Beach big data geographic space intelligent application service method based on large model Agent

Abstract

The invention provides a beach big data geographic space intelligent application service method based on a large model Agent, and relates to the technical field of geographic space data artificial intelligence processing and analysis. According to the method, a distributed platform comprising a parallel computing cluster, a spatial database and a large language model is constructed, and standardized storage and metadata management of multi-source remote sensing, vector and attribute data are supported. The system generates a spatio-temporal query instruction through natural language analysis, completes cross-source matching and spatial calculation, outputs an analysis result in a multi-modal form such as a map, a chart and a text, and realizes intelligent and interactive service of geographic data processing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geographic spatial data artificial intelligence processing and analysis, in particular to a beach big data geographic spatial intelligent application service method based on a large model Agent. BACKGROUND

[0002] With the development of remote sensing images, Internet of Things sensors, and social media technologies, the coastal zone continuously generates high-resolution, multi-dimensional data, forming beach big data represented by images, vectors, texts, and sensor values. Traditional relational databases and single-machine geographic information system (GIS) software have been unable to meet the demand for real-time retrieval and comprehensive analysis of massive, multi-source beach data. In recent years, spatial big data processing platforms based on distributed computing and storage architectures, such as in-memory computing cluster systems and column-oriented distributed databases, have gradually become mainstream, enabling parallel computing of PB-level data on large-scale node clusters. However, such systems still generally rely on manual writing of complex query statements or data processing procedures, making it difficult for ordinary business personnel to use them.

[0003] In recent years, the semantic understanding and reasoning capabilities of general artificial intelligence language models have rapidly improved, and their application scope has gradually expanded to the field of geographic information services. The latest research results explore the automatic conversion of users' natural language questions into executable database query instructions and geographic spatial analysis processes through "human-like assistant systems," and complete graphic display and text report generation, aiming to enable non-professional users to complete professional geographic spatial tasks through dialogue. Meanwhile, in-memory-driven elastic data processing architectures and geographic coding spatial indexes are becoming increasingly mature, supporting large-scale parallel data aggregation, cropping, statistics, and map output, providing bottom support for the response speed and expansion capability of geographic intelligent systems.

[0004] Current geographic information intelligent systems still have a significant gap between semantic understanding and actual queries:

[0005] ①When processing beach-related tasks, there is often a lack of field-specific mapping and multi-layered intent recognition for multi-source heterogeneous data (such as beach morphology, slope, water quality, and tourist density), resulting in low query accuracy.

[0006] ②The data storage, semantic analysis, spatio-temporal calculation, and visualization functions within the system are often loosely integrated by different components, with weak coupling between the overall running process, and without automatic error correction and multi-round self-evolution capabilities.

[0007] ③The output result is single in form, making it difficult to provide multi-modal information integrating heat maps, three-dimensional models, data charts, and natural language analysis reports, limiting the assistance capability for decision-makers. SUMMARY

[0008] In order to overcome the deficiencies of the prior art, the purpose of the present application is to provide a beach big data geospatial intelligent application service method based on a large model Agent, which realizes intelligent and automatic processing and interactive display of beach multi-source spatio-temporal data, and improves the convenience, intelligence and adaptability of geospatial application services.

[0009] In order to achieve the above-mentioned purpose, the present application provides the following scheme:

[0010] A beach big data geospatial intelligent application service method based on a large model Agent, comprising:

[0011] A distributed storage-computing environment oriented to beach multi-source spatio-temporal data is constructed; the distributed storage-computing environment comprises a software and hardware platform composed of a parallel computing node cluster and a column-oriented spatial database, and a large language model Agent;

[0012] The cleaned and standardized remote sensing raster, vector form and attribute records are imported into the spatial database, and a metadata table of record data type, spatial reference system and update time is generated;

[0013] Natural language or voice query requests from a user terminal are received, and the query requests are routed to the large language model Agent residing in the distributed storage-computing environment;

[0014] The query requests are subjected to entity extraction, intent recognition and structured conversion by the large language model Agent, to generate machine-readable spatio-temporal query and analysis instructions;

[0015] Cross-source data matching, spatio-temporal filtering, aggregation statistics, spatial clipping and numerical filtering are completed in the spatial database according to the spatio-temporal query and analysis instructions, to obtain analysis results;

[0016] The analysis results are output in the form of dynamic maps, statistical charts and natural language reports in a multi-modal form.

[0017] Preferably, the distributed storage-computing environment oriented to beach multi-source spatio-temporal data is constructed, comprising:

[0018] A parallel computing cluster composed of multiple physical computing nodes is started; the computing nodes are connected by an interconnection network to build a communication link, and a memory-driven elastic computing framework is deployed to support distributed execution of spatial computing tasks;

[0019] Components of the column-oriented spatial database are deployed in the computing nodes, a spatial index mechanism based on latitude and longitude coding is used to build data row keys, and the spatial database and the parallel computing cluster are communicatively integrated to constitute a unified storage-computing integrated platform.

[0020] Integrating the large language model agent into the storage-computing integrated platform; deploying the large language model agent in an independent computing container and configuring a semantic parsing interface, an inference executor, and a model parameter access channel to support the conversion of natural language into spatial analysis tasks;

[0021] Initialize the system configuration file, set the spatial data access interface, model service call path and analysis module scheduling rules, and publish the application interface address that can be used for service calls after initialization is completed.

[0022] Preferably, the cleaned and standardized remote sensing raster, vector form and attribute records are imported into the spatial database, and a metadata table is generated that records the data type, spatial reference system and update time, including:

[0023] Performing unified coordinate system conversion on beach remote sensing images, vector layers, and structured attribute data from multiple source platforms, and verifying spatial geometry validity, field integrity, and timestamp consistency to obtain verified data files; the data files include the remote sensing raster, the vector form, and the attribute records;

[0024] The verified data files are classified according to data types to obtain classified data; wherein the remote sensing grid is indexed by path according to time series, the vector form is classified by layer name, and the attribute record is tabulated according to primary key mapping;

[0025] Using a row key strategy based on a geocoding mechanism, each of the classified data is written into a corresponding column family or table space of the spatial database, and the import status and batch information are recorded;

[0026] According to the type, spatial reference system, time stamp and data source identifier of each of the classified data written into the spatial database, a structured metadata record is generated and written into the dedicated metadata table.

