferably, the specific architecture of the dynamically extended knowledge base module is as follows:

[0023] (1) Basic layer: stores physical table structure information such as table name, storage type, and creation time; HBase table types are generally divided into two types: frequency-level aggregation and time-level aggregation;

[0024] (2) Business layer: defines field-level semantic attributes and relationships;

[0025] (3) Mapping layer: establishes a bidirectional mapping between natural language vocabulary and database fields, and supports dynamic expansion to adapt to new business terms;

[0026] (4) Intelligent management: Automatically synchronize table structure changes through the metadata update engine, and combine with the consistency check module to ensure real-time alignment of metadata with the HBase table schema.

[0027] Preferably, the specific architecture of the high-performance MCP data interface module is as follows:

[0028] (1) Protocol parsing layer: supports JSON format input and parses query elements;

[0029] (2) Parameter verification layer: Verify the legitimacy of fields based on knowledge base metadata and filter out illegal query requests;

[0030] (3) Result encapsulation layer: converts the raw data returned by HBase into a structured format that adapts to the output requirements of large models and supports the generation of natural language descriptions.

[0031] Preferably, in step (2), the specific operations of intent classification and term resolution are as follows:

[0032] (21) Query intent classification: The LoRA enhanced domain model module uses the query intent classification task in the dual-task learning mechanism to determine whether the user question involves data query, with a confidence level of ≥95%;

[0033] (22) Term parsing: Through the bidirectional mapping between natural language vocabulary and database fields in the mapping layer, the natural language terms in the user query are parsed into corresponding database fields and associated with the corresponding data tables.

[0034] Preferably, in step (3), the specific operations of query logic generation are as follows:

[0035] (31) Metadata acquisition: The LoRA enhanced domain model module obtains relevant metadata from the dynamically extended knowledge base module;

[0036] (32) Query parameter generation: Combine the acquired metadata and query logic template library to generate query parameters that comply with HBase specifications.

[0037] Preferably, in step (4), the specific operations of data query and result return are as follows:

[0038] (41) Protocol parsing: The protocol parsing layer of the high-performance MCP data interface module parses query parameters and extracts query elements such as table names, column families, and filter conditions;

[0039] (42) Parameter verification: The parameter verification layer verifies the legitimacy of fields based on the knowledge base metadata and filters out illegal query requests;

[0040] (43) Query optimization: The query optimization layer converts natural language filter conditions into HBase native filters through predicate push-down technology;

[0041] (44) Data query: call the HBase interface to execute the query;

[0042] (45) Result encapsulation: The result encapsulation layer converts the raw data returned by HBase into a structured format to adapt to the output requirements of large models;

[0043] (46) Result return: The LoRA enhanced domain large model module organizes the structured data into natural language reports or structured tables and feeds them back to the user.

[0044] The beneficial effects of the present invention are: realizing natural language intelligent query of radio monitoring data, performing well in radio monitoring data query, being able to quickly and accurately process users' natural language query requests, significantly improving the efficiency and accuracy of data query, being able to realize efficient conversion from natural language input to precise data query, lowering technical threshold and improving query performance, being able to be widely used in data query and analysis in the field of radio monitoring, and providing strong technical support for radio management and decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0046] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0047] like Figure 1 In the described embodiment, a large-model-based end-to-end radio monitoring data query method includes a LoRA-enhanced domain large model module, a dynamically extended knowledge base module, and a high-performance MCP data interface module; the LoRA-enhanced domain large model module uses the LLaMA-Factory framework in combination with LoRA technology to perform domain adaptation on the 32B Deepseek-r1-distill-qwen large model to achieve accurate understanding of radio professional semantics; the dynamically extended knowledge base module constructs a three-level metadata management system to provide structured data semantic support for the large model; the high-performance MCP data interface module designs a three-layer protocol architecture to achieve efficient interaction between the large model and HBase.

