Knowledge graph-based marine survey historical data correlation analysis and retrieval system

Abstract

The application relates to the field of marine information technology and data processing, and discloses a marine survey historical data correlation analysis and retrieval system based on a knowledge graph, which comprises a self-adaptive relation kernel compiling engine, a user query intention, a field semantic graph, a dynamically selected and compiled relation kernel encapsulating a field algorithm, a generated relation discovery execution plan, a dynamic hypergraph construction module executing the plan, a calculation of generated relation instances representing high-order correlations as hyperedges by calling the relation kernel to act on original data, a dynamically constructed dynamic hypergraph with space-time data field nodes as vertices and the relation instances as hyperedges, and a pattern matching on the dynamic hypergraph to retrieve complex correlation patterns and provide traceable analysis results. The application realizes the calculation-driven dynamic construction of data relations, can efficiently and flexibly discover and analyze high-order correlations in marine space-time data.

Description

Technical Field

[0001] This invention relates to the field of marine information technology and data processing, specifically to a knowledge graph-based system for the association analysis and retrieval of historical marine survey data. Background Technology

[0002] Marine science research and applications heavily rely on the analysis and mining of massive, multi-source, and heterogeneous historical survey data. This data, including satellite remote sensing, buoy data, ship observations, and numerical simulation results, exhibits typical spatiotemporal characteristics, high dimensionality, and dynamic changes. The core challenge facing the field of marine data science is how to effectively extract entities from this vast amount of data, discover the relationships between entities, and construct a knowledge system capable of supporting complex queries.

[0003] Currently, relational databases, geographic information systems (GIS), or knowledge graphs are commonly used to process this type of data. While relational databases provide structured storage, their rigid table structures are inflexible when expressing complex, many-to-many relationships between marine phenomena. This often requires numerous complex join operations, resulting in low query efficiency and difficulty in intuitively reflecting the intrinsic connections between phenomena. Geographic information systems, on the other hand, focus on the spatial attributes and geometric relationships of data, but have limited ability to reveal non-spatial semantic relationships based on physical processes or statistical laws.

[0004] To better represent the interconnections between data, knowledge graph technology has been introduced into marine data management. Existing knowledge graph construction methods typically rely on a preprocessing workflow, involving data extraction, transformation, and loading to import identified entities and predefined relationship types into a graph database. Knowledge graphs constructed in this way are inherently static. When new observational data continuously flows in or data is updated, the graph cannot reflect these changes in real time, resulting in a delay and disconnect between its content and the real-time state of the underlying data.

[0005] Furthermore, traditional knowledge graphs are mostly based on simple graph models, where a binary relationship is represented by an edge connecting two nodes. However, many important phenomena in the ocean, such as a front formed by the interaction of multiple water masses, or a red tide caused by the combined effects of specific wind and current fields, are essentially high-order relationships involving multiple entities. In simple graphs, these high-order relationships are difficult to express directly and clearly, usually requiring the introduction of additional intermediate nodes for workarounds. This not only increases the complexity of the graph structure but also makes querying such complex patterns extremely cumbersome and unintuitive. Therefore, existing technologies still have significant limitations in the dynamic discovery of relationships, the effective expression of high-order relationships, and the flexible querying of complex association patterns. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a knowledge graph-based system for the association analysis and retrieval of historical marine survey data. This system solves the problems of existing data analysis methods, which rely on static knowledge bases and simple graph structures, making it difficult to discover relationships on demand for dynamically changing marine data, and are unable to effectively express and query high-order associations of multiple entities.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a knowledge graph-based marine survey historical data association analysis and retrieval system, comprising: This invention provides a knowledge graph-based system and method for the association analysis and retrieval of historical marine survey data. The core technical solution of this invention lies in constructing a dynamic relationship discovery mechanism driven by query intent and guided by domain knowledge. This mechanism does not rely on predefined static relationship links, but rather adaptively orchestrates and executes relationship kernels encapsulated with domain algorithms, directly acting on the original high-dimensional data. It dynamically generates hypergraph structures representing high-order associations in a computationally driven manner, thereby achieving flexible, efficient retrieval and traceable analysis of complex association patterns in marine data.

[0008] The first aspect of this invention provides a knowledge graph-based system for the association analysis and retrieval of historical marine survey data. This system includes: a data processing and spatiotemporal data field construction module, configured to process and encapsulate raw marine survey data into spatiotemporal data field nodes carrying the original high-dimensional dataset; a domain semantic graph management module, configured to store and manage a domain semantic graph defining the associations between marine domain concepts, relation types, methods, and data types; a relation kernel function library module, configured to provide a set of relation kernels encapsulating specific domain algorithms; and an adaptive relation kernel orchestration engine module, configured to parse user queries into... The system constructs a query intent and, based on the structured query intent and the domain knowledge in the domain semantic graph management module, selects and orchestrates relation kernels from the relation kernel function library module to generate a relation discovery execution plan; a dynamic hypergraph construction and maintenance module is configured to execute the relation discovery execution plan, and generates relation instances as hyperedges by calling relation kernels to dynamically construct and maintain a dynamic hypergraph with spatiotemporal data field nodes as vertices and relation instances as hyperedges; and a patterned query and analysis module is configured to execute patterned queries on the dynamic hypergraph to retrieve subgraph instances that satisfy user-defined query patterns.

[0009] In one specific implementation, the relation kernel is a computational function that directly acts on the original high-dimensional dataset encapsulated within one or more spatiotemporal data field nodes. By executing a preset domain algorithm, it calculates to discover and quantify specific associations between the spatiotemporal data field nodes.

[0010] In one specific implementation, the relationship instance is a structured data object generated after the relationship kernel is successfully computed. The relationship instance includes: a set of associated nodes, used to identify all spatiotemporal data field nodes constituting the relationship; a relationship kernel type identifier, used to specify the specific relationship kernel that generated the relationship instance, ensuring the traceability of the relationship; and a set of quantified attributes, used to store the specific numerical results calculated by the relationship kernel, such as correlation coefficients, gradient values, or similarity scores, providing a quantitative basis for the strength and characteristics of the relationship.