[0027] Preferably, receiving a natural language or voice query request from a user terminal and routing the query request to a large language model agent residing in the distributed storage-computing environment includes:

[0028] Receive a user's query content at the user terminal through a graphical user interface or a voice input interface, wherein the query content is a natural language sentence in text form, or a text string transcribed by a voice recognition module;

[0029] Performing format preprocessing on the query content, and extracting time expressions, geographical areas, and keywords of analysis indicators in the query content, and verifying whether the query content meets semantic analysis requirements according to the keywords;

[0030] Packaging the query content meeting the requirements into a structured query request; the query request packet includes original query text, semantic label annotation results, user identification, and query timestamp;

[0031] Routing and sending the query request to a large language model Agent residing in the distributed storage-computing environment through a task scheduling interface in the distributed storage-computing environment.

[0032] Preferably, the query request is subjected to entity extraction, intent recognition, and structured conversion by the large language model Agent to generate machine-readable spatio-temporal query and analysis instructions, including:

[0033] Performing word segmentation, part-of-speech tagging, and syntax analysis on the natural language query text in the query request to extract semantic elements of geographical location, time range, target attribute, and operation intent;

[0034] Analyzing the semantic elements based on a pre-trained intent classification model to identify the task type corresponding to the query request; the task type includes statistics, comparison, prediction, trend analysis, or spatial relationship judgment;

[0035] Mapping the extracted semantic elements to field names and data table structures in the spatial database, and generating an intermediate instruction template containing field names, operators, filtering conditions, and spatial ranges;

[0036] Analyzing and formatting the intermediate query instruction template to generate a structured spatio-temporal query and analysis instruction; the spatio-temporal query and analysis instruction is expressed in structured query language statements, object-oriented JSON expressions, or graph computing task description syntax.

[0037] Preferably, according to the spatio-temporal query and analysis instruction, cross-source data matching, spatio-temporal filtering, aggregation statistics, spatial clipping, and numerical filtering are completed in the spatial database to obtain analysis results, including:

[0038] According to the field conditions, time range, and spatial range in the spatio-temporal query and analysis instruction, the field index and spatial index functions of the spatial database are called to determine the original data records matching the instruction;

[0039] Matching and association operations are performed on data from different data sources in the original data records according to the primary key field, time field or spatial location field set in the spatio-temporal query and analysis instruction, so as to realize cross-source data matching.

[0040] After completing the cross-source data matching, a screening operation is performed according to the time range and spatial boundary conditions set in the spatio-temporal query and analysis instruction, and only data records meeting the conditions are retained to complete the spatio-temporal screening process and obtain screened data records.

[0041] Aggregation statistics are performed on the screened data records according to the aggregation field and statistical method set in the spatio-temporal query and analysis instruction to obtain aggregation statistical results.

[0042] Based on the aggregation statistical results, spatial clipping is performed according to the spatial region boundary specified in the spatio-temporal query and analysis instruction to obtain the clipping results; the clipping mode of the spatial clipping includes administrative district limitation, grid division and buffer clipping.

[0043] Numerical filtering operations are performed on the clipping results according to the numerical threshold or Boolean logic condition in the spatio-temporal query and analysis instruction to obtain the analysis results.

[0044] Preferably, the analysis results are output in the form of multi-modal dynamic maps, statistical charts and natural language reports, including:

[0045] Based on the geographic coordinates, spatial boundaries and index fields contained in the analysis results, a dynamic map containing layer superposition, color coding and zoom interaction is drawn through a map rendering engine.

[0046] According to the statistical fields and time series in the analysis results, a data visualization component is called to generate a two-dimensional statistical chart.

[0047] The structured fields and semantic labels in the analysis results are input into a natural language generation template to generate an automatic text report describing the geographic features, statistical profiles and spatio-temporal change trends of the queried target.

[0048] The dynamic map, the two-dimensional statistical chart and the automatic text report are integrated into an interactive interface, and switching views, viewing layer details or exporting result files are supported.

[0049] Preferably, the two-dimensional statistical chart includes a line chart, a column chart, a pie chart and a heat map.

[0050] A beach big data geographic spatial intelligent application service method based on a large model Agent includes:

[0051] A distributed environment construction unit is configured to construct a distributed storage-computing environment for beach multi-source spatio-temporal data; the distributed storage-computing environment comprises a software and hardware platform composed of a parallel computing node cluster and a column-oriented spatial database and a large language model Agent;

[0052] A data access and standardized storage unit is configured to import cleaned and standardized remote sensing raster, vector and attribute records into the spatial database and generate a metadata table of record data types, spatial reference systems and update times;

[0053] A query request receiving and routing unit is configured to receive a natural language or voice query request from a user terminal and route the query request to the large language model Agent residing in the distributed storage-computing environment;

[0054] A semantic analysis and instruction generation unit is configured to perform entity extraction, intent recognition and structured conversion on the query request by the large language model Agent to generate machine-readable spatio-temporal query and analysis instructions;

[0055] A spatial computing and analysis execution unit is configured to complete cross-source data matching, spatio-temporal filtering, aggregation statistics, spatial clipping and numerical filtering in the spatial database according to the spatio-temporal query and analysis instructions to obtain analysis results;

[0056] A multi-modal result presentation unit is configured to output the analysis results in the form of dynamic maps, statistical charts and natural language reports.

[0057] According to the embodiments of the present application, the following technical effects are provided:

[0058] The present application can automatically analyze natural language or voice query requests of users, generate machine-readable spatio-temporal query and analysis instructions, and efficiently complete multi-source data matching and analysis processing in the spatial database, finally output the analysis results in the form of multi-modal, realize intelligent and automatic processing and interactive display of beach multi-source spatio-temporal data, and improve the convenience, intelligence and adaptability of geographic spatial application services. BRIEF DESCRIPTION OF DRAWINGS

[0059] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0060] Figure 1A method flowchart is provided for the embodiment of the present application.