[0048] Among them, the specific architecture of the LoRA enhanced domain large model module is as follows:

[0049] (1) Lightweight fine-tuning architecture: inserting a low-rank adapter into the Transformer layer freezes more than 98% of the original model parameters and only trains about 2 million adapter parameters;

[0050] (2) Dual-task learning mechanism: Simultaneously train the "query intent classification" and "query logic generation" tasks. The former determines whether the user's question involves data query with an accuracy rate of ≥95%, and the latter generates query parameters that conform to the HBase structure based on the knowledge base metadata.

[0051] (3) Domain knowledge injection: Through the term mapping table and query logic template library, the model is guided to generate standardized query logic, supporting complex scenarios such as single-table query, multi-table association, and time range filtering.

[0052] The specific architecture of the dynamically extended knowledge base module is as follows:

[0053] (1) Basic layer: stores physical table structure information such as table name, storage type, and creation time; HBase table types are generally divided into two types: frequency-level aggregation and time-level aggregation;

[0054] The following is the HBase table metadata aggregated at the frequency level with a 15-minute granularity:

[0055]

[0056] The following is the HBase table metadata aggregated at a 15-minute granularity:

[0057]

[0058]

[0059] (2) Business layer: defines field-level semantic attributes and relationships;

[0060] (3) Mapping layer: establishes a bidirectional mapping between natural language vocabulary and database fields, and supports dynamic expansion to adapt to new business terms;

[0061] (4) Intelligent management: Automatically synchronize table structure changes through the metadata update engine, and combine with the consistency check module to ensure real-time alignment of metadata with the HBase table schema.

[0062] The specific architecture of the high-performance MCP data interface module is as follows:

[0063] (1) Protocol parsing layer: supports JSON format input and parses query elements;

[0064] (2) Parameter verification layer: Verify the legitimacy of fields based on knowledge base metadata and filter out illegal query requests;

[0065] (3) Result encapsulation layer: converts the raw data returned by HBase into a structured format that adapts to the output requirements of large models and supports the generation of natural language descriptions.

[0066] The specific steps are as follows:

[0067] (1) User question input: The user submits a query in natural language;

[0068] (2) Intent classification and term analysis: The LoRA enhanced domain model module determines that the problem is a data query problem, and uses the mapping layer to parse the natural language terms into database fields and associate them with the corresponding data tables;

[0069] The specific operations of intent classification and term resolution are as follows:

[0070] (21) Query intent classification: The LoRA enhanced domain model module uses the query intent classification task in the dual-task learning mechanism to determine whether the user question involves data query, with a confidence level of ≥95%;

[0071] (22) Term parsing: Through the bidirectional mapping between natural language vocabulary and database fields in the mapping layer, the natural language terms in the user query are parsed into the corresponding database fields. For example, "maximum level" is parsed into the field "max_dp" and associated with the corresponding data table.

[0072] (3) Query logic generation: The LoRA enhanced domain model module combines the metadata of the dynamically extended knowledge base module to generate query parameters that comply with HBase specifications, including table name, target field, and filter conditions;

[0073] The specific operations of query logic generation are as follows:

[0074] (31) Metadata acquisition: The LoRA enhanced domain model module obtains relevant metadata from the dynamically extended knowledge base module, including field-level semantic attributes, association relationships, etc.

[0075] (32) Query parameter generation: Combine the acquired metadata and query logic template library to generate query parameters that comply with HBase specifications, including table name, target field, filter conditions, etc.

[0076] (4) Data query and result return: The high-performance MCP data interface module calls the HBase interface to execute the query, and returns the result after the optimization layer reduces invalid data scanning; the LoRA enhanced domain large model module organizes the data into natural language reports or structured tables and feeds them back to the user;

[0077] The specific operations for data query and result return are as follows:

[0078] (41) Protocol parsing: The protocol parsing layer of the high-performance MCP data interface module parses query parameters and extracts query elements such as table names, column families, and filter conditions;

[0079] (42) Parameter verification: The parameter verification layer verifies the legitimacy of fields based on the knowledge base metadata and filters out illegal query requests;

[0080] (43) Query optimization: The query optimization layer converts natural language filter conditions into HBase native filters through predicate push-down technology;

[0081] (44) Data query: call the HBase interface to execute the query;

[0082] (45) Result encapsulation: The result encapsulation layer converts the raw data returned by HBase into a structured format to adapt to the output requirements of large models;

[0083] (46) Result return: The LoRA enhanced domain large model module organizes the structured data into natural language reports or structured tables and feeds them back to the user.