[0011] In one specific implementation, the query intent parsing unit of the adaptive relational kernel orchestration engine module transforms the user query into a formalized query intent. : ; in, For the set of vertices of the target concept, For the target relation type vertex set, For the required set of spatiotemporal data field types, , and These constraints are time, space, and attribute constraints, respectively. Subsequently, its relation core intelligent orchestration unit selects and combines relation cores based on this intent and the domain semantic graph, and the execution plan generation unit generates a relation discovery execution plan. The plan is defined as follows: ( ); in, This is a set of relational kernel computation tasks. Let be the set of directed edges representing the data dependencies between tasks. Each computational task... Structured as follows: ; in, The identifier for the relation kernel to be invoked. For the input spatiotemporal data field node set, This is the set of operating parameters for the relational kernel.

[0012] In one specific implementation, the dynamic hypergraph construction and maintenance module executes the relation discovery execution plan, uses the calculated relation instances as hyperedges, and dynamically constructs the dynamic hypergraph. : ( ); in, It is a set of spatiotemporal data field nodes. This is the set of hyperedges that constitute relation instances. When a spatiotemporal data field node associated with a relation instance is updated, this module triggers a re-evaluation of that relation instance to update the topology and attributes of the hypergraph, ensuring its timeliness.

[0013] In one specific implementation, the pattern-based query and analysis module formalizes the user-defined query pattern into a query template hypergraph. : ; in, To query the set of vertex variables, To query the set of hyperedge variables, and These are the constraints applied to vertices and hyperedges, respectively. The hypergraph pattern matching algorithm in this module filters the candidate set of the dynamic hypergraph and performs a backtracking-based subgraph isomorphic search on the filtered candidate set to find all matching subgraph instances. The module's result presentation and associative tracing unit, while presenting the matching results, provides users with complete tracing information for each relation instance, including its source relation kernel type, input data nodes, and quantification attributes, achieving transparent tracing from high-level knowledge to the original data.

[0014] A second aspect of this invention provides a method for retrieving historical marine survey data based on knowledge graphs, the method comprising the following steps: S1: Receive user queries and parse the user queries into structured query intents; S2: Based on the structured query intent and the preset domain semantic graph, select and arrange relation kernels from the relation kernel function library to generate a relation discovery execution plan; S3: Execute the relationship discovery execution plan, and dynamically construct a dynamic hypergraph with spatiotemporal data field nodes as vertices and relationship instances as hyperedges by calling the relationship kernel to calculate and generate relationship instances. S4: Receive the user-defined query pattern and perform pattern matching on the dynamic hypergraph to find subgraph instances that satisfy the query pattern as query results; S5: Present the query results and provide association tracing information for the relational instances in the results.

[0015] This invention provides a knowledge graph-based system for the association analysis and retrieval of historical marine survey data. It offers the following advantages: 1. This invention establishes an adaptive relation kernel orchestration engine. This engine dynamically selects and organizes relation kernel functions to generate an execution plan based on the user's query intent and the domain semantic graph, enabling on-demand computation and dynamic discovery of data relationships. This computation-driven model eliminates the reliance on static, predefined knowledge graphs, tightly coupling knowledge generation with the user's analytical intent. This ensures the real-time nature and relevance of discovered relationships, overcoming the problems of knowledge solidification and delayed updates in traditional methods.

[0016] 2. This invention constructs a dynamic hypergraph by using relation instances generated by relation kernel calculation as hyperedges connecting multiple spatiotemporal data field nodes. This hypergraph can directly and clearly express the high-order multi-entity relationships that are prevalent in marine phenomena. Compared to the traditional simple graph approach that uses intermediate nodes for workarounds, the hypergraph structure is more concise and closer to physical reality in terms of model, making the definition and matching of queries for complex relationship patterns more intuitive and efficient, and significantly improving the system's ability to analyze and express complex marine phenomena.

[0017] 3. This invention records the relation kernel type, input data nodes, and quantification attributes of the relation instance, and provides a link tracing function, achieving complete traceability from high-level knowledge to raw data. Users can not only see the link results, but also clearly understand how the link was calculated using a specific algorithm and based on which raw data. This ensures the transparency, interpretability, and reproducibility of the entire analysis process, greatly enhancing the credibility of the analysis results. Attached Figure Description

[0018] Figure 1 This is a system architecture diagram of the present invention; Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example: Please see the appendix Figure 1 This invention provides a knowledge graph-based system for the association analysis and retrieval of historical marine survey data, including: This invention provides a system for correlation analysis and retrieval of historical marine survey data. The system may include: a data processing and spatiotemporal data field construction module 100, a domain semantic graph management module 200, a relation kernel function library module 300, an adaptive relation kernel orchestration engine module 400, a dynamic hypergraph construction and maintenance module 500, and a pattern-based query and analysis module 600.

[0021] Specifically, the data processing and spatiotemporal data field construction module 100 is responsible for receiving and standardizing raw marine survey data, encapsulating it into structured spatiotemporal data field nodes. These STDF nodes constitute the basic elements of the hypergraph.

[0022] The domain semantic graph management module 200 stores and manages meta-knowledge in the marine science domain, including concepts, analysis methods, data types, and their interrelationships, providing semantic support for the system's intelligent decision-making. The relation kernel function library module 300 contains a series of predefined, extensible computation functions, each designed to discover and quantify specific types of associations from input STDF nodes and instantiate these associations as hyperedges.

[0023] The adaptive relation kernel orchestration engine module 400 is the intelligent core of the system. Based on the high-level query intent proposed by the user, combined with the knowledge in the domain semantic graph management module 200 and the currently available STDF node information, it intelligently selects, combines and optimizes the relation kernels in the relation kernel function library module 300 to generate an efficient relation discovery execution plan.

[0024] The dynamic hypergraph construction and maintenance module 500 is responsible for calling relation kernel functions based on the execution plan generated by the adaptive relation kernel orchestration engine module 400, and dynamically adding the discovered relationships to the hypergraph in the form of hyperedges. This module is also responsible for the continuous updating and maintenance of the hypergraph to reflect changes in data and relationships.

