A method and system for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks, a terminal and a storage medium

Abstract

This invention discloses a method, system, terminal, and storage medium for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks. The method includes: acquiring multi-source sensing data and preprocessing the multi-source sensing data to obtain preprocessed multi-source sensing data; performing entity modeling processing on the preprocessed multi-source sensing data to obtain carbon source and sink entities, and performing feature modeling and relation modeling processing on the carbon source and sink entities to obtain carbon source and sink entity features and carbon source and sink entity relations; performing graph mapping processing on the carbon source and sink entity features and carbon source and sink entity relations to obtain graph nodes and graph edges, and constructing a graph substructure based on the carbon source and sink entities, graph nodes, and graph edges; and performing storage update processing on the graph substructure to obtain a spatiotemporal dynamic knowledge graph of carbon sources and sinks. This invention, by acquiring the entity features and entity relations of carbon sources and sinks and performing dynamic storage updates, constructs a spatiotemporal dynamic knowledge graph of carbon sources and sinks, which can effectively improve the estimation accuracy of carbon sources and sinks.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, system, terminal, and computer-readable storage medium for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks. Background Technology

[0002] With the increasing severity of global climate change, the dynamic changes of carbon sources and sinks in terrestrial ecosystems (carbon sources and sinks include carbon sources and carbon sinks; carbon sources refer to the processes, activities, or mechanisms by which carbon is released into the atmosphere from carbon reservoirs, such as deforestation and coal-fired power generation; carbon sinks, on the contrary, refer to the processes, activities, or mechanisms by which carbon dioxide is absorbed from the atmosphere through various measures, thereby reducing the concentration of greenhouse gases in the atmosphere) have become a core issue in environmental monitoring, carbon emission control, and ecological policy formulation.

[0003] Current technologies primarily rely on single-type data sources such as remote sensing imagery or ground monitoring data, combined with static modeling methods, to estimate carbon fluxes. However, when faced with multi-source heterogeneous data, multi-scale spatial structures, and complex ecosystems subject to frequent human disturbance, it is difficult to accurately characterize the spatiotemporal dynamic evolution of carbon sources and sinks, resulting in poor accuracy in estimating the dynamic changes of carbon sources and sinks.

[0004] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0005] The main objective of this invention is to provide a method, system, terminal, and computer-readable storage medium for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks. This invention aims to solve the problem in the prior art that it is difficult to accurately characterize the spatiotemporal dynamic evolution of carbon sources and sinks when facing complex ecosystems with multi-source heterogeneous data, multi-scale spatial structures, and frequent human interference, resulting in poor accuracy in estimating the dynamic changes of carbon sources and sinks.

[0006] To achieve the above objectives, the present invention provides a method for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks, the method comprising the following steps:

[0007] Acquire multi-source sensing data and preprocess the multi-source sensing data to obtain preprocessed multi-source sensing data;

[0008] The preprocessed multi-source sensing data is subjected to entity modeling to obtain carbon source sink entities. The carbon source sink entities are then subjected to feature modeling and relation modeling to obtain carbon source sink entity features and carbon source sink entity relations.

[0009] The carbon source and sink entity features and carbon source and sink entity relationships are processed by graph mapping to obtain graph nodes and graph edges, and a graph substructure is constructed based on the carbon source and sink entities, the graph nodes and graph edges;

[0010] The storage and update process of the graph substructure is performed to obtain a spatiotemporal dynamic knowledge graph of carbon sources and sinks.

[0011] Optionally, in the method for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks, the multi-source sensing data includes structured data and unstructured text corpus data; the preprocessed multi-source sensing data includes target structured data and target text corpus data.

[0012] The acquisition of multi-source sensing data and the preprocessing of the multi-source sensing data to obtain preprocessed multi-source sensing data specifically include:

[0013] The structured data is acquired from the multi-source sensing data, wherein the structured data includes remote sensing image data, ground sensing monitoring data, meteorological and carbon flux statistics data, and socio-economic activity data;

[0014] The structured data is processed by data completion, standardization, outlier handling, and deduplication to obtain the target structured data.

[0015] The unstructured text corpus data in the multi-source perception data is obtained, and entity information extraction processing is performed on the unstructured text corpus data to obtain the target text corpus data.

[0016] Optionally, the method for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks, wherein the step of performing entity modeling processing on the preprocessed multi-source sensing data to obtain carbon source and sink entities, and performing feature modeling and relation modeling processing on the carbon source and sink entities to obtain carbon source and sink entity features and carbon source and sink entity relations, specifically includes:

[0017] The preprocessed multi-source sensing data is judged according to entity function to obtain carbon source sink entities, wherein the carbon source sink entities include carbon source entities and carbon sink entities;

[0018] The carbon source sink entity is subjected to feature modeling processing to obtain carbon source sink entity features, wherein the carbon source sink entity features include semantic features, spatial features, temporal features and attribute features;

[0019] Relationship modeling is performed on the carbon source and sink entities to obtain carbon source and sink entity relationships, which include conceptual relationships, spatial relationships, temporal relationships, and attribute relationships.

[0020] Optionally, in the method for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks, the step of performing feature modeling processing on the carbon source and sink entities to obtain carbon source and sink entity features specifically includes:

[0021] The semantic features are obtained by performing entity name normalization, entity type code encoding, and entity identifier code construction on the carbon source sink entity.

[0022] The spatial features are obtained by extracting latitude and longitude coordinates and determining spatial representation methods for the carbon source and sink entities.

[0023] The carbon source and sink entities are subjected to time series table construction and time topology construction to obtain the time features;

[0024] The carbon source and sink entities are processed by assigning entity attributes to obtain the attribute features.

[0025] Optionally, the method for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks, wherein the step of performing relational modeling on the carbon source and sink entities to obtain the carbon source and sink entity relationships specifically includes:

[0026] The concept relationship is obtained by performing hierarchical structure identification and semantic role parsing on the carbon source and sink entities.