[0061] Figure 2 A storage and computing overall architecture schematic diagram is provided for the embodiment of the present application.

[0062] Figure 3 A spatial data storage table structure design schematic diagram based on HBase is provided for the embodiment of the present application.

[0063] Figure 4 A processing framework design schematic diagram based on a large language model is provided for the embodiment of the present application.

[0064] Figure 5 An overall work flowchart is provided for the embodiment of the present application.

[0065] Figure 6 A tool chain work flowchart is provided for the embodiment of the present application. DETAILED DESCRIPTION

[0066] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0067] The purpose of the present application is to provide a beach big data geospatial intelligent application service method based on a large model Agent, which realizes intelligent and automatic processing and interactive display of beach multi-source spatio-temporal data, and improves the convenience, intelligence and adaptability of geospatial application services.

[0068] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0069] Figure 1 A method flowchart is provided for the embodiment of the present application, as shown in Figure 1 The present application provides a beach big data geospatial intelligent application service method based on a large model Agent, which includes:

[0070] Step 100: Construct a distributed storage-computing environment for beach multi-source spatio-temporal data; the distributed storage-computing environment includes a software and hardware platform composed of a parallel computing node cluster and a column-oriented spatial database, and a large language model Agent;

[0071] Step 200: Import the cleaned and standardized remote sensing raster, vector form and attribute records into the spatial database, and generate a metadata table that records data type, spatial reference system and update time;

[0072] Step 300: receiving a natural language or voice query request from a user terminal, and routing the query request to a large language model agent residing in a distributed storage-computing environment;

[0073] Step 400: Perform entity extraction, intent recognition, and structural conversion on the query request through the large language model agent to generate machine-readable spatiotemporal query and analysis instructions;

[0074] Step 500: completing cross-source data matching, spatiotemporal screening, aggregate statistics, spatial clipping, and numerical filtering in the spatial database according to the spatiotemporal query and analysis instructions to obtain analysis results;

[0075] Step 600: Output the analysis results in a multimodal form of dynamic maps, statistical charts, and natural language reports.

[0076] Preferably, a distributed storage-computing environment for multi-source spatiotemporal data of beaches is constructed, including:

[0077] Launching a parallel computing cluster consisting of multiple physical computing nodes; establishing communication links between the computing nodes through an internet network, and deploying a memory-driven elastic computing framework to support the distributed execution of spatial computing tasks;

[0078] Deploying the components of the spatial database for column storage in the computing nodes, constructing data row keys using a spatial index mechanism based on latitude and longitude encoding, and integrating the spatial database with the parallel computing cluster to form a unified storage-computing integrated platform;

[0079] Integrating the large language model agent into the storage-computing integrated platform; deploying the large language model agent in an independent computing container and configuring a semantic parsing interface, an inference executor, and a model parameter access channel to support the conversion of natural language into spatial analysis tasks;

[0080] Initialize the system configuration file, set the spatial data access interface, model service call path and analysis module scheduling rules, and publish the application interface address that can be used for service calls after initialization is completed.

[0081] In the present invention, Figure 2To achieve semantic perception and structured retrieval of large model agents for multi-source heterogeneous spatial data, uniform structured storage and metadata annotation processing must be performed on beach remote sensing images, vector boundaries, and attribute records. This process differs from simple storage behavior under the traditional HBase table structure, introducing cross-type field mapping mechanisms, spatial row key encoding strategies, and multi-class meta-information registration mechanisms to ensure that downstream natural language queries can directly drive unified access and reasoning calls for heterogeneous data through semantic instructions.

[0082] Specifically, the embodiment first automatically identifies the structure type of the data source according to the original format, distinguishing between remote sensing grids (such as.tif), spatial vectors (such as.geojson,.shp), and attribute table data (such as.csv or.parquet). After identification, the embodiment uses a uniform coordinate reference system (EPSG:4326) as a standard reference, calls the GDAL interface to perform coordinate transformation and precision verification on vector and raster data, and if there is a reference system missing or boundary illegal condition, the embodiment will automatically abort the storage and generate an error record. For attribute records with inconsistent field names, the embodiment will use a field alias mapping table to uniformly map semantic synonyms to standard fields, such as "beach area" to "area" and "grain grade" to "grain_grade", and write the mapping relationship into the field interpretation table for subsequent large model alignment.

[0083] After standardization, the data is written into the spatial database according to the preset column family structure, where the "geo" column family saves the object geometry structure (represented by WKT / WKB encoding), the "attr" column family saves the parsed attribute fields and their values, and the "time" column family records the data collection or storage time (uniformly using ISO-8601 format). To improve spatial query performance, the embodiment uses the object geometric center point-based Geohash spatial encoding mechanism to construct the HBase row key, with the key format being "Geohash encoding + data type identifier + data timestamp", such as "wq6r8tvzgds2_RASTER_20240601". This design maintains spatial adjacency and facilitates distributed positioning and scheduling of data in large-scale node clusters.

[0084] In the data writing process, the embodiment automatically extracts meta information and synchronously writes into the specially designed metadata table. The metadata table records the type (RASTER, VECTOR or TABULAR) of each piece of data, data source ID, unified spatial reference system code, bounding box coordinate, resolution or sampling rate, field list and unit, latest update time and the like. These metadata are not only used for subsequent task scheduling module to query optimization, but also provide necessary support for semantic tasks such as field perception, unit understanding and type inference performed by the large model Agent. The standardized warehousing and metadata management mechanism constitutes the key link of the present application which is different from the existing spatial data processing scheme, and lays the executable and schedulable basic environment for natural language driven intelligent spatio-temporal query.