[0025] The pattern-based query and analysis module 600 provides a user interface that allows users to define and submit high-level query patterns. This module performs pattern matching on the hypergraph maintained by the dynamic hypergraph construction and maintenance module 500 to discover complex related events and phenomena that match the user-specified patterns, and presents the analysis results to the user.

[0026] Through the organic integration and collaborative work of the above modules, this system realizes the transformation from raw marine survey data to high-order related knowledge, supporting marine science researchers to conduct in-depth exploratory analysis and knowledge discovery.

[0027] This invention provides a method for correlation analysis and retrieval of historical marine survey data, comprising the following steps: S1, Data Processing and Spatiotemporal Data Field Construction: Receive and standardize heterogeneous raw marine survey data, divide the data according to its spatiotemporal characteristics and type, and encapsulate it into structured spatiotemporal data field nodes, which serve as the basic vertices of the hypergraph.

[0028] S2, Domain Semantic Graph Initialization and Maintenance: Construct and continuously update the domain semantic graph. This DSG explicitly represents the concepts, analytical methods, data types and their interrelationships within the marine science domain, providing high-level knowledge support for intelligent relationship discovery.

[0029] S3, Relational Kernel Function Library Preparation: Develop and maintain a relational kernel function library. Each RK function in the library is designed to take one or more STDF nodes as input and, based on the raw data characteristics inside these STDF nodes, discover and quantify specific types of associations through computation.

[0030] S4, Adaptive Relation Core Orchestration: Receives the user's high-level query intent, combines domain knowledge from the DSG with currently available STDF node information, intelligently selects, combines, and parameterizes relation cores from the RK function library, and generates an optimized relation discovery execution plan.

[0031] S5, Dynamic Hypergraph Construction and Update: Based on the execution plan generated by Adaptive Relation Kernel Orchestration S4, the corresponding relation kernel functions are called to instantiate the discovered associations as hyperedges. These hyperedges connect the relevant STDF nodes and are dynamically added to the hypergraph. Simultaneously, the hypergraph's topology and hyperedge attributes are maintained according to data updates and relation changes.

[0032] S6, Pattern-Based Query and Association Analysis: Provides an interface for users to define high-level query patterns and performs pattern matching within a dynamic hypergraph. It identifies complex association events and phenomena that match user-specified patterns and visualizes the matching results, supporting users in performing association tracing analysis.

[0033] The raw data processing unit in the data processing and spatiotemporal data field construction module 100 is responsible for receiving, preprocessing and standardizing various types of historical marine survey data.

[0034] The workflow of the raw data processing unit includes the following steps: Data Ingestion and Initial Sorting: The system continuously receives raw survey data from different data interfaces or repositories. The received data is initially sorted and categorized according to its type and timestamp.

[0035] Data cleaning and denoising: Performing a series of quality control operations on the ingested data to identify, correct, or remove outliers and noise caused by observation errors, sensor malfunctions, transmission interruptions, etc. For a given data sequence... If a certain data point satisfy (in The mean of the sequence. The standard deviation of the sequence. If the threshold is met, it is marked as an anomaly. Or, if the condition is met... or (in It is the first quartile. It is the third quartile. , If the threshold is used, it is marked as an anomaly.

[0036] Format conversion and data integration: Data from different sources and formats is uniformly converted into a standard data model within the system. This involves spatial coordinate system transformation. Timestamps are standardized to Coordinated Universal Time (UTC) with consistent time precision. For data with different spatial or temporal resolutions, resampling or interpolation can be performed according to preset strategies.

[0037] Metadata extraction and standardization: Extract key metadata from the original data files, including but not limited to data acquisition time, geographical location, sensor type, measurement parameters, data resolution, data provider, and data quality identifiers.

[0038] The spatiotemporal data field node generation unit in the data processing and spatiotemporal data field construction module 100 is responsible for structuring and encapsulating the preprocessed and standardized raw marine survey data according to its inherent spatiotemporal characteristics to form the basic vertices of the hypergraph system—the spatiotemporal data field nodes.

[0039] The core innovation of this unit lies in its encapsulation of the raw, high-dimensional survey dataset itself as nodes in a hypergraph, rather than abstract entities or metadata summaries commonly found in traditional knowledge graphs. In this way, the graph's topology directly reflects the inherent relationships between data fields, allowing the subsequent relationship discovery process to directly address the physical essence of the data, rather than its artificial abstraction.

[0040] The generation process of spatiotemporal data field nodes includes the following steps: Spatiotemporal gridding and data slicing: The system divides the entire ocean research area into a series of discrete four-dimensional spatiotemporal units according to a preset spatial resolution and temporal granularity. Spatial partitioning can use regular grids or adaptive grids. Temporal partitioning is based on the frequency of data acquisition, observation period, or analysis requirements, dividing the time axis into continuous or discrete time segments. Each four-dimensional spatiotemporal unit... The spatiotemporal range covered by a potential STDF node is defined.

[0041] Raw dataset aggregation and encapsulation: For each defined spatiotemporal unit, the spatiotemporal data field node generation unit aggregates all processed raw survey data of the same type that fall within that unit. These data are not further abstracted or simplified, but are presented as the raw dataset. It is completely encapsulated into the corresponding STDF node.

[0042] Formal definition and metadata association of STDF nodes: The final generated STDF nodes Formalizable definition: ; in: This represents the three-dimensional extent of the STDF node in geographic space, where Longitude Latitude For depth.

[0043] This indicates the time range covered by the STDF node, where The start time, This is the end time.

[0044] It is the original high-dimensional survey dataset encapsulated within the STDF node, and its structure and content depend on the data type. It contains numerical data from all actual observations or model calculations within this spatiotemporal region.

[0045] This is a set of metadata associated with the STDF node. This metadata originates from extraction and standardization results, including data type identifiers, sensor type, raw data resolution, data quality identifiers, and some basic statistical overview information within the data field. This overview information is... Lightweight descriptions for quick filtering and prediction.