[0027] The spatial relationships of the carbon source and sink entities are obtained by using GIS topology analysis methods to identify geospatial relationships and determine spatial conditions.

[0028] The time relationship is obtained by performing time-series evolution path extraction on the carbon source and sink entities.

[0029] The attribute relationships are obtained by performing attribute feature extraction and attribute similarity calculation on the carbon source and sink entities.

[0030] Optionally, the spatiotemporal dynamic knowledge graph construction method for carbon source sinks, wherein the step of performing graph mapping processing on the carbon source sink entity features and carbon source sink entity relationships to obtain graph nodes and graph edges, and constructing a graph substructure based on the carbon source sink entities, the graph nodes, and the graph edges, specifically includes:

[0031] The semantic features, spatial features, temporal features, and attribute features are subjected to graph mapping processing to obtain multiple graph nodes;

[0032] The conceptual relationships, spatial relationships, temporal relationships, and attribute relationships are subjected to graph mapping processing to obtain multiple graph edges;

[0033] The graph is constructed based on the carbon source and sink entities, multiple graph nodes, and multiple graph edges to obtain a graph substructure.

[0034] Optionally, the method for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks, wherein the step of storing and updating the graph substructure to obtain the spatiotemporal dynamic knowledge graph of carbon sources and sinks specifically includes:

[0035] When new carbon source and sink data is received, a state node chain is constructed based on the new carbon source and sink data;

[0036] The spatiotemporal structure and attribute structure in the graph substructure are determined, and the graph substructure is three-dimensionally linked with the state node chain according to the spatiotemporal structure and the attribute structure to obtain the initial carbon source sink knowledge graph;

[0037] The initial carbon source and sink knowledge graph is subjected to dynamic incremental writing and dynamic tracking processing to obtain a spatiotemporal dynamic knowledge graph of carbon sources and sinks.

[0038] Furthermore, to achieve the above objectives, the present invention also provides a spatiotemporal dynamic knowledge graph construction system for carbon sources and sinks, wherein the spatiotemporal dynamic knowledge graph construction system for carbon sources and sinks includes:

[0039] The data preprocessing module is used to acquire multi-source sensing data and preprocess the multi-source sensing data to obtain preprocessed multi-source sensing data.

[0040] The entity feature and relationship extraction module is used to perform entity modeling processing on the preprocessed multi-source sensing data to obtain carbon source sink entities, and to perform feature modeling processing and relationship modeling processing on the carbon source sink entities to obtain carbon source sink entity features and carbon source sink entity relationships.

[0041] The initial graph construction module is used to perform graph mapping processing on the carbon source and sink entity features and carbon source and sink entity relationships to obtain graph nodes and graph edges, and to construct graph substructures based on the carbon source and sink entities, the graph nodes and graph edges;

[0042] The graph dynamic update module is used to store and update the graph substructure to obtain a carbon source-sink spatiotemporal dynamic knowledge graph.

[0043] In this invention, multi-source sensing data is acquired and preprocessed to obtain preprocessed multi-source sensing data. Entity modeling is then performed on the preprocessed multi-source sensing data to obtain carbon source sink entities. Feature modeling and relation modeling are then performed on these carbon source sink entities to obtain carbon source sink entity features and carbon source sink entity relationships. Graph mapping is then performed on the carbon source sink entity features and carbon source sink entity relationships to obtain graph nodes and graph edges. A graph substructure is constructed based on the carbon source sink entities, graph nodes, and graph edges. Finally, the graph substructure is stored and updated to obtain a spatiotemporal dynamic knowledge graph of carbon source sinks. This invention, by acquiring multi-source sensing data, constructing carbon source sink entity features and carbon source sink entity relationships based on the multi-source sensing data, and then constructing a graph substructure based on these features and relationships, and introducing a time evolution mechanism to form a spatiotemporal dynamic knowledge graph of carbon source sinks, can effectively improve the estimation accuracy of carbon source sinks. Attached Figure Description

[0044] Figure 1 This is a flowchart of a preferred embodiment of the spatiotemporal dynamic knowledge graph construction method for carbon source sinks of the present invention;

[0045] Figure 2 This is a schematic diagram of the structured data processing flow of a preferred embodiment of the spatiotemporal dynamic knowledge graph construction method for carbon source sinks of the present invention.

[0046] Figure 3 This is a schematic diagram of the process for extracting carbon source sink entity information, which is a preferred embodiment of the spatiotemporal dynamic knowledge graph construction method for carbon source sinks of the present invention.

[0047] Figure 4 This is a schematic diagram of the carbon source sink entity modeling, which is a preferred embodiment of the spatiotemporal dynamic knowledge graph construction method of the present invention.

[0048] Figure 5 This is a schematic diagram of the carbon source sink entity relationship extraction process, which is a preferred embodiment of the spatiotemporal dynamic knowledge graph construction method of the present invention.

[0049] Figure 6 This is a schematic diagram of the carbon source sink entity storage and update structure, which is a preferred embodiment of the spatiotemporal dynamic knowledge graph construction method of the present invention.

[0050] Figure 7 This is a structural diagram of a preferred embodiment of the spatiotemporal dynamic knowledge graph construction system for carbon source sinks of the present invention;

[0051] Figure 8 This is a structural diagram of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0053] With the intensification of global climate change, the dynamic changes of carbon sources and sinks in terrestrial ecosystems (carbon sources refer to the processes, activities, or mechanisms by which carbon is released into the atmosphere from carbon reservoirs, such as deforestation and coal-fired power generation; carbon sinks, on the other hand, refer to the processes, activities, or mechanisms by which carbon dioxide is absorbed from the atmosphere through various measures, thereby reducing the concentration of greenhouse gases in the atmosphere. Major carbon sinks worldwide include forest carbon sinks, grassland carbon sinks, arable land carbon sinks, soil carbon sinks, and marine carbon sinks) have become a core issue in environmental monitoring, carbon emission control, and ecological policy formulation. As a key link in the carbon cycle of ecosystems, the heterogeneity of carbon sources and sinks in time and space directly affects the accuracy of regional carbon budget assessments and the scientific nature of regulation. Therefore, how to achieve dynamic, high-precision, and intelligent monitoring of carbon sources and sinks has become a core technical challenge that urgently needs to be solved in current carbon cycle research and management practices.