[0085] In the present application, as shown in Figure 3 The collaborative design of the Spark distributed computing framework and the HBase spatial database constitutes the basic execution layer of the system to realize high concurrency and high throughput intelligent service, and the two realize the closed loop support from natural language analysis to spatio-temporal analysis result output through unified data scheduling logic and index structure. First, the Spark framework not only carries out the execution tasks of data reading, writing and intermediate processing, but also realizes spatial partitioning and parallel distribution of large-scale beach remote sensing and vector data through the RDD mechanism, avoiding the bottleneck of traditional GIS serial processing. In the data inflow stage, the embodiment performs format identification and label annotation on structured data (such as attribute records) and unstructured data (such as remote sensing images), which provides a parsing basis for subsequent field mapping and semantic call. At the cluster configuration level, the embodiment adopts static resource quota combined with dynamic scheduling mechanism to ensure that the semantic query can be executed in the specified Worker node after being translated into a Spark task, and to keep the calculation state traceable.

[0086] Correspondingly, the design of the HBase table structure not only follows the general column family division principle, but also introduces the spatial index row key mechanism and field cluster design logic according to the high-dimensional characteristics of geographic spatial data. All spatial fields are uniformly mapped to the geo column family, time attributes are independently entered into the time column family, and attribute fields are entered into the attr column family. The row key is composed of Geohash encoding + data identification + timestamp, which not only supports range retrieval, but also corresponds to the query range parameters generated by the semantic analysis module one by one. In order to improve the efficiency of downstream spatio-temporal clipping and aggregation, the embodiment designs a strategy based on primary key compression and column family index precalculation, supplemented by regional hierarchical slice control, to realize near real-time distributed query response. In terms of data consistency, all cross-source merged writing operations are controlled through transaction lock, and field integrity check rules are introduced to avoid misinterpretation of fields or misplacement of field meanings by the model Agent, so as to ensure the consistent linkage of natural language and data semantics.

[0087] Through the configuration and optimization of the above-mentioned Spark and HBase collaborative mechanism, an end-to-end closed-loop path from natural language to structured instructions to distributed execution is achieved, supporting the core capabilities of geospatial intelligent applications driven by the large model proposed in this invention.

[0088] Preferably, the cleaned and standardized remote sensing raster, vector form and attribute records are imported into the spatial database, and a metadata table is generated that records the data type, spatial reference system and update time, including:

[0089] Performing unified coordinate system conversion on beach remote sensing images, vector layers, and structured attribute data from multiple source platforms, and verifying spatial geometry validity, field integrity, and timestamp consistency to obtain verified data files; the data files include the remote sensing raster, the vector form, and the attribute records;

[0090] The verified data files are classified according to data types to obtain classified data; wherein the remote sensing grid is indexed by path according to time series, the vector form is classified by layer name, and the attribute record is tabulated according to primary key mapping;

[0091] Using a row key strategy based on a geocoding mechanism, each of the classified data is written into a corresponding column family or table space of the spatial database, and the import status and batch information are recorded;

[0092] According to the type, spatial reference system, time stamp and data source identifier of each of the classified data written into the spatial database, a structured metadata record is generated and written into the dedicated metadata table.

[0093] In the present invention, in order to achieve unified query and semantically perceptible processing of multi-source beach spatial data, remote sensing rasters, vector forms and attribute records must be written into the spatial database in a structured manner, and metadata tables that support semantic instruction driving must be constructed simultaneously. This process requires not only conventional format cleaning of the data, but also the execution of coordinate system unification, spatial structure legitimacy verification and cross-type field consistency verification before storage. Among them, the coordinate conversion adopts a unified spatial reference system (such as WGS84) by default to ensure that the subsequent spatial index and semantic map are consistent. Figure 1 In the vector layer, for complex polygon structures, this embodiment needs to perform topological validity detection to eliminate self-intersecting error boundaries to ensure that the graphics can be correctly parsed. The timestamps of attribute records and layer data must be aligned within the specified error tolerance, otherwise they are considered to be unable to be jointly stored, thereby ensuring that semantic queries are based on the same spatiotemporal section. After the verification is passed, the data will be classified by type: remote sensing rasters will generate multi-level path indexes according to time series; vector morphologies will be merged according to layer themes; attribute records will be columnarized through primary key hash mapping to form an input format compatible with the HBase table structure.

[0094] In the process of formally writing data into the database, the application adopts a row key generation strategy with space position as the core, constructs a triple index key combining data type and time label, and realizes a space, time and semantic perceptible data organization method. Specifically, the main row key of each data is composed of a space identifier generated by Geohash (or Z-curve) encoding and a data type suffix, a collection timestamp, which ensures that the reverse index can be hit through the spatial position when querying. Classified data is written into predefined column families, among which remote sensing image data is written into the "geo-raster" family, vector data is written into the "geo-vector" family, and attribute data is written into the "attr" family. Each batch import operation automatically records the batch number and import status, supports failure rollback and result tracking. The embodiment generates corresponding metadata records in real time while writing, and the record fields include data type, spatial reference system code, bounding rectangle boundary, main field list and its unit, data time label and data source identifier, and are written into an independent metadata table. This metadata table not only serves as the field perception basis of the semantic analysis module, but also supports dynamic field matching, unit conversion and spatio-temporal range reasoning under semantic driving, and is an indispensable bridge link in the application. This mechanism ensures the semantic closed loop between unstructured spatial data and structured query intent, and is the underlying guarantee for supporting spatial semantic reasoning of large language model Agent.

[0095] Preferably, a natural language or voice query request from a user terminal is received and routed to a large language model Agent residing in the distributed storage-computing environment, including:

[0096] The user's query content is received by the user terminal through a graphical user interface or a voice input interface, and the query content is a natural language sentence in text form or a text string converted by a voice recognition module;

[0097] The query content is preprocessed and formatted, and the time expression, geographic area, and analysis index keywords in the query content are extracted, and it is verified whether the query content meets the semantic analysis requirements according to the keywords;

[0098] The query content that meets the requirements is encapsulated as a structured query request; the query request packet includes the original query text, semantic label annotation results, user identification, and query timestamp;

[0099] The query request is routed and sent to the large language model Agent residing in the distributed storage-computing environment through a task scheduling interface in the distributed storage-computing environment.