[0046] STDF Node Storage and Indexing: The generated STDF nodes are persistently stored in a distributed storage system or a spatiotemporal database. To support efficient retrieval, the system constructs a multi-dimensional index structure for each STDF node, enabling rapid searching and extraction based on spatiotemporal range, data type, or metadata attributes.

[0047] Through the above process, the system transforms massive amounts of raw marine survey data into a series of structured, self-contained STDF nodes.

[0048] The core of the Domain Semantic Graph Management Module 200 is the Domain Semantic Graph, which is an independent meta-knowledge graph used to explicitly model concepts, theories, analytical methods, their interrelationships, and applicable conditions within the field of marine science. It provides advanced semantic guidance and reasoning foundation for subsequent adaptive relation kernel orchestration engines.

[0049] A domain semantic graph It can be formally defined as a set of vertices and a set of edges, represented as: ; in: yes The set of vertices in the domain represents various entities and abstract concepts within the domain. This set contains the following four main types of vertices: Conceptual vertices. These vertices represent abstract or concrete concepts in oceanography.

[0050] Method vertices. These vertices represent the specific algorithms, models, or analytical processes used to identify, analyze, or infer oceanic concepts or relationships.

[0051] : Data type vertices. These vertices represent the required STDF data type, that is, the category of the original dataset encapsulated inside the STDF node.

[0052] : Relation type vertices. These vertices represent the semantic types of various quantified associations that the relation kernel may produce.

[0053] yes The set of directed edges in the diagram represents the semantic relationships between vertices. These edges provide structured paths for automated reasoning and intelligent orchestration.

[0054] The construction of this system employs ontology engineering methods, where domain experts formalize marine science knowledge. This enables... It is not just a data dictionary, but a knowledge system that can be understood and reasoned by machines, effectively supporting the adaptive relational kernel orchestration engine to make intelligent decisions.

[0055] The specific implementation of a semantic-driven mechanism includes the following steps: Formal mapping of query intent: The adaptive relational kernel orchestration engine module 400 receives high-level queries from user input.

[0056] Semantic guidance selection of relation kernels: After determining the target concept vertices, the adaptive relation kernel orchestration engine module 400... It performs graph traversal. Starting from the target concept vertex, it searches along predefined semantic relationship edges to find all method vertices that can identify or analyze these concepts. These were the method vertices that were found. The relation kernel functions corresponding to ) constitute the candidate relation kernel set for executing the current query task.

[0057] Data-driven feasibility verification: For each candidate relation kernel, the adaptive relation kernel orchestration engine module 400 continues in... It performs reasoning from the corresponding method vertex ( Depart, along REQUIRES DATATYPE side( ) Find the data type vertex required for the relation kernel to execute ( Then, the engine queries the currently available spatiotemporal data field nodes in the system to verify whether there are STDF nodes that meet the data type requirements within the user-spatiotemporal range. Only candidate relation kernels whose data requirements are met are retained, thus ensuring the feasibility of subsequent execution plans.

[0058] Dynamic orchestration of the analysis process: when a single relational kernel is insufficient to handle complex queries, The structure provides a basis for the combination of multiple relation kernels. The adaptive relation kernel orchestration engine module 400 utilizes... COMPOSED OF An edge or multi-step graph path reasoning process is used to construct a relation kernel calculation flow.

[0059] Context-dependent parameterization guidelines: The vertices and edges in the data can be attached with attributes, storing domain expert knowledge and experience parameters.

[0060] Through the above mechanism By transforming static domain knowledge into dynamic instructions that drive the relationship discovery process, the system can transcend fixed algorithm calling patterns and achieve semantic understanding-based, adaptive, and intelligent data analysis.

[0061] The relation kernel function library module 300 provides a set of extensible relation kernel functions. In embodiments of the present invention, a relation kernel is a computational function that encapsulates specific domain knowledge or data analysis algorithms. Its core principle lies in directly acting on the original high-dimensional dataset within one or more spatiotemporal data field nodes to discover and quantify the potential correlations between these data fields through computation.

[0062] A relational kernel It can be formally defined as a function: (3) in: This represents a specific type of relational kernel function, such as a thermal front detection kernel or a spatial co-occurrence analysis kernel.

[0063] This is the input to the relation kernel, representing a set of spatiotemporal data field nodes to be analyzed, where... ≥1.

[0064] This is a set of internal parameters associated with the relation kernel, such as the statistical significance threshold, physical model coefficients, and hyperparameters of the feature extraction algorithm. These parameters can be dynamically adjusted by the adaptive relation kernel orchestration engine module 400 according to the context.

[0065] This is the output of the relation kernel, representing a relation instance that has been successfully detected.

[0066] This indicates that the relation kernel failed to detect an association that meets the preset conditions between the input STDF nodes.

[0067] Taking a relation kernel used to detect thermal fronts as an example, its implementation principle includes the following steps: Input STDF nodes: This relation kernel receives at least two spatially adjacent STDF nodes with the data type "temperature field" as input.

[0068] Raw data extraction and computation: This relation kernel directly extracts the original high-dimensional temperature dataset encapsulated within the input STDF nodes. Subsequently, temperature gradients were calculated numerically across the spatial adjacency interfaces of these datasets. The magnitude of the temperature gradient on a two-dimensional horizontal plane... It can be represented as: ; in It's temperature. and These are spatial coordinates. The calculation is performed directly on the original data points.

[0069] Association determination: The calculated gradient value and the kernel parameters of this relationship. Preset gradient threshold A comparison is made. If a continuous region exists on the adjacent interface whose gradient value consistently exceeds a threshold, the comparison is made. If the relation kernel determines that there is a thermal front association between these input STDF nodes, then the relation kernel will determine that there is a thermal front association between them.

[0070] Output: If an association is determined to exist, the relation kernel will generate a relation instance. Output the result; if no gradient anomaly region satisfying the conditions is detected, output an empty set. .

[0071] Other types of relation kernels can encapsulate different algorithms.