[0054] Current technologies primarily rely on single-source data sources such as remote sensing imagery or ground-based monitoring data, combined with static modeling methods, for carbon flux estimation. While these techniques have some applicability, they exhibit significant limitations when faced with multi-source heterogeneous data, multi-scale spatial structures, and complex ecosystems frequently disturbed by human activity. For example, against the backdrop of urban expansion and land-use change, the processes of carbon source and sink changes are complex and dynamic, making it difficult for traditional methods to accurately characterize their spatiotemporal dynamic evolution. Furthermore, although some studies have attempted to improve estimation accuracy by introducing data assimilation and multi-source fusion models, significant deficiencies remain in structural modeling capabilities, making it difficult to express the complex nonlinear and multi-level interactions between carbon flux and the environment and human activities.

[0055] In recent years, knowledge graphs, as an intelligent technology capable of depicting complex semantic relationships between entities and supporting logical reasoning, have been increasingly applied in fields such as carbon market regulation and carbon emission management, demonstrating promising development prospects. In related research, knowledge graphs are primarily used to integrate structured carbon emission data and corporate carbon activity records, achieving preliminary information association and visualization. However, these applications are mostly focused on the expression and summarization of static information, and a unified methodological system oriented towards natural ecological processes and possessing spatiotemporal dynamic modeling capabilities has not yet been formed, making it difficult to meet the practical needs of modeling and large-scale intelligent cognition in the evolution of carbon sources and sinks.

[0056] Existing knowledge graph construction methods in the carbon cycle field generally rely on structured statistical information, resulting in static graph structures and simple node semantics. This makes them ill-suited for modeling the core characteristics of carbon flux, such as its spatiotemporal evolution and complex interactions. Therefore, existing technologies still face the following key challenges:

[0057] 1. Insufficient ability to fuse multi-source data: Most current studies are based on single-type data and lack a systematic integration and spatiotemporal alignment mechanism for multi-source heterogeneous information such as remote sensing images, ground flux observations, meteorological variables and socioeconomic data. This results in limited dimensions of the constructed map information and insufficient accuracy and completeness of carbon source and sink modeling results.

[0058] 2. Weak static and dynamic modeling capabilities of graph structures: Most traditional graphs do not incorporate time evolution mechanisms, making it difficult to capture the evolution of relationships between various elements in the carbon cycle system over time; when dealing with the coupling modeling of natural processes and human activities, they also lack the ability to express nonlinear and multi-scale interactions, which restricts their further promotion in complex application scenarios such as carbon source and sink monitoring, prediction and reasoning.

[0059] Therefore, to address the aforementioned issues, this invention proposes a multi-source sensing collaborative-driven method for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks. This method standardizes remote sensing imagery, carbon flux observations, environmental factors, and human activity data using a unified spatial basis, constructs multi-type entity nodes and semantic relationships, and introduces a temporal evolution mechanism to form a dynamic graph structure. Compared with existing technologies, this invention possesses advantages such as strong dynamic modeling capabilities and high multi-source data fusion, making it suitable for various application scenarios such as carbon flux monitoring, ecological compensation assessment, and regional carbon governance, and has broad practical application value.

[0060] The method for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks according to a preferred embodiment of the present invention, such as... Figure 1 As shown, the method for constructing the spatiotemporal dynamic knowledge graph of carbon sources and sinks includes the following steps:

[0061] Step S10: Acquire multi-source sensing data and preprocess the multi-source sensing data to obtain preprocessed multi-source sensing data. The multi-source sensing data includes structured data and unstructured text corpus data; the preprocessed multi-source sensing data includes target structured data and target text corpus data.

[0062] The data used in this invention includes five categories: remote sensing image data, ground sensor monitoring data, meteorological and carbon flux statistics data, socio-economic activity data, and unstructured text corpus data, which constitute the multi-source sensing input for the spatiotemporal dynamic modeling of carbon sources and sinks (i.e., the multi-source sensing data in this invention).

[0063] Specifically, the structured data in the multi-source sensing data is acquired, wherein the structured data includes remote sensing image data, ground sensing monitoring data, meteorological and carbon flux statistics data, and socio-economic activity data.

[0064] The main sources and core fields of various data types are shown in Table 1 below:

[0065] Table 1: Multi-source sensing data table

[0066]

[0067]

[0068] The remote sensing image data sources include, but are not limited to, remote sensing platforms such as MODIS, Sentinel, and Landsat, with resolutions ranging from 30m to 500m and continuous observation capabilities at regional scales. Ground monitoring data comes from the National Ecological and Environmental Observation Network and regional flux tower systems, providing high-frequency carbon exchange information at the minute to hourly level. Meteorological and carbon flux statistics come from meteorological bureaus, ecological and environmental bureaus, and statistical yearbooks. Socioeconomic activity data integrates publicly available government statistics, urban big data platforms, and third-party industry databases. Unstructured text corpus data mainly includes carbon monitoring reports, ecological survey archives, carbon market documents, scientific research papers, and industry development books, containing a large amount of tacit knowledge and expert experience information, which is an important foundation for semantic enhancement modeling.

[0069] The structured data is processed by data completion, standardization, outlier handling, and deduplication to obtain the target structured data.

[0070] The data extraction in this invention comprises two parts: entity extraction and relation extraction of structured and unstructured text data. For the extraction of carbon source and sink entity information, a dual-channel parallel processing strategy of structured and unstructured data is adopted to achieve high-quality extraction of entity nodes and unified semantic representation.