[0100] Preferably, the query request is subjected to entity extraction, intent recognition and structured conversion by the large language model Agent to generate machine-readable spatio-temporal query and analysis instructions, including:

[0101] The natural language query text in the query request is subjected to word segmentation, part-of-speech tagging and syntactic analysis to extract semantic elements of geographic location, time range, target attribute and operation intent;

[0102] The semantic elements are analyzed based on a pre-trained intent classification model to identify the task type corresponding to the query request; the task type includes statistics, comparison, prediction, trend analysis or spatial relationship judgment;

[0103] The extracted semantic elements are mapped to field names and data table structures in the spatial database, and an intermediate instruction template containing field names, operators, filtering conditions and spatial ranges is generated;

[0104] The intermediate query instruction template is parsed and formatted to generate the structured spatio-temporal query and analysis instructions; the spatio-temporal query and analysis instructions are expressed in structured query language statements, object-oriented JSON expressions or graph computing task description syntax.

[0105] In the present application, in order to realize efficient semantic docking between user query and large language model, a set of query request construction mechanism with semantic constraint and task driving ability needs to be executed after natural language or voice input. This mechanism is not only to simply forward the user input to the large model, but also to introduce pre-processing and structure packaging links on the terminal side or access layer to analyze and verify the legality of the user input content. The user inputs the query content through a graphical user interface or a voice input interface, and after the text type is recognized, the content is sent to the pre-processing module. The module performs standardized preprocessing on the natural language text, extracts the geographic area (such as "the first sea bathing place in Qingdao in the past five years"), the time range (such as "from 2018 to now") and the analysis attribute keywords (such as "beach width", "number of tourists", "water quality type", etc.), and judges whether the input meets the minimum semantic element combination requirement. Only when the content has the three core elements of geographic entity, time and space limitation and analysis dimension at the same time, the query content will be converted into a structured query request package, and the semantic label annotation result, user identification, query timestamp and other meta information will be attached to route to the large language model Agent processing end in a unified format.

[0106] After receiving the structured query request, the large language model agent first identifies the elements and enhances the semantic labels of the input content through its embedded natural language understanding capability, maps the time expressions, geographic entities and attribute categories involved in the query into a parameter structure that can correspond one by one with the database fields. Subsequently, the embodiment calls the built-in intent classification module, divides the query into statistical query, comparative analysis, trend identification or spatial relationship analysis according to the identified keyword combination and sentence pattern, and further refines it into "sum", "average", "frequency statistics" and other specific sub-tasks. On this basis, the large model executes the structured conversion logic to map the semantic elements in natural language to specific fields and table names in the spatial database, and automatically generates machine-readable query statements containing field names, operators, filtering conditions, and spatio-temporal ranges. If the input involves a spatial range (such as "beach section within 500 meters from the main entrance"), the embodiment will call the geographic spatial extension tool to generate spatial clipping conditions and embed them in the query logic. Finally, the optimized query statement is sent to the backend execution module as a complete analysis instruction, ensuring that the query intent understood by the model can be accurately and efficiently parsed and responded by the spatial database, building a closed-loop path from natural language to spatio-temporal query tasks. This mechanism is one of the key implementations that distinguish the present invention from traditional database calling methods, directly supporting the core goal of "unstructured input driving structured calculation".

[0107] Figure 4 The core mechanism of the natural language query processing flow in the present invention is demonstrated, and the whole process from the user inputting query content on the terminal, through query content receiving, semantic preprocessing, structured packaging, to the large language model Agent completing element identification, intent classification and instruction generation, and finally submitting the generated spatio-temporal query and analysis instruction to the spatial database for execution is fully described. The connection between the modules in the figure clearly represents the logical chain of input-analysis-execution, highlighting the automatic conversion path from unstructured query intent to structured spatial task realized by the model driving in the present invention, which is the key architecture diagram supporting the natural language spatial intelligent service in the embodiment.

[0108] Preferably, according to the spatio-temporal query and analysis instruction, cross-source data matching, spatio-temporal filtering, aggregation statistics, spatial clipping and numerical filtering are completed in the spatial database to obtain the analysis result, including:

[0109] According to the field conditions, time range and space range in the spatio-temporal query and analysis instruction, the field index and spatial index functions of the spatial database are called to determine the original data records matching the instruction;

[0110] Performing matching and association operations on data from different data sources in the original data records according to the primary key field, time field, or spatial location field set in the spatiotemporal query and analysis instructions to achieve cross-source data matching;

[0111] After completing the cross-source data matching, a filtering operation is performed according to the time range and spatial boundary conditions set by the spatiotemporal query and analysis instructions, and only data records that meet the conditions are retained, completing the spatiotemporal filtering process and obtaining the filtered data records;

[0112] Performing aggregate statistics on the filtered data records according to the aggregate fields and statistical methods set in the spatiotemporal query and analysis instructions to obtain aggregate statistical results;

[0113] Based on the aggregated statistical results, spatial clipping is performed according to the spatial region boundaries specified in the spatiotemporal query and analysis instructions to obtain the clipping results; the clipping methods of the spatial clipping include administrative area limitation, grid division and buffer zone clipping;

[0114] A numerical filtering operation is performed on the clipping result according to the numerical threshold or Boolean logic condition in the spatiotemporal query and analysis instruction to obtain the analysis result.

[0115] In the present invention, after receiving the spatiotemporal query and analysis instruction generated by the large language model agent, this embodiment uses the instruction as the task execution entry, and by parsing the field screening conditions, time range and spatial range information therein, calls the field index and spatial index module in the spatial database for joint retrieval, and quickly locates data records that meet the semantic query intent. In order to support the fusion processing of multi-source heterogeneous data, this embodiment introduces a cross-source matching mechanism, and performs matching and association logic on records from different data tables or data sources based on uniformly set primary key fields, time fields or spatial location fields. This logic supports one-to-one, one-to-many, many-to-many and other data mapping forms, and can realize the automatic correspondence between remote sensing images and field monitoring data, as well as the spatial connection of vector boundaries and statistical attributes. Through this mechanism, the original data can be dynamically constructed into a logically unified and computable data view before the task is executed, thereby ensuring the integrity and adaptability of the spatiotemporal query logic.