[0072] When a relation kernel function successfully detects a correlation that meets preset conditions among the nodes of the input spatiotemporal data field, it generates a structured relation instance. This relation instance acts as a hyperedge in the dynamic hypergraph, used to represent and quantify a multi-entity, high-order relation, and is the core carrier for this system to realize the transformation from data to knowledge.

[0073] The generation and structure of relation instances specifically include the following steps: Formal definition of a relation instance: After the relation kernel function completes its calculation and determines the existence of an association, the system will generate a relation instance based on its calculation process and results. This relation instance It is a structured data object, which can be formally defined as: ; Related Subjects and Type Definitions: Relationship Instances Components and The basic subject and type of the association are defined.

[0074] It is a set of associated nodes, containing all spatiotemporal data field nodes that participated in this relation kernel calculation and were determined to be associated. Since this set can contain multiple nodes, this structure can directly represent higher-order associations between more than two data fields.

[0075] This is a relation kernel type identifier that specifies the particular relation kernel function that generated the relation instance. This identifier links the discovered associations to the underlying physical or statistical model, providing a basis for the interpretability of the associations.

[0076] Quantitative description of relationships: Relationship instances Components and The discovered relationships should be described objectively and quantitatively.

[0077] It is a set of quantified attributes that stores the specific numerical results calculated by the relation kernel function. This attribute set makes the relationship no longer a simple matter of existence or non-existence, but rather a precise quantification.

[0078] It is a confidence score used to quantify the reliability or statistical significance of the association.

[0079] Spatiotemporal context and semantic description of relationships: Relation instances Components and It provides the spatiotemporal context and human-understandable semantics of this association.

[0080] The spatiotemporal validity range of the association itself is defined.

[0081] It is a semantic label, such as a strong heat front or a significant positive correlation. This label is usually formed by... It is mapped and associated with the relation type vertices in the domain semantic graph, providing users with an intuitive and easy-to-understand description.

[0082] The query intent parsing unit in the adaptive relational kernel orchestration engine module 400 is responsible for converting high-level information requests input by the user into structured query intents that the system can understand and process. User queries can be in natural language or input through a structured interactive interface. The key to this unit is accurately mapping these informal expressions to the domain concepts and relationships defined in the domain semantic graph management module 200.

[0083] The process of parsing query intent includes the following steps: Query Input and Preliminary Processing: The system receives queries submitted by users. .like For natural language text, the system will perform preliminary linguistic analysis.

[0084] Semantic keyword extraction: Identifying and extracting key terms with clear domain semantics from the pre-processed query text or structured components. These terms typically point to specific marine phenomena, attributes, events, and the types of associations that the user is interested in and expects to analyze.

[0085] Alignment with concepts in the domain semantic graph: The extracted semantic keywords are precisely mapped and aligned to the corresponding vertex types in the domain semantic graph.

[0086] Concept vertex alignment: The marine phenomena or entities described in the user query will be aligned to... Concept vertices in .

[0087] Relationship type vertex alignment: The association types that users expect to discover, such as related, will be aligned to... Relationship type vertices in If the relationship type is not explicitly specified in the query, the system will infer the default relationship type based on the context or preset rules.

[0088] Data type vertex alignment: User-mentioned observed variables or data characteristics, such as marine biomass, will be aligned to... Data type vertex in .

[0089] Spatiotemporal and Attribute Constraint Extraction: In addition to core semantic elements, the query intent parsing unit also identifies and extracts various constraints contained in the query. These constraints define the scope and specific conditions of relation discovery. Specifically, these include: Time Range Constraints: These are parsed into specific... Time period. Spatial constraints: These are resolved into specific time periods. Geographic spatial scope. Data attribute constraints: These are numerical or logical restrictions on data attributes within nodes of a spatiotemporal data field. Relational attribute constraints: These are constraints on the attributes of future generated relational instances.

[0090] Construction of a formal query intent: After the above steps, all extracted and aligned information is integrated and structured to generate a formal query intent. .Should As an explicit instruction for an adaptive relational kernel orchestration engine, it can be represented as: ; in: It is the set of target concept vertices related to the user's query.

[0091] It is the set of vertices representing the target relation type that the user expects to discover.

[0092] It is a collection of spatiotemporal data field types that may be needed to complete the query.

[0093] Includes one or more time constraints, such as time periods. .

[0094] Includes one or more spatial constraints, such as geographic range. .

[0095] Includes numerical or logical constraints on data field node attributes or relation instance attributes.

[0096] The relation kernel intelligent orchestration unit in the adaptive relation kernel orchestration engine module 400 is responsible for processing the structured query intents generated by the query intent parsing unit. As input, the relation kernels in the relation kernel function library module 300 are dynamically selected, combined, and parameterized by the domain knowledge in the domain semantic graph management module 200 to generate an efficient relation discovery execution plan.

[0097] The specific process of relational kernel intelligent orchestration includes the following steps: Candidate relation kernel identification and screening: The intelligent relation kernel orchestration unit first identifies and filters candidates based on formal query intent. Target concept vertex in Relationship type vertices with target In the domain semantic graph Graph search is performed within this unit. Zhongyu Starting from the relevant concept vertices, trace along the USES or DETECTS edges to identify all possible method vertices used for detecting or analyzing these concepts. Each method vertex Corresponding to a relation kernel function Simultaneously, this unit verifies these relational kernel functions. Able to generate or with It matches the relationship type specified in the code.

[0098] Data requirements and availability verification: For each candidate relation core selected, the relation core intelligent orchestration unit further verifies... Query the corresponding method vertex in the middle Connected REQUIRES DATATYPE Edges are used to determine the data type of vertices required for the relation kernel to execute. The unit then checks the current status of the system. Spatiotemporal data field nodes available within the defined spatiotemporal range.

[0099] The combination and arrangement of relation cores: when query intent When a task is complex and requires the collaboration of multiple relation cores, the intelligent orchestration unit of the relation cores, based on... The knowledge in the data is combined into relational kernels.