[0071] It should be noted that the aforementioned multi-source data may exhibit inconsistencies in format and fields due to differences in data provision platforms or acquisition times. However, the core information fields required by this invention are universally applicable across similar data types. Because multi-source sensing data exhibits certain instability during acquisition, processing, and transmission, particularly remote sensing and sensor monitoring data which often contain missing data, noise, and outliers, rigorous data preprocessing is necessary before formal data modeling. This preprocessing mainly includes, but is not limited to, the following aspects:

[0072] 1. Data Missing: If a record has missing key fields (such as timestamp or spatial coordinates), this invention will first try to fill them in by interpolation or logical inference; if it cannot be filled in, the record will be removed.

[0073] 2. Inconsistent data formats: such as non-standard time formats and inconsistent spatial coordinate systems, need to be uniformly converted to standard formats (this invention uses ISO timestamps and WGS-84 coordinate system) for subsequent fusion modeling.

[0074] 3. Data anomalies: If the data shows carbon flux values ​​exceeding the theoretical threshold, mismatch between monitoring station locations and images, or sudden changes in the thermal intensity of social activities, judgments and corrections will be made based on historical distribution characteristics, boundary rules, and domain knowledge.

[0075] 4. Data redundancy and duplication: During the data source integration process, there may be duplicate records. It is necessary to establish a unique identifier through a time-space joint primary key to remove duplicates and ensure the integrity and uniqueness of the dataset.

[0076] like Figure 2 As shown, for the processing of structured data, firstly, multi-source structured data from remote sensing observations, environmental monitoring, and ground sensor data are integrated to unify the spatial and temporal reference systems. Missing values, duplicate records, and outliers are then cleaned to ensure data consistency across spatial and temporal dimensions. Subsequently, based on the spatial coverage of the data, the cleaned observation data is allocated to corresponding semantic entities according to area proportions, and the attribute fields of these entities are constructed. This step includes:

[0077] 1. Project all spatial coordinate fields in structured data to the WGS-84 geographic coordinate system. Furthermore, standardize all time fields using the "year-month-day-hour" format, supporting time index alignment between day and hour scales.

[0078] 2. Missing fields are filled using the K-nearest neighbor imputation method. For any missing sample x... q Let its neighboring samples be x1, x2, ..., xn. K Then its interpolation value Represented as:

[0079]

[0080] Among them, y i Let K be the observation value of the i-th similar sample, and K be the number of neighbors.

[0081] 3. Calculate outlier boundaries using the mean μ and standard deviation σ, and remove all data records that meet the following conditions:

[0082] |x-μ|>3σ;

[0083] Where x is the original observation value.

[0084] 4. Construct a composite primary key based on spatial coordinates and timestamps, identify and delete duplicate records with the same primary key, and retain only the record with complete information.

[0085] 5. Based on the spatial overlap area between each observation data point and the target entity, the observation value is assigned to the entity using an area-weighted allocation method. Let A be the overlap area between the observation data unit and the i-th entity. i The total overlapping area is ∑ j A j Then assign a weight w i for:

[0086]

[0087] Ultimately, the data value v is assigned to the i-th entity as v. i =w i ·v.

[0088] 6. Map the cleaned and assigned index values ​​(such as carbon flux, temperature, pollutant concentration, etc.) to the structured attribute fields of the corresponding entities and store them in key-value format to form a complete set of entity attributes.

[0089] The unstructured text corpus data in the multi-source perception data is obtained, and entity information extraction processing is performed on the unstructured text corpus data to obtain the target text corpus data.

[0090] like Figure 3 As shown, for unstructured text data, the system first accesses unstructured corpus resources, including online texts and research reports in the carbon source sink field. Semantic paragraphs are extracted using a text segmentation module, and a CARE prompt word template based on a large language model is constructed to complete entity information extraction. This step includes:

[0091] 1. Construct a prompt input template, and fill in the text segment (Context), extraction action (Action), output structure (Result), and few-sample prompt (Example) in sequence.

[0092] 2. Call the large language model to perform the extraction task and identify information such as the name, classification, spatial location, and functional attributes of carbon source and sink entities.

[0093] 3. Convert the extracted results into a structured format and perform preliminary validation using regular expressions and template rules to determine whether they conform to the established data structure.

[0094] 4. Data that does not meet the requirements should be re-extracted or manually labeled and confirmed to ensure the accuracy and parsability of semantic entity information.

[0095] Step S20: Perform entity modeling processing on the preprocessed multi-source sensing data to obtain carbon source sink entities, and perform feature modeling and relation modeling processing on the carbon source sink entities to obtain carbon source sink entity features and carbon source sink entity relations.

[0096] This invention proposes a carbon source-sink knowledge graph representation model that integrates spatiotemporal semantics, and constructs the carbon source-sink graph ontology structure using the entity-attribute-relation (EAR) triple paradigm. The model construction includes the following steps:

[0097] First, by identifying the functions in the carbon cycle process, entities (here, "entity" refers to objective object units extracted from multi-source data that have independent semantic identifiers, spatial locations, and functional attributes) are determined to be either carbon source entities (with CO2 emission characteristics) or carbon sink entities (with CO2 absorption or fixation characteristics). Furthermore, carbon source entities are classified according to emission type into industrial emission facilities, transportation systems, and agricultural activity units, while carbon sink entities are classified according to ecological function and spatial affiliation into forest ecosystems, wetland systems, and marine carbon sink areas.

[0098] like Figure 4 As shown, a four-dimensional feature representation system is used to model carbon source and sink entities, including semantic features, spatial features, temporal features, and attribute features. Then, based on the functional classification of entities, four basic semantic relationships—conceptual relationships, spatial relationships, temporal relationships, and attribute relationships—are extracted respectively.

[0099] Specifically, the preprocessed multi-source sensing data is discriminated according to entity function to obtain carbon source and sink entities, wherein the carbon source and sink entities include carbon source entities and carbon sink entities.