[0116] After the cross-source data matching is completed, the embodiment filters the associated data records according to the time range and the space boundary condition set in the analysis instruction, using the clipping operator and the interval filtering mechanism of the database, and only retains a data subset satisfying the space-time constraint. On this basis, according to the aggregation field and the statistical method (including but not limited to average, sum, maximum and minimum value, frequency count, etc.), the aggregation calculation is performed to generate the intermediate statistical result for spatial analysis. Subsequently, the embodiment performs spatial clipping on the aggregation result according to the clipping boundary parameters specified in the analysis instruction, using the administrative area limitation, the regular grid division or the buffer clipping strategy. The spatial clipping operation not only ensures that the analysis result has a clear spatial boundary attribute, but also provides a basis for graphical rendering in the multi-modal output stage. Finally, the embodiment performs numerical filtering operation on the clipped result according to the preset numerical threshold or Boolean logic condition in the analysis instruction, and removes invalid data points or statistical outliers that do not meet the query target, and outputs the structured analysis result. The execution process is the core link of converting the natural language analysis result into a spatial database computable instruction, embodies the integrated task processing mechanism of "semantic driving-data fusion-space calculation-result filtering" in the present application, and is an indispensable calculation link for realizing the spatial intelligent query of the embodiment.

[0117] Preferably, the analysis result is output in the form of a dynamic map, a statistical chart and a natural language report in a multi-modal form, including:

[0118] Based on the geographic coordinates, the space boundary and the index field contained in the analysis result, a dynamic map including layer superposition, color coding and zoom interaction is drawn through a map rendering engine;

[0119] According to the statistical field and the time sequence in the analysis result, a data visualization component is called to generate a two-dimensional statistical chart;

[0120] The structured field and the semantic label in the analysis result are input into a natural language generation template to generate an automatic text report describing the geographic features, the statistical profile and the spatio-temporal change trend of the query target;

[0121] The dynamic map, the two-dimensional statistical chart and the automatic text report are integrated into an interactive interface, and switching views, viewing layer details or exporting result files are supported.

[0122] Preferably, the two-dimensional statistical chart includes a line chart, a column chart, a pie chart and a heat map.

[0123] In the present application, to realize the intuitive understanding and multi-dimensional interaction of the user to the analysis result, the embodiment constructs a multi-modal output mechanism based on geographic coordinates, attribute indicators and semantic labels. After completing the generation of the structured analysis result, the embodiment first parses the spatial boundary and geographic coordinate field in the analysis result, and organizes the spatial rendering order according to the layer classification information, calls the map rendering engine to draw a dynamic map in the form of layer superposition, color coding and zoom interaction. The map rendering not only shows the geographic range of the beach and the spatial distribution state of the analysis indicators, but also supports the click response of a single spatial unit and the attribute value linkage query, so that the user can visually browse and target in the spatial range. At the same time, according to the time series field and statistical indicators in the analysis result, the embodiment dynamically selects the appropriate chart type, calls the graphic visualization engine to generate two-dimensional charts including line charts, column charts, pie charts and heat maps, and displays the distribution trend of the selected indicators in the time or spatial dimension. The data refresh logic is automatically associated during the chart generation process, ensuring that the graphics are updated in real time when the user switches different indicators or regions, improving the visual decision-making efficiency.

[0124] In addition to graphical output, the embodiment also introduces an automatic report generation mechanism based on semantic label driving. After the fields in the analysis result and their corresponding labels are structurally matched in the pre-defined template, the embodiment calls the natural language generation module to convert the indicator values, trend direction, comparison results and other contents into descriptive and explanatory natural language texts, and outputs an analysis report including the attribute profile of the target geographic object, the change trend of the key indicators, the abnormal point prompt and other contents. The report supports synchronous linkage with the map layer and the chart module on the page side, and the user can automatically locate the corresponding paragraph in the report by clicking the spatial unit. Finally, the embodiment integrates the above dynamic map, statistical chart and natural language report into the front-end interactive interface, and provides functions such as view switching, layer hiding, exporting PDF or GeoJSON, so that the analysis result can be freely switched and expressed in the spatial, graphical and semantic dimensions. This module is a direct embodiment of the "readable, visual and explainable" intelligent service result of the present application, supporting the complete closed-loop presentation of the model analysis result on the user side, and constituting a key link of human-computer interaction from semantic query to intelligent feedback.

[0125] Figure 5The workflow in the system is composed of five stages of data access, semantic understanding, instruction generation, task execution and result output. The user submits a natural language or voice query request through the terminal interface, and the system routes the request to the large language model Agent in the distributed environment. The Agent is responsible for performing element identification and intent classification, and combining the field information in the metadata table to complete the query structuring. The structured spatio-temporal query and analysis instruction is submitted to the spatial database to start the matching, filtering, aggregation and spatial clipping calculation process of cross-source data, and finally form the structured analysis result. The analysis result is sent to the multi-modal output module to generate output content in three ways of map rendering, chart drawing and natural language report in parallel. Figure 4 The system directly reflects the running mechanism of the system in the path of "natural language-structured understanding-parallel computing-multidimensional expression", which is the overall framework of the intelligent spatial question and answer service realized by the system.

[0126] As shown in Figure 6 , the tool chain module receives natural language query text from the user side, first performs word segmentation, part-of-speech tagging and keyword extraction through the semantic preprocessing submodule to obtain preliminary semantic elements. Then, in the intent classification stage, the system calls the fine-tuned large model to perform task type discrimination, classifies the query content into statistical, comparative, trend analysis and other query targets, and further refines it into "sum", "count", "spatial filtering" and other operation instructions. Then, the tool chain calls the field mapping module to map the identified query elements to the specific fields and data table structure in the database, and constructs an intermediate query expression template with field name, filtering condition, spatial boundary and other contents. After syntax analysis and scheduling adaptation processing, the template generates the final spatio-temporal query and analysis instruction that can be recognized and executed by the database.