[0084] The specific implementation examples are as follows:

[0085] (1) The user submits a query in natural language through the interactive interface, for example, "Query the average power level of the 87MHz to 108MHz frequency bands in Xihu District from 6:00 PM to 7:00 PM on May 23, 2025." The system receives the query and enters the intelligent analysis process, automatically identifying the geographical area, time range, frequency parameters, and target indicator (average power level).

[0086] (2) Intent classification and term analysis. The specific steps are as follows:

[0087] (21) The LoRA enhanced domain model uses the query intent classification task in dual-task learning to determine whether the user question is a data query class, and the confidence threshold is set to ≥95%.

[0088] Example: For non-query questions such as "What is the signal coverage in Xihu District?", the following guidance message is returned: "Please enter your specific data query requirements, for example: Query the parameters of XX frequency at XX time in XX area."

[0089] (22) Through the bidirectional mapping table of the mapping layer, natural language terms are parsed into database fields and associated tables:

[0090]

[0091] (3) Query logic generation, the specific steps are as follows:

[0092] (31) Metadata acquisition, achieving accurate association between geographic areas and task IDs through a two-layer mapping table;

[0093] (32) District and County Code and Site Mapping Table

[0094]

[0095]

[0096] (33) Site name and task ID mapping table (only relevant fields are shown)

[0097]

[0098] (34) Geocoding analysis: User inputs "Xihu District" → queries the district and county code table to obtain AREA_CODE = 330106 → obtains the corresponding SITE_ID = MONITOR_SITE_001;

[0099] (35) Task ID retrieval: query the MFID field in the bo_task_history table through SITE_ID = MONITOR_SITE_001 → obtain TASKID = TASK_20250523_001;

[0100] (36) ROWKEY construction: Generate ROWKEY=TASK_20250523 (format: task ID_timestamp).

[0101] (37) Combine metadata with HBase specifications to generate query parameters including table name, target field, and filter conditions:

[0102] Table name: monitor:bo_spectrum_net_txtfile_15m_taskid_time (frequency-level 15-minute granularity aggregation HBase table);

[0103] Target field: avg field under column family amp (corresponding to average level indicator);

[0104] Filter by:

[0105] Constructing compound filter conditions:

[0106] (371) Task ID prefix matching: ROWKEY starts with [TASKID_timestamp];

[0107] (372) Frequency range check: The starting frequency is equal to 87 MHz and the ending frequency is equal to 108 MHz.

[0108] (4) Data query and result return, the specific steps are as follows:

[0109] (41) Protocol parsing, MCP interface parses query parameters and extracts key elements:

[0110] Table name: monitor:bo_spectrum_net_txtfile_15m_taskid_time;

[0111] Filter conditions: Task ID prefix match + frequency range check + timestamp range.

[0112] (42) Business rule verification:

[0113] Frequency range compliance: The user-entered frequency must be within the STARTFREQ / ENDFREQ range of the bo_task_history table;

[0114] Time validity: The query time must be within the task STARTTIME / ENDTIME;

[0115] (43) Through predicate push-down technology, natural language filter conditions are converted into HBase native filter logic (such as row key prefix matching + column value verification), reducing full table scans and improving query efficiency;

[0116] (44) Execute HBase query operation (pseudo code):

[0117] (441) Create a scan object and set filter conditions;

[0118] (442) Call the HBase interface to perform the scan;

[0119] (443) Traverse the scan results and extract the target column family and field data;

[0120] (444) Close the scanner to release resources;

[0121] (45) Result encapsulation, converting HBase raw data into a structured format (example JSON):

[0122] {

[0123] "Region": "West Lake District",

[0124] "Time Range": "2025-05-23 18:00~19:00",

[0125] "Frequency range": "87MHz~108MHz",

[0126] "Average level": [7.5, 7.8, 8.0, 8.2] / / Segment data at 15-minute granularity

[0127] }

[0128] (46) The results are returned to the LoRA model to organize the data into natural language reports

[0129] Natural language report: "From 18:00 to 19:00 on May 23, 2025, the average power levels of the 87-108 MHz frequency band in Xihu District were: 18:00-18:15 (7.5), 18:15-18:30 (7.8), 18:30-18:45 (8.0), and 18:45-19:00 (8.2)."