[0100] For the entity modeling of carbon source sinks, this invention is based on internationally accepted functional classification standards, combined with the carbon emission and carbon sequestration mechanisms in the carbon cycle process, to systematically define and classify carbon source sink entities, specifically including the following steps:

[0101] 1. Carbon Source / Sink Function Identification: Based on the functional performance of an entity in the carbon cycle, it is first determined whether it is a carbon source entity (i.e., possessing CO2 emission characteristics) or a carbon sink entity (i.e., possessing CO2 absorption or fixation characteristics). This classification is based on the entity's activity characteristics and industry attributes (indirectly inferring whether an entity's behavior in the carbon cycle is "emitting CO2" or "absorbing or fixing CO2" through its activity characteristics and industry attributes, rather than simply determining whether it is emitting or absorbing through direct observation).

[0102] 2. Classification of Carbon Source and Carbon Sink Entities: If an entity is identified as a carbon source, it is further classified into three subcategories based on its dominant emission type: industrial emission facilities, transportation systems, and agricultural activity units. If an entity is identified as a carbon sink, it is classified into three subcategories based on its ecological function and spatial ecosystem affiliation: forest ecosystems, wetland systems, and marine carbon sink areas, as shown in Table 2 below.

[0103] Table 2: Classification of Carbon Sources and Sinks

[0104]

[0105] The carbon source sink entity is subjected to entity name normalization, entity type code encoding, and entity identifier code construction to obtain the semantic features; the carbon source sink entity is subjected to latitude and longitude coordinate extraction and spatial representation method discrimination to obtain the spatial features; the carbon source sink entity is subjected to time series table construction and time topology construction to obtain the time features; and the carbon source sink entity is subjected to entity attribute assignment to obtain the attribute features.

[0106] The feature modeling process for carbon source and sink entities (i.e., the set of partitioned carbon source and sink entities) is as follows:

[0107] This invention proposes a four-dimensional feature representation system, which models features from four dimensions: semantic features, spatial features, temporal features, and attribute features.

[0108] Regarding semantic features: In terms of semantic dimension, a conceptual system for carbon source and sink entities is constructed, and a three-level coding structure is used to achieve standardized expression and accurate mapping of entity semantic attributes. This process includes the following three components:

[0109] 1. Entity Name Normalization Process: Normalize the original entity name (the original name of the carbon source / sink entity) N. raw First, the Word2Vec word vector model is used to map it into a semantic vector representation. Let the entity name consist of a set of terms N. raw ={w1,w2,…,w n}, each term w i (for N) raw One of them obtains its corresponding word vector representation through a pre-trained Word2Vec model.

[0110]

[0111] Where d is the dimension of the embedding vector. Represents the set of real numbers. Let represent a d-dimensional real vector space.

[0112] The overall word vector representation of the entity name is obtained by averaging the word vectors of all terms.

[0113]

[0114] Where n is the number of terms.

[0115] Similarly, for the international standard terminology database Each candidate matching standard term in Similarly, its corresponding semantic vector is constructed based on Word2Vec.

[0116]

[0117] Where m is the number of candidate matching criteria terms, w' k for The kth candidate matching standard term in the.

[0118] Then, based on the word vectors of the original entity names and standard terms obtained from the calculation, the cosine similarity sim between them is calculated. j :

[0119]

[0120] Finally, the standard term with the highest semantic similarity is selected from the candidate matches as the normalization result N. norm :

[0121]

[0122] This step involves selecting the standard terms with the highest semantic similarity as the normalization result. This result serves as the basis for standard naming of entity nodes, relationship construction, entity alignment, and ontology mapping in subsequent graph construction, ensuring the consistency of semantic expression and the stability of the graph structure during the fusion of cross-source heterogeneous data.

[0123] 2. Entity Type Code Encoding Structure: This invention constructs a standard classification tree structure for carbon source and sink entities based on the IPCC National Greenhouse Gas Inventory Guidelines, employing a three-level system of "domain – category – subcategory" to hierarchically represent functional categories. The functional category C of the entity is... f The mapping is a three-level encoding structure, namely the Entity Type Code (ETC), defined as follows: C f The input variables are used to obtain ETC through IPCC semantic mapping. This step is to assign a functional category C to each entity. f Provides standardized mapping encoding for ETC, the expression is:

[0124] ETC=S i -C j -SC k ;

[0125] Wherein: S i ∈S represents the i-th carbon source or carbon sink sector, such as energy, industrial processes, forestry, etc. For the classification within the j-th domain; This is a subclass under the k-th category. S represents the set of primary functional areas for carbon sources or sinks, such as the main areas defined in the IPCC National Greenhouse Gas Inventory Guide, including energy, industrial processes and product use, agriculture, forestry and land use, and waste. This represents the set of secondary categories corresponding to the primary domain Si, used to further subdivide the types of emission or absorption activities within that functional domain. For example, in the "Energy" domain, secondary categories might include energy industry, manufacturing and construction, transportation, and others.

[0126] 3. Entity Identifier Code Construction: Extract the geographical coordinates of the entity. The spatial information, along with the timestamp T in ISO8601 format, is encoded using the Geohash geogrid coding function to obtain the spatial location code G, expressed as:

[0127] G = GridEncode(lat,lon);

[0128] Finally, the entity type code ETC, spatial code G, and timestamp T are concatenated to generate an entity identifier code with spatiotemporal uniqueness (the entity identifier code is used to uniquely locate a spatiotemporal entity instance, which is different from the semantic features that characterize the entity category and attributes; the two play different roles in constructing the carbon source sink knowledge graph).

[0129] For spatial features: Based on spatial dimensions, extract the latitude and longitude coordinates of the entity and determine its spatial representation method according to the entity type. If it is a point-source entity (such as a thermal power plant or a farm), it is directly represented by coordinate points; if it is a surface-type entity (such as a forest or wetland), its extent is represented by a vector surface or a raster layer.