[0127] Corresponding to the above method, the embodiment also provides a beach big data geographic spatial intelligent application service method based on a large model Agent, comprising:

[0128] A distributed environment construction unit is configured to construct a distributed storage-computing environment for beach multi-source spatio-temporal data; the distributed storage-computing environment comprises a software and hardware platform composed of a parallel computing node cluster and a column-oriented spatial database, and a large language model Agent;

[0129] A data access and standardization storage unit is configured to import cleaned and standardized remote sensing raster, vector and attribute records into the spatial database, and generate a metadata table of record data type, spatial reference system and update time;

[0130] A query request receiving and routing unit is configured to receive a natural language or voice query request from a user terminal, and route the query request to a large language model Agent residing in the distributed storage-computing environment.

[0131] a semantic analysis and instruction generation unit configured to perform entity extraction, intent recognition and structured conversion on the query request by the large language model agent, and generate machine-readable spatio-temporal query and analysis instructions;

[0132] a spatial computing and analysis execution unit configured to perform cross-source data matching, spatio-temporal filtering, aggregation statistics, spatial clipping and numerical filtering in the spatial database according to the spatio-temporal query and analysis instructions, and obtain analysis results;

[0133] a multi-modal result presentation unit configured to output the analysis results in the form of dynamic maps, statistical charts and natural language reports.

[0134] The beneficial effects of the present application are as follows:

[0135] The present application proposes an intelligent query method combining a large language model and a distributed spatial computing framework, for the first time introducing a natural language interaction mechanism into beach multi-source geographic data application services, realizing a closed-loop processing process from user semantic input to structured spatio-temporal instruction generation, significantly reducing the professional GIS operation threshold, enabling non-professional users to quickly obtain spatial analysis results through natural language, and improving the human-computer interaction intelligence level and service universality of the system.

[0136] By constructing a distributed storage-computing environment integrating a large language model agent, the present application realizes unified storage and structured management of remote sensing raster, vector boundary and attribute data, and supports semantic-driven field analysis, index mapping and spatial clipping operations. Compared with the traditional multi-system splicing method, the present application provides an integrated path from query understanding, instruction construction to computing scheduling, avoiding data migration, semantic inconsistency and other problems, greatly improving processing efficiency and system stability.

[0137] The present application adopts a Geohash spatial encoding row key and semantic metadata table joint indexing mechanism, enabling the large language model to dynamically perceive field meaning, time range and spatial structure when analyzing user intent, thereby accurately generating spatio-temporal query and analysis instructions that meet database execution requirements, ensuring high consistency between query results and user semantic intent. This structural and semantic collaborative design significantly improves query accuracy and system robustness.

[0138] In terms of analysis result output, the present application constructs a multi-modal expression mechanism, supporting synchronous rendering and presentation of structured results in the form of dynamic maps, statistical charts and natural language analysis reports, meeting user needs in multiple scenarios such as visual analysis, trend judgment and report generation, expanding the explainability and application range of geographic information systems, and having good promotion value and industrial landing prospects.

[0139] The various embodiments described in this specification are presented for the purpose of illustrating the principles of the present application and its best mode of operation. Each of the embodiments described in this specification has been provided for the purpose of illustration only and the same or similar parts of each embodiment are mutually referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple and the relevant parts are referred to the description of the method.

[0140] The principles and implementations of the present application are described herein by using specific examples. The above description of the embodiments is only used to help understand the method of the present application and its core idea. Meanwhile, for those skilled in the art, the specific implementation and application range of the present application can be changed according to the idea of the present application. In summary, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A beach big data geospatial intelligent application service method based on a large model agent, characterized in that: include: Build a distributed storage and computing environment for multi-source spatiotemporal data on beaches; The distributed storage-computing environment includes a hardware and software platform consisting of a parallel computing node cluster and a column-oriented spatial database, as well as a large language model agent; Importing the cleaned and standardized remote sensing raster, vector morphology and attribute records into the spatial database, and generating a metadata table that records data type, spatial reference system and update time; Receiving a natural language or voice query request from a user terminal and routing the query request to a large language model agent residing in the distributed storage-computing environment; The large language model agent performs entity extraction, intent recognition, and structural conversion on the query request to generate machine-readable spatiotemporal query and analysis instructions; According to the spatiotemporal query and analysis instructions, cross-source data matching, spatiotemporal screening, aggregate statistics, spatial clipping and numerical filtering are completed in the spatial database to obtain analysis results; The analysis results are output in multimodal forms such as dynamic maps, statistical charts and natural language reports.

2. The beach big data geospatial intelligent application service method based on large model agent according to claim 1 is characterized in that: Build a distributed storage and computing environment for multi-source spatiotemporal data on beaches, including: Launching a parallel computing cluster consisting of multiple physical computing nodes; establishing communication links between the computing nodes through an internet network, and deploying a memory-driven elastic computing framework to support the distributed execution of spatial computing tasks; Deploying the components of the spatial database for column storage in the computing nodes, constructing data row keys using a spatial index mechanism based on latitude and longitude encoding, and integrating the spatial database with the parallel computing cluster to form a unified storage-computing integrated platform; Integrating the large language model agent into the storage-computing integrated platform; deploying the large language model agent in an independent computing container and configuring a semantic parsing interface, an inference executor, and a model parameter access channel to support the conversion of natural language into spatial analysis tasks; Initialize the system configuration file, set the spatial data access interface, model service call path and analysis module scheduling rules, and publish the application interface address that can be used for service calls after initialization is completed.