[0130] (5) Domain Adaptation and Expansion

[0131] Lightweight fine-tuning: When adding a new business term (such as "band occupancy"), simply add "band occupancy" → band_occupancy in the mapping layer and perform incremental fine-tuning (cycle ≤ 12 hours);

[0132] Metadata synchronization: When the HBase table structure changes (such as adding a new column family real_time_data), the metadata update engine automatically synchronizes the table schema to ensure real-time adaptation of the query logic generated by the large model.

[0133] The implementation examples are as follows:

[0134] Take the user's query for "the average level of the 87MHz to 108MHz frequency band in Xihu District from 18:00 to 19:00 on May 23, 2025" as an example:

[0135] Intent classification: The large model determines it as a data query (95% confidence level);

[0136] Terminology parsing: Through the mapping between the district and county code table and the task table, obtain TASKID = TASK_20250523 and frequency parameters;

[0137] Query execution: Generates HBase filter conditions to retrieve corresponding ROWKEY data in the monitor:bo_spectrum_net_txtfile_15m_taskid_time table, reducing the time consumption by 40% compared to traditional methods.

[0138] Results returned: The large model integrates 15-minute granularity data into user-readable reports with an accuracy rate of ≥95%.

[0139] At this point, an end-to-end radio monitoring data query method based on a large model is completed.

[0140] After experimental verification, the end-to-end radio monitoring data query method based on the large model of the present invention has the following advantages:

[0141] Efficient field adaptation: LoRA technology compresses 98% of fine-tuning parameters, supports conventional computing power deployment, and the adaptation cycle is ≤12 hours;

[0142] Accurate semantic conversion: Professional terminology recognition rate ≥ 95%, query logic and HBase table structure matching degree ≥ 95%, significantly reducing semantic ambiguity errors;

[0143] Low-code extension: When adding new data tables / fields, only the knowledge base metadata needs to be updated, shortening the business response cycle from 7 days to less than 1 day.

[0144] In summary, this invention enables natural language intelligent querying of radio monitoring data and efficient domain adaptation. Using LoRA technology, it compresses the scale of fine-tuning parameters by over 98%, reducing video memory requirements and enabling domain adaptation in conventional computing environments with a fine-tuning cycle of ≤12 hours. It also achieves precise semantic conversion, improves professional terminology recognition, and generates query logic with a ≥95% match between the query logic and the HBase table structure, significantly reducing query errors caused by semantic ambiguity. Furthermore, low-code business expansion allows for new data tables or fields to be added, requiring only updating the knowledge base metadata. Adaptation can be completed through incremental fine-tuning, shortening the business response cycle from one week of workdays with traditional methods to less than one day. This method can be widely applied to data query and analysis in the radio monitoring field, providing strong technical support for radio management and decision-making.

Claims

1. A large-scale model-based end-to-end radio monitoring data query method, characterized in that: It includes a LoRA enhanced domain model module, a dynamically expanded knowledge base module, and a high-performance MCP data interface module. The LoRA enhanced domain model module uses the LLaMA-Factory framework combined with LoRA technology to adapt the 32B Deepseek-r1-distill-qwen model to the domain, achieving accurate understanding of radio professional semantics. The dynamic expansion knowledge base module builds a three-level metadata management system to provide structured data semantic support for the large model; the high-performance MCP data interface module designs a three-layer protocol architecture to achieve efficient interaction between the large model and HBase. The specific operation steps are as follows: (1) User question input: The user submits a query in natural language; (2) Intent classification and term analysis: The LoRA enhanced domain model module determines that the problem is a data query problem, and uses the mapping layer to parse the natural language terms into database fields and associate them with the corresponding data tables; (3) Query logic generation: The LoRA enhanced domain model module combines the metadata of the dynamically extended knowledge base module to generate query parameters that comply with HBase specifications, including table name, target field, and filter conditions; (4) Data query and result return: The high-performance MCP data interface module calls the HBase interface to execute the query, and returns the result after the optimization layer reduces invalid data scanning; the LoRA enhanced domain large model module organizes the data into natural language reports or structured tables and feeds them back to the user.