[0130] For time features: Regarding the time dimension, firstly, obtain the time-series monitoring dataset D related to carbon source and sink entities. time The data sources include remote sensing imagery, ground sensor sampling, and continuous monitoring from fixed stations. Assuming this dataset contains high-frequency observation points ranging from minutes to hours, it can be represented as a time series D. time ={(t k ,v k)|k=1,2,…,N}(k represents the index of the k-th observation point in the time series, i.e., the k-th record of data sampling; N represents the total number of observation points in the time series dataset, i.e., there are N records in total, where t k For timestamps, v k This represents the observed value at the corresponding time point (such as carbon flux, status indicator, etc.).

[0131] To achieve unified modeling, the original data (D) time The original time-series monitoring dataset (which is the carbon source and sink entity data) is sliced ​​according to a fixed time interval Δt to construct a structured time-series table T. (Δt) Each row represents an observation record of a certain entity within a certain time segment. For example, for entity e... i Its observation data at different time periods Represented as:

[0132]

[0133] in, For entity e i The observation at the nth time point.

[0134] Furthermore, to express the dynamic evolution characteristics of entities in the time dimension, a temporal topology G is constructed with temporal nodes as vertices. time (Constructed based on observational data). In the temporal topology G time In the diagram, three typical types of time edges are defined (a time edge connects two temporal nodes in the graph structure, representing a certain evolutionary relationship): the first is a sequential edge, which represents a linear evolutionary relationship over time. Let's assume that for entity e... i There are two adjacent observation records and If t is satisfied k+1 -t k =Δt, then at node and Add sequential edges e between seq , denoted as: Second, periodic edges: used to express periodic patterns such as seasonal variations, diurnal variations, and seasonal fluctuations. If two observation records exist... Satisfy |t k+p -t k |≈n×T cycle , among which, T cycle Given a known period length (e.g., 24 hours or 12 months), where n is a positive integer, a periodic edge is formed by connecting two nodes, denoted as: Thirdly, trend edges: used to indicate the continuous upward or downward trend of carbon flux, assuming that for observed data... If its corresponding observed value satisfies (i.e., an upward trend), or (i.e., a downward trend), then add trend edges e sequentially between each node. trend , denoted as: e trend =(v k →v k+1 ),

[0135] Attribute Features: Entity attributes (including static and dynamic attributes) are assigned values ​​according to a "static-dynamic" dual-modal feature system. Static attributes include information that does not change over time, such as entity area, land use type, and functional type, and are directly extracted from remote sensing interpretation results. Dynamic attributes are dynamically updated based on remote sensing inversion parameters (such as NDVI, land surface temperature, etc.) and sensor observation data (such as CO2 concentration, soil moisture, etc.). All attributes are stored uniformly using a field encoding method, and a relationship is established between them and entity nodes.

[0136] The carbon source and sink entities are subjected to hierarchical structure identification and semantic role parsing to obtain the conceptual relationship; GIS topology analysis methods are used to perform geospatial relationship identification and spatial condition judgment on the carbon source and sink entities to obtain the spatial relationship; temporal evolution path extraction is performed on the carbon source and sink entities to obtain the temporal relationship; attribute feature extraction and attribute similarity calculation are performed on the carbon source and sink entities to obtain the attribute relationship.

[0137] The process of modeling the relationship between carbon source and sink entities is as follows:

[0138] Establishing conceptual relationships: First, for each carbon source / sink entity, identify its hierarchical structure within the functional classification system. By analyzing the semantic role of the entity in the carbon cycle system, construct conceptual relationships such as "belongs to (is_a)," "belongs to (part_of)," and "transforms to (transform_to)." For example, in the ontology model, establish an "is_a" relationship between "thermal power plant" and "stationary emission source," and establish a "part_of" relationship between "forest" and "terrestrial carbon sink system." These conceptual relationships are formally modeled using the Web Ontology Language (OWL), and the modeling results are stored in the semantic layer of the graph.

[0139] Constructing Spatial Relationships: Based on the acquired entity boundary and spatial coordinate data, GIS topology analysis methods are used to identify the geospatial relationships between entities. The spatial extent of entities is mapped using a grid and buffer zones are calculated, determining whether they meet spatial conditions such as adjacency (adjacent to), influence (influence range), or contain (contain by). For example, remote sensing image analysis can be used to determine whether an industrial source is adjacent to a surrounding forest carbon sink area; or the impact range of a road on urban green space can be identified based on the road buffer zone. This step relies on the fusion processing of remote sensing inversion results and entity boundary data to generate spatial edge connections and update them in real time.

[0140] Here, buffer calculation refers to the calculation of buffers based on entity boundaries and spatial coordinate data, using buffer analysis tools in GIS software (such as ArcGIS and QGIS).

[0141] The specific steps are as follows: After importing the vector boundary data of the entity, call the buffer function, set the buffer distance parameter, and the corresponding buffer polygon will be generated. The buffer represents the spatial influence range of the entity, which is convenient for subsequent spatial overlay analysis to determine the adjacency, containment and influence relationships between entities.

[0142] Establishing temporal relationships: For entities with temporal evolution characteristics, based on historical monitoring data and management event records, their temporal evolution paths are extracted, and relationship types such as "time_sequence," "lag_effect," and "periodic_pattern" are constructed according to the time axis. For example, starting from the carbon tax implementation date, the time delay of carbon flux changes in the region is compared to determine whether there is a significant lag response relationship. This process combines time slicing technology to establish a nested temporal relationship network to express dynamic evolution patterns at different scales.

[0143] Establishing attribute relationships: In the attribute relationship modeling process, the static features and dynamic indicators of entities are first extracted based on their original attribute data. Static features include information that does not change over time, such as entity area, functional type, and land use pattern. Dynamic indicators include remote sensing inversion parameters (such as NDVI and LST) and observational variables (such as carbon emissions and CO2 concentration). All attributes are uniformly encoded into vector form, let entity e... i The attribute vector is represented as Where d represents the attribute dimension.