3. The beach big data geospatial intelligent application service method based on large model agent according to claim 1 is characterized in that: Import the cleaned and standardized remote sensing raster, vector morphology and attribute records into the spatial database, and generate a metadata table that records data type, spatial reference system and update time, including: Performing unified coordinate system conversion on beach remote sensing images, vector layers, and structured attribute data from multiple source platforms, and verifying spatial geometry validity, field integrity, and timestamp consistency to obtain verified data files; the data files include the remote sensing raster, the vector form, and the attribute records; The verified data files are classified according to data types to obtain classified data; wherein the remote sensing grid is indexed by path according to time series, the vector form is classified by layer name, and the attribute record is tabulated according to primary key mapping; Using a row key strategy based on a geocoding mechanism, each of the classified data is written into a corresponding column family or table space of the spatial database, and the import status and batch information are recorded; According to the type, spatial reference system, time stamp and data source identifier of each of the classified data written into the spatial database, a structured metadata record is generated and written into the dedicated metadata table.

4. The beach big data geospatial intelligent application service method based on large model agent according to claim 1 is characterized in that: Receiving a natural language or voice query request from a user terminal and routing the query request to a large language model agent residing in the distributed storage-computing environment, including: Receive a user's query content on the user terminal through a graphical user interface or a voice input interface, wherein the query content is a natural language sentence in text form, or a text string transcribed by a voice recognition module; Performing formatting preprocessing on the query content, extracting keywords of time expression, geographical area, and analysis index in the query content, and verifying whether the query content meets the semantic parsing requirements based on the keywords; Encapsulate the query content that meets the requirements into the structured query request; the query request package includes the original query text, semantic tagging results, user ID and query timestamp; The query request is routed and sent to the large language model agent residing in the distributed storage-computing environment through the task scheduling interface in the distributed storage-computing environment.

5. The beach big data geospatial intelligent application service method based on large model agent according to claim 1 is characterized in that: The large language model agent performs entity extraction, intent recognition, and structural conversion on the query request to generate machine-readable spatiotemporal query and analysis instructions, including: Performing word segmentation, part-of-speech tagging, and syntactic analysis on the natural language query text in the query request to extract semantic elements of geographic location, time range, target attributes, and operation intention; Analyze the semantic elements based on a pre-trained intent classification model to identify the task type corresponding to the query request; the task type includes statistics, comparison, prediction, trend analysis, or spatial relationship judgment; Mapping the extracted semantic elements to the field names and data table structure in the spatial database, and generating an intermediate instruction template including the field name, operator, filter condition and spatial range; The intermediate query instruction template is parsed and formatted to generate the structured spatiotemporal query and analysis instructions; the spatiotemporal query and analysis instructions are expressed using structured query language statements, object-oriented JSON expressions or graph computing task description syntax.

6. The beach big data geospatial intelligent application service method based on large model agent according to claim 1 is characterized in that: According to the spatiotemporal query and analysis instructions, cross-source data matching, spatiotemporal screening, aggregate statistics, spatial clipping and numerical filtering are completed in the spatial database to obtain analysis results, including: According to the field conditions, time range and space range in the spatiotemporal query and analysis instructions, calling the field index and space index functions of the spatial database to determine the original data records that match the instructions; Performing matching and association operations on data from different data sources in the original data records according to the primary key field, time field, or spatial location field set in the spatiotemporal query and analysis instructions to achieve cross-source data matching; After completing the cross-source data matching, a filtering operation is performed according to the time range and spatial boundary conditions set by the spatiotemporal query and analysis instructions, and only data records that meet the conditions are retained, completing the spatiotemporal filtering process and obtaining the filtered data records; Performing aggregate statistics on the filtered data records according to the aggregate fields and statistical methods set in the spatiotemporal query and analysis instructions to obtain aggregate statistical results; Based on the aggregated statistical results, spatial clipping is performed according to the spatial region boundaries specified in the spatiotemporal query and analysis instructions to obtain the clipping results; the clipping methods of the spatial clipping include administrative area limitation, grid division and buffer zone clipping; A numerical filtering operation is performed on the clipping result according to the numerical threshold or Boolean logic condition in the spatiotemporal query and analysis instruction to obtain the analysis result.

7. The beach big data geospatial intelligent application service method based on large model agent according to claim 1 is characterized in that: Output analysis results in multimodal forms such as dynamic maps, statistical charts, and natural language reports, including: Based on the geographic coordinates, spatial boundaries and indicator fields included in the analysis results, a map rendering engine is used to draw a dynamic map including layer overlays, color coding and zoom interaction; Based on the statistical fields and time series in the analysis results, calling the data visualization component to generate a two-dimensional statistical chart; Input the structured fields and semantic tags in the analysis results into a natural language generation template to generate an automatic text report describing the geographic characteristics, statistical overview, and spatiotemporal change trends of the query target; The dynamic map, the two-dimensional statistical chart and the automatic text report are integrated into an interactive interface, and the user is supported to switch views, view layer details or export result files.

8. The beach big data geospatial intelligent application service method based on large model agent according to claim 7 is characterized in that: The two-dimensional statistical charts include: line charts, bar charts, pie charts and heat maps.

9. A beach big data geospatial intelligent application service method based on a large model agent, characterized in that: include: Distributed environment construction unit, used to build a distributed storage and computing environment for multi-source spatiotemporal data of beaches; The distributed storage-computing environment includes a hardware and software platform consisting of a parallel computing node cluster and a column-oriented spatial database, as well as a large language model agent; A data access and standardized storage unit is used to import cleaned and standardized remote sensing raster, vector form and attribute records into the spatial database, and generate a metadata table that records data type, spatial reference system and update time; A query request receiving and routing unit, configured to receive a natural language or voice query request from a user terminal and route the query request to a large language model agent residing in the distributed storage-computing environment; A semantic parsing and instruction generation unit, configured to perform entity extraction, intent recognition, and structural conversion on the query request through the large language model agent, and generate machine-readable spatiotemporal query and analysis instructions; A spatial calculation and analysis execution unit, configured to perform cross-source data matching, spatiotemporal screening, aggregate statistics, spatial clipping, and numerical filtering in the spatial database according to the spatiotemporal query and analysis instructions to obtain analysis results; The multimodal result presentation unit is used to output the analysis results in a multimodal form of dynamic maps, statistical charts and natural language reports.

Images