[0100] Dynamic adjustment and optimization of relation kernel parameters: For selected and combined relation kernels, the relation kernel intelligent orchestration unit adjusts the parameters according to the parameters. Included attribute constraints ( )as well as Corresponding method vertex The default or recommended parameter information carried, regarding the internal parameters of the relation kernel. Perform dynamic configuration. For example, if If a correlation coefficient greater than 0.8 is specified, the internal threshold of the correlation analysis kernel will be set to 0.8. A certain method vertex in The cell also takes into account its computational complexity or applicability attributes and optimizes accordingly.

[0101] Generate a relation discovery execution plan: After the above steps, the relation kernel intelligent orchestration unit finally generates an optimized relation discovery execution plan. This plan is a structured instruction set that explicitly specifies the sequence of relation kernels to be executed, the input STDF nodes for each relation kernel, and the corresponding parameter configurations. This execution plan is passed to the execution plan generation unit for subsequent execution by the dynamic hypergraph construction and maintenance module 500.

[0102] The execution plan generation unit in the adaptive relation kernel orchestration engine module 400 is responsible for transforming the optimized relation kernel calculation process output by the intelligent relation kernel orchestration unit into a structured, directly executable relation discovery execution plan. This execution plan provides explicit instructions for the subsequent dynamic hypergraph construction and maintenance module 500 to actually call the relation kernel function and construct hyperedges.

[0103] The execution plan generation process includes the following steps: Implementation Plan Formal definition: Relationship discovery execution plan It is constructed as a directed acyclic graph, where each node represents a relational kernel computation task to be executed, and each edge represents a data dependency between tasks. This plan can be formally defined as: ( ); in: It is a set of relational kernel computation tasks.

[0104] It is a set of directed edges representing the dependencies between tasks. If there exists a path from... point to The edge represents Execution depends on The output.

[0105] Structure of the S243.2 relational kernel computation task: Each computational task in They all encapsulate all the information needed to execute a specific relation kernel instance, and can be represented as: ; in: It is the identifier of the specific relation kernel function to be called, which corresponds to a relation kernel in relation kernel function library module 300. It is the set of spatiotemporal data field nodes required for the execution of this task. These nodes are determined by filtering based on the spatiotemporal constraints of the query intent and data availability. This is the set of runtime parameters for the relation kernel instance. These parameters are dynamically adjusted and optimized by the relation kernel intelligent orchestration unit based on the domain semantic graph and query constraints. This describes the expected output type of the task, namely, a relation instance. The structure and semantic tags. It includes additional runtime constraints that may need to be followed when performing the task, such as the priority of computing resources and fault tolerance strategies.

[0106] Task dependency construction: For complex queries involving multiple relation core combinations, the execution plan generation unit will generate the execution plan according to the logical order determined by the relation core intelligent orchestration unit. Establish dependencies between tasks.

[0107] Execution plan output and readiness: The final generated execution plan It is a structured representation that can be directly parsed and executed by a distributed computing framework. It contains all the information needed to perform relation discovery tasks, eliminating the need for manual intervention. This plan is then passed to the dynamic hypergraph construction and maintenance module 500 to trigger the actual relation kernel computation and hyperedge generation process.

[0108] The dynamic hypergraph construction and maintenance module 500 is responsible for executing the execution plan generated by the adaptive relation kernel orchestration engine module 400. Furthermore, through computation-driven methods, discrete spatiotemporal data field nodes are connected to dynamically construct or expand the core data structure of the system, namely a dynamic hypergraph.

[0109] The construction process of a dynamic hypergraph specifically includes the following steps: Execution Plan Reception and Parsing: The Dynamic Hypergraph Construction and Maintenance Module 500 receives the Relationship Discovery Execution Plan output by the Execution Plan Generation Unit. This module parses the directed acyclic graph structure in the plan, identifies the sequence of relational kernel computation tasks to be executed, and their dependencies.

[0110] Execution of relational kernel computation tasks: The module executes according to the execution plan. It schedules computing resources to execute each relational kernel computation task. For each task, the system calls its specified relational kernel function. and the set of input spatiotemporal data field nodes defined in the task. and runtime parameter set As an actual parameter of a function.

[0111] Relation instance generation: When a relation kernel function Once the execution is complete and a satisfying association is successfully detected between the input STDF nodes, it generates a structured relation instance. This relation instance It includes information such as associated participating nodes, quantification attributes, confidence levels, and spatiotemporal validity.

[0112] Construction and expansion of dynamic hypergraphs: The system maintains a dynamic hypergraph. Its formal definition is: ( ); in: It is the collection of all spatiotemporal data field nodes that have been generated in the system.

[0113] It is the set of all generated relation instances in the system. Each generated non-empty relation instance... In this embodiment, it is regarded as a connection to its associated node set. The hyperedge. Dynamically added to the set of hyperedges of the hypergraph, i.e., the operation is performed: ; Through this operation, the originally discrete STDF nodes are connected by newly generated hyperedges, thus explicitly expressing the intrinsic relationships between data fields in the hypergraph's topology. This construction process is data-driven and dynamically emergent because the hypergraph's connection structure is a direct result of the relation kernel's computation on the original data, rather than based on pre-defined static links.

[0114] The Dynamic Hypergraph Construction and Maintenance Module 500 is not only responsible for the initial construction of the dynamic hypergraph, but also formulates and executes a series of maintenance strategies to ensure the hypergraph's real-time performance, accuracy, and consistency. As a dynamic knowledge carrier, the nodes and hyperedges of a hypergraph may change over time, with the influx of new data, or updates to domain knowledge. An effective maintenance mechanism is crucial for supporting continuous relationship discovery and high-precision knowledge retrieval.

[0115] The maintenance strategy for dynamic hypergraphs mainly includes the following steps: Spatiotemporal data field node update and lifecycle management: As new raw marine survey data is continuously acquired and processed, the system generates new STDF nodes and adds them to the vertex set of the hypergraph. In addition, existing STDF nodes may change due to data updates, corrections, or precision improvements. When an STDF node is updated, the system marks the node and records its update time. For STDF nodes that are no longer active, have exceeded their validity period, or have been replaced, maintenance policies can mark them as expired or move them to the historical archive area for management. To maintain its scale and keep it active.