[0144] To measure the attribute similarity between entities, cosine similarity is used to calculate the similarity score sim(e) between any two entity attribute vectors. i ,e j), defined as follows:

[0145]

[0146] Among them, e i With e j Representing two entities, and This represents the corresponding attribute vector, which measures the attribute similarity between entities. i With e j These are two entities whose attribute similarity needs to be measured. and It is its corresponding attribute vector.

[0147] When the similarity value is greater than or equal to the set threshold τ sim When an edge is defined as having a "similarity relationship" between entities, it is denoted as:

[0148]

[0149] like Figure 5 As shown, the extraction of carbon source-sink entity relationships is carried out through the following process:

[0150] For extracting conceptual relationships: First, unstructured text corpora from the carbon source and sink domain, including pre-defined documents, research reports, and standard documents, are collected and input into a Large Language Model (LLM) module capable of relationship extraction. Next, standardized prompt word templates are constructed, whose structure includes: semantic context, extraction task type (e.g., "identify subclasses" or "identify inclusion relationships"), expected output structure, and example samples.

[0151] For spatial relationship identification: After semantic extraction, the system further calls the spatial structure data interface to obtain the vector boundary files and corresponding geographic coordinate data of each carbon source and sink entity (such as green space, street block, urban area, etc.). Based on the spatial location relationship between entities, a plot-level spatial topology determination method is used to identify whether there is an adjacency relationship or containment relationship between entities.

[0152] Specifically, when two entities share a common line segment or vertex, the system determines them to be adjacent and adds a "spatial adjacency" edge to the graph; when all boundary points of an entity (such as a green space) are located inside another entity (such as a street block), and the two boundaries do not intersect or overlap, they are determined to be contained and a "spatial containment" edge is added to the graph.

[0153] For the temporal and attribute relationships of spatiotemporally changing entities, the system constructs time-series relationships and attribute coupling relationships based on observation data of carbon source and sink entities at different time scales. This describes the dynamic evolution of entities in the time dimension and the linkage patterns between multiple attributes. The process includes two steps: temporal relationship modeling and attribute relationship identification, which are described in detail below:

[0154] 1. Temporal Relationship Modeling: First, extract the observed attribute sequences of carbon source and sink entities at multiple time points, denoted as t1, t2, ..., t n Each time point corresponds to an entity state snapshot, including core attribute indicators such as current carbon emissions, flux intensity, and stationary capacity. The system maps these time-series states to graph nodes and constructs two types of temporal semantic edges: one type is "belonging to time point t". i The state-assigned edge is used to represent the association between a snapshot of a certain attribute and a time node; another type is "the next state is t". i+1 A directed time evolution edge is used to describe the path of an entity as it evolves from its current time state to the next time state.

[0155] To further model the evolution trend of attributes over time, the system analyzes each attribute sequence. The ARIMA (Autoregressive Integrated Moving Average) model is applied for fitting and prediction. The standard form of this model is as follows:

[0156]

[0157] Among them, y t Let c represent the attribute value at time t, p be the order of the autoregressive (AR) term, q be the order of the moving average (MA) term, and φ be the value of the attribute at time t. i θ is the coefficient of the autoregressive term. j The coefficient of the moving average term, ∈ t-j Error term from past time, ∈ t This is a Gaussian white noise term.

[0158] To select the optimal combination of ARIMA model orders (p, d, q), where d represents the difference order (how many differencing orders are performed on the original sequence to achieve stationarity), the system uses the Akaike Information Criterion (AIC) as the model selection metric. The formula for calculating AIC is as follows:

[0159] AIC = 2k - 2ln(L);

[0160] The definition of AIC is essentially a balance between the overall goodness of fit and complexity of a model. It relies on the model's log-likelihood function value L, rather than explicitly writing out each observation. Here, k represents the number of free parameters in the model, and L is the model's fit value under maximum likelihood estimation. The system iterates through multiple model combinations within the candidate parameter range, calculates the corresponding AIC values, and selects the model with the smallest AIC as the optimal fit for the target attribute.

[0161] If the fitting results show a significant lag effect, the system adds a "temporal dependency" semantic edge to the entity in the knowledge graph to characterize the dynamic regularity of the entity's attributes in the time dimension.

[0162] 2. Attribute Relationship Identification: First, to eliminate differences in magnitude and units among different attributes, a normalization process is performed on each attribute sequence using the following linear normalization formula:

[0163]

[0164] Where, x i This represents the original value of the attribute (i.e., the attribute observation value before normalization), x' i Let be the normalized i-th observation value, and min(x) and max(x) represent the minimum and maximum values ​​of the attribute in all samples, respectively. The normalized attribute values ​​are limited to the interval [0, 1].

[0165] Furthermore, the present invention applies each pair of entity attribute vectors (such as...) The cosine similarity is calculated to quantify the directional consistency between the two pairs. The formula is as follows:

[0166]

[0167] Among them, a i With b i Representing attribute vectors respectively The normalized value in the i-th dimension.

[0168] When the similarity result sim>0.85, it is considered that there is a significant similarity or coupling relationship between the attribute pairs. The system will add an "attribute-related" semantic edge to the graph and label the relationship type as "high similarity".

[0169] Step S30: Perform graph mapping processing on the carbon source and sink entity features and carbon source and sink entity relationships to obtain graph nodes and graph edges, and construct a graph substructure based on the carbon source and sink entities, the graph nodes and the graph edges.

[0170] The above features (i.e., carbon source and sink entity features) and relationships (carbon source and sink entity relationships) are mapped to graph nodes and edges, and a graph substructure containing state nodes, spatial nodes, time nodes and attribute nodes is constructed.

[0171] Specifically, the semantic features, spatial features, temporal features, and attribute features are subjected to graph mapping processing to obtain multiple graph nodes; the conceptual relationships, spatial relationships, temporal relationships, and attribute relationships are subjected to graph mapping processing to obtain multiple graph edges; and graph construction processing is performed based on the carbon source-sink entities, the multiple graph nodes, and the multiple graph edges to obtain a graph substructure.