2. The end-to-end radio monitoring data query method based on a large model according to claim 1 is characterized in that: The specific architecture of the LoRA enhanced domain large model module is as follows: (1) Lightweight fine-tuning architecture: insert a low-rank adapter into the Transformer layer, freeze more than 98% of the original model parameters, and only train 2 million adapter parameters; (2) Dual-task learning mechanism: Simultaneously train the "query intent classification" and "query logic generation" tasks. The former determines whether the user's question involves data query with an accuracy rate of ≥95%, and the latter generates query parameters that conform to the HBase structure based on the knowledge base metadata. (3) Domain knowledge injection: Through the term mapping table and query logic template library, guide the model to generate standardized query logic.

3. The end-to-end radio monitoring data query method based on a large model according to claim 1 is characterized in that: The specific architecture of the dynamically extended knowledge base module is as follows: (1) Basic layer: stores physical table structure information such as table name, storage type, and creation time; HBase table types are generally divided into two types: frequency-level aggregation and time-level aggregation; (2) Business layer: defines field-level semantic attributes and relationships; (3) Mapping layer: establishes a bidirectional mapping between natural language vocabulary and database fields, and supports dynamic expansion to adapt to new business terms; (4) Intelligent management: Automatically synchronize table structure changes through the metadata update engine, and combine with the consistency check module to ensure real-time alignment of metadata with the HBase table schema.

4. The end-to-end radio monitoring data query method based on a large model according to claim 1 is characterized in that: The specific architecture of the high-performance MCP data interface module is as follows: (1) Protocol parsing layer: supports JSON format input and parses query elements; (2) Parameter verification layer: Verify the legitimacy of fields based on knowledge base metadata and filter out illegal query requests; (3) Result encapsulation layer: converts the raw data returned by HBase into a structured format that adapts to the output requirements of large models and supports the generation of natural language descriptions.

5. The end-to-end radio monitoring data query method based on a large model according to claim 2 or 3, characterized in that: In step (2), the specific operations of intent classification and term resolution are as follows: (21) Query intent classification: The LoRA enhanced domain model module uses the query intent classification task in the dual-task learning mechanism to determine whether the user question involves data query, with a confidence level of ≥95%; (22) Term parsing: Through the bidirectional mapping between natural language vocabulary and database fields in the mapping layer, the natural language terms in the user query are parsed into corresponding database fields and associated with the corresponding data tables.

6. The end-to-end radio monitoring data query method based on a large model according to claim 2 or 3, characterized in that: In step (3), the specific operations of query logic generation are as follows: (31) Metadata acquisition: The LoRA enhanced domain model module obtains relevant metadata from the dynamically extended knowledge base module; (32) Query parameter generation: Combine the acquired metadata and query logic template library to generate query parameters that comply with HBase specifications.

7. The end-to-end radio monitoring data query method based on a large model according to claim 4 is characterized in that: In step (4), the specific operations of data query and result return are as follows: (41) Protocol parsing: The protocol parsing layer of the high-performance MCP data interface module parses query parameters and extracts query elements such as table names, column families, and filter conditions; (42) Parameter verification: The parameter verification layer verifies the legitimacy of fields based on the knowledge base metadata and filters out illegal query requests; (43) Query optimization: The query optimization layer converts natural language filter conditions into HBase native filters through predicate push-down technology; (44) Data query: call the HBase interface to execute the query; (45) Result encapsulation: The result encapsulation layer converts the raw data returned by HBase into a structured format to adapt to the output requirements of large models; (46) Result return: The LoRA enhanced domain large model module organizes the structured data into natural language reports or structured tables and feeds them back to the user.

Images