[0116] Re-evaluation and refresh of relation instances: When re-evaluating and refreshing a relation instance When a related STDF node is updated, or when a relation kernel definition in the relation kernel function library is revised, the system triggers a re-evaluation of the relation instance. Specifically, the dynamic hypergraph maintenance module identifies all affected hyperedges. For these hyperedges, the system can: Re-execute the relation kernel: Call the original relation kernel that generated the hyperedge. Using the updated STDF nodes as input, new relation instances are recalculated and generated. If there are differences between the old and new relation instances, the hyperedges are updated. Attributes Confidence level or spatiotemporal validity .

[0117] Incremental update: For some relational kernels that support incremental calculation, the system can calculate only the part of the data affected by the change, so as to save resources.

[0118] Validation and Deletion: If the relation kernel cannot detect the association again after re-evaluation, the corresponding hyperedge will be removed from the kernel. Removed from the middle.

[0119] Relationship discovery for new nodes: When a new STDF node is added to the hypergraph Subsequently, the maintenance module will work in conjunction with the adaptive relation kernel orchestration engine module 400 to discover new relations that these new nodes may form. The system can proactively trigger the relation kernel orchestration process, executing relevant relation kernels for newly added STDF nodes and their neighboring spatiotemporal data fields, in order to generate new hyperedges and expand the network. This process can be periodic or event-driven.

[0120] Hypergraph Structure Optimization and Redundancy Elimination: As the hypergraph continues to grow, the maintenance module will periodically perform structural optimization. This includes optimizing structures for hypergraphs with extremely low confidence scores. For hyperedges that have not been queried or accessed for a long time, the system can mark them as low priority, or even delete or archive them according to preset policies, to reduce the complexity and storage burden of the hypergraph. In addition, the system will also perform data consistency checks to ensure that the STDF nodes associated with the hyperedges are in good working order. It still exists and is valid.

[0121] Through the aforementioned dynamic maintenance strategy, the ocean spatiotemporal data field supergraph can continuously reflect the latest ocean observation and analysis results, ensuring that it serves as a comprehensive, accurate, and real-time updated ocean knowledge graph, providing users with continuously valuable insights.

[0122] In the hypergraph patterned query and analysis module 600, the query pattern definition and expression unit is responsible for providing a structured way for users or upper-layer applications to define what they want to do in the dynamic hypergraph. Complex association patterns retrieved in the search.

[0123] The definition and expression of query patterns, and their specific implementation process, include the following steps: Formal definition of a query schema: A query schema It is formally defined as a query template hypergraph. This template hypergraph describes the target subgraph structure to be matched, and is defined as follows: ; in: It is a collection of query vertex variables. Each query vertex variable As a placeholder, it represents a space within a dynamic hypergraph. The matched STDF node.

[0124] It is a collection of query hyperedge variables. Each query hyperedge variable Connected One or more vertex variables in the array represent a relation instance (hyperedge) to be matched.

[0125] It is a set of variables applied to the query vertex. The set of constraints on.

[0126] It is a set of variables applied to the query hyperedge. The set of constraints on.

[0127] Vertex constraints Concretization: Vertex constraint set This is used to precisely specify the characteristics that the STDF nodes to be matched should possess. These constraints can include: Data type constraint: limits the data type of the STDF nodes matched by a certain query vertex variable.

[0128] Spatiotemporal range constraint: limits the spatiotemporal range of the matched STDF nodes. and It must intersect with or be contained within the specified spatiotemporal region.

[0129] Internal data attribute constraints: the original dataset encapsulated within the STDF node. Constraints are imposed on the statistical properties.

[0130] Hyperedge constraints Concretization: Hyperedge Constraint Set These constraints are used to precisely specify the characteristics that the relation instances to be matched should possess. These constraints can include: Relation type constraint: Restricts the relation instance matched by a certain query hyperedge variable to be generated by a relation kernel of a specific type.

[0131] Quantified attribute constraints: The set of quantified attributes for a relation instance. Constraints are imposed on the values ​​in the data.

[0132] Confidence constraint: Confidence score for relation instances Set a threshold.

[0133] Associative node number constraint: limits the number of vertices connected by the matched hyperedge.

[0134] The query pattern can be expressed as follows: the formalized query pattern described above. It needs to be expressed in a machine-readable format for subsequent query engine processing.

[0135] The Hypergraph Pattern Matching Algorithm Unit is the core computing engine of the Hypergraph Pattern Query and Analysis Module 600. It is responsible for receiving structured query patterns. and in dynamic hypergraph The algorithm performs an efficient subgraph matching algorithm to find all subgraph instances that satisfy the query pattern.

[0136] The specific implementation process of the hypergraph pattern matching algorithm includes the following steps: Candidate set selection and preprocessing: Before starting the formal graph matching process, the algorithm first utilizes the query pattern. Vertex constraints defined in and hyperedge constraints For large dynamic hypergraphs Pre-screening and pruning are carried out.

[0137] Backtracking-based subgraph isomorphic search: The algorithm employs a subgraph isomorphic search method suitable for hypergraphs, typically a backtracking algorithm with pruning optimization. The process begins with an empty matching map and progressively expands it until a complete match is found or the current path is deemed invalid. The algorithm selects an unmatched query vertex variable as the starting point and starts from its candidate set. Select one candidate STDF node for trial matching. If a query superedge Connected the query vertex set Then the hypergraph and hyperedge it matches It must be connected exactly. The set of matched STDF nodes.

[0138] Results collection and output: When the algorithm successfully generates a query pattern... When all vertex and hyperedge variables have found a valid match that satisfies all constraints, it means that in a dynamic hypergraph... A subgraph instance matching the query pattern was found.

[0139] The specific implementation process of this unit includes the following steps: Multi-view result presentation: For each subgraph instance that satisfies the query pattern returned by the query, the system provides multiple presentation methods to adapt to different analysis needs.