[0172] Step S40: Perform storage update processing on the graph substructure to obtain a carbon source-sink spatiotemporal dynamic knowledge graph.

[0173] The time nodes are serialized based on their temporal and spatial evolution characteristics, driving the internal state updates and switching of entity nodes, thereby enabling dynamic adjustment and precise construction of the attribute characteristics of carbon source and sink entities under different time and space conditions.

[0174] Specifically, when new carbon source and sink data is received, a state node chain is constructed based on the new carbon source and sink data.

[0175] like Figure 6 As shown, the process for storing and updating carbon source and sink entities is as follows:

[0176] 1. Constructing a state node chain: For each newly added carbon source / sink data, extract its timestamp, spatial location, and attribute values, and determine whether it belongs to an existing entity's evolutionary sequence. If a historical state node already exists, add a new current state node and connect the previous and subsequent states through "change relationship" edges; if it is the first time it is added, create an evolutionary chain with the current state as the initial node.

[0177] The spatiotemporal structure and attribute structure in the graph substructure are determined, and the graph substructure is three-dimensionally linked with the state node chain according to the spatiotemporal structure and attribute structure to obtain the initial carbon source sink knowledge graph.

[0178] 2. Linking Spatiotemporal and Attribute Information: Each state node is bound to its corresponding time point, spatial location, and attribute value. After adding a new state node, it needs to be aligned and completed with the existing spatiotemporal and attribute structures in the graph to ensure data consistency and integrity.

[0179] In the time dimension, the timestamp information of the state nodes is extracted and matched with existing time nodes in the graph. If the time node does not yet exist, a corresponding time node is added, and a pointing relationship is established between the state node and the time node to achieve structural docking of time information.

[0180] In the spatial dimension, the spatial coordinates or addresses carried by the status nodes are compared with the spatial entity nodes already constructed in the graph. If the spatial location falls within the range of an existing spatial unit, it is attached to that spatial entity node; if no match is found, a new spatial entity node is added and a link is established.

[0181] At the attribute level, based on predefined attribute field mapping rules, the original field names in the state nodes are mapped one-to-one with the standard attribute fields in the graph. When field types are inconsistent, type conversion is performed according to the graph attribute structure. If a state node is missing some key attribute fields, the historical values ​​of those fields are extracted from its previous time-series state node, or the set default values ​​are used to supplement them, thus completing the attribute completion.

[0182] After completing the above three-dimensional connection, if the attribute value or spatial position of the current state node differs from that of the previous node, it is determined to be an evolutionary behavior, and the previous and previous state nodes are linked through the "change relationship" edge; if the spatial position changes, its connection relationship with the spatial entity node is updated synchronously to ensure that the spatial reference is correct.

[0183] The initial carbon source and sink knowledge graph is subjected to dynamic incremental writing and dynamic tracking processing to obtain a spatiotemporal dynamic knowledge graph of carbon sources and sinks.

[0184] 3. Dynamic Incremental Writing: In scenarios involving continuous access to observation data, a dynamic incremental writing mechanism is employed. Only state nodes whose attributes or spatial information have changed are compared with the previous state node and written. For items without changes, no write operation is performed. After each write, a changed edge is established to connect with the previous state node, keeping the original historical data unchanged and ensuring the continuity and traceability of the state chain.

[0185] 4. Temporal Relationship Backtracking and Reasoning: Dynamic tracking of carbon source and sink entities is achieved through the temporal topology and path retrieval functions of the graph database. First, based on the temporal topology, a path retrieval algorithm is used to track the evolution trajectory of any entity at different time points, obtaining its attribute change paths and state transition processes. Specifically, the path retrieval algorithm connects the state nodes of the target entity according to the temporal order based on the time edges in the graph database, forming a temporal path of entity evolution. Assume the entity's state node is S(t... i ), where t i Let S(t) represent the i-th time point. j ) represents the entity at time point t in the j-th time period. j The state nodes. The path retrieval algorithm calculates the path length P(S(t) between state nodes. i ),S(t jThe process of state transition is represented by )). During path retrieval, the system gradually establishes the attribute change path of an entity at different times based on sequential edges, periodic edges, and trend edges. By comparing the states before and after, the system infers the triggering factors of state changes and their spatiotemporal evolution laws.

[0186] Furthermore, by combining historical monitoring data, a causal inference algorithm is used to trace the causal origins of attribute changes and state transitions. The causal inference algorithm first extracts the historical monitoring dataset {D(t1), D(t2), ..., D(t...}}. n )}, where D(t) i (t) represents time point t i The algorithm uses observed attribute data. Based on a causal graph model, it constructs a causal relationship network between attribute changes to infer the attribute change ΔA(t) at a certain time point. i Is it affected by the previous state S(t)? i-1 ) or external factors E(t) i The impact of ). Specifically, the causal inference model can be expressed as:

[0187] ΔA(t i )=f(S(t i-1 ), E(t) i ), ∈);

[0188] Wherein, ΔA(t) i ) represents attribute changes, f is the causal inference function, and ∈ represents the error term. This inference process helps identify key factors affecting changes in the state of an entity.

[0189] Furthermore, such as Figure 7 As shown, based on the above-mentioned method for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks, this invention also provides a system for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks, wherein the system for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks includes:

[0190] The data preprocessing module 51 is used to acquire multi-source sensing data and preprocess the multi-source sensing data to obtain preprocessed multi-source sensing data.

[0191] The entity feature and relationship extraction module 52 is used to perform entity modeling processing on the preprocessed multi-source sensing data to obtain carbon source sink entities, and to perform feature modeling processing and relationship modeling processing on the carbon source sink entities to obtain carbon source sink entity features and carbon source sink entity relationships.

[0192] The initial graph construction module 53 is used to perform graph mapping processing on the carbon source and sink entity features and carbon source and sink entity relationships to obtain graph nodes and graph edges, and to construct graph substructures based on the carbon source and sink entities, the graph