[0140] List View: Displays all matching subgraph instances in a structured list format. Each instance can be expanded to view summary information about the specific spatiotemporal data field nodes and relation instances that make up that instance, such as the spatiotemporal range and data type of the nodes, as well as the relation type and key quantization attributes of the hyperedges.

[0141] Topological view: This view graphically renders matching subgraph instances, where STDF nodes are vertices and relation instances are hyperedges connecting them. This view visually represents the topological structure of complex patterns.

[0142] Geospatial View: Overlays the spatiotemporal validity range of matched STDF nodes and relationship instances onto the Geographic Information System map. Users can observe the distribution, extent, and spatiotemporal evolution of these patterns within a geospatial context. These views are interconnected; a user's selection or highlighting action in one view will be synchronously reflected in other views.

[0143] Relationship tracing mechanism: To ensure the reliability of the analysis results, the system provides a complete tracing path for each relationship instance. Because each relationship instance... Its source information is recorded during generation, enabling a traceability mechanism. When a user selects a specific relationship instance in the results presentation interface, the system can immediately access and display the traceability information contained in its metadata, mainly including: relation kernel type : Clearly indicate which relation kernel function was used to calculate and generate the relation.

[0144] Input data nodes List all STDF nodes that participated in this calculation and formed this association.

[0145] Quantification attributes This displays all the quantitative indicators calculated by the relation kernel, such as the maximum gradient value and correlation coefficient, which are the direct numerical basis for determining the validity of the relation.

[0146] Relation kernel parameters: Records the specific parameters used when executing the relation kernel. This allows the calculation process to be accurately reproduced.

[0147] Interactive source tracing and data drill-down: Based on the provided source tracing information, users can perform interactive drill-down analysis. Users can not only see how the correlation was generated, but also further examine the original data that produced the correlation.

[0148] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A knowledge graph-based system for the association analysis and retrieval of historical marine survey data, characterized in that, include: The data processing and spatiotemporal data field construction module is configured to process and encapsulate raw marine survey data into spatiotemporal data field nodes carrying the original high-dimensional dataset. The domain semantic graph management module is configured to store and manage a domain semantic graph that defines the relationships between marine domain concepts, relation types, methods, and data types. The relation kernel function library module is configured to provide a set of relation kernels that encapsulate domain-specific algorithms. The adaptive relation kernel orchestration engine module is configured to parse user queries into structured query intents, and based on the structured query intents and the domain knowledge in the domain semantic graph management module, select and orchestrate relation kernels from the relation kernel function library module to generate a relation discovery execution plan. The dynamic hypergraph construction and maintenance module is configured to execute the relation discovery execution plan. It generates relation instances as hyperedges by calling the relation kernel to dynamically construct and maintain a dynamic hypergraph with spatiotemporal data field nodes as vertices and relation instances as hyperedges. The patterned query and analysis module is configured to perform patterned queries on the dynamic hypergraph to retrieve subgraph instances that satisfy a user-defined query pattern.

2. The system according to claim 1, characterized in that, The relation kernel is a computational function configured to directly act on the original high-dimensional dataset within one or more spatiotemporal data field nodes, and to discover and quantify the correlation between the spatiotemporal data field nodes through a preset algorithm.

3. The system according to claim 1, characterized in that, The relation instance is a structured data object, including: The associated node set is used to identify all spatiotemporal data field nodes involved in the association; The relation kernel type identifier is used to specify the relation kernel that generated this relation instance; The quantized attribute set is used to store the numerical results obtained from relation kernel calculations.

4. The system according to claim 1, characterized in that, The adaptive relation kernel orchestration engine module includes: A query intent parsing unit is configured to extract semantic keywords from user queries and align them to concept vertices and relation type vertices in the domain semantic graph to construct the structured query intent; and The relation kernel intelligent orchestration unit is configured to select, combine, and parameterize relation kernels based on the query intent and the association between method vertices and data type vertices in the domain semantic graph.

5. The system according to claim 4, characterized in that, The adaptive relation kernel orchestration engine module further includes an execution plan generation unit, configured to generate a directed acyclic graph consisting of relation kernel computation tasks as the relation discovery execution plan by combining and parameterizing the relation kernels.

6. The system according to claim 1, characterized in that, The dynamic hypergraph construction and maintenance module is also configured to: trigger a re-evaluation of the relation instance when the spatiotemporal data field node associated with a relation instance is updated, so as to update or remove the hyperedge corresponding to the relation instance.

7. The system according to claim 1, characterized in that, The patterned query and analysis module includes a query pattern definition and expression unit, configured to formalize the user-defined query pattern into a query template hypergraph consisting of query vertex variables, query hyperedge variables, vertex constraints, and hyperedge constraints.

8. The system according to claim 7, characterized in that, The patterned query and analysis module further includes a hypergraph pattern matching algorithm unit, configured to: filter candidate sets of vertices and hyperedges in the dynamic hypergraph, and perform a backtracking-based subgraph isomorphic search on the filtered candidate set to find all subgraph instances that match the query template hypergraph.

9. The system according to claim 1, characterized in that, The patterned query and analysis module further includes a result presentation and association tracing unit, configured to provide association tracing for each relation instance when presenting query results. The association tracing displays the relation kernel type, input data nodes, and quantified attributes that generated the relation instance.

10. A method for retrieving and analyzing historical marine survey data based on knowledge graphs, and a system for retrieving and analyzing historical marine survey data based on knowledge graphs according to any one of claims 1-9, characterized in that, Includes the following steps: S1: Receive user queries and parse the user queries into structured query intents; S2: Based on the structured query intent and the preset domain semantic graph, select and arrange relation kernels from the relation kernel function library to generate a relation discovery execution plan; S3: Execute the relationship discovery execution plan, and dynamically construct a dynamic hypergraph with spatiotemporal data field nodes as vertices and relationship instances as hyperedges by calling the relationship kernel to calculate and generate relationship instances. S4: Receive the user-defined query pattern and perform pattern matching on the dynamic hypergraph to find subgraph instances that satisfy the query pattern as query results; S5: Present the query results and provide association tracing information for the relational instances in the results.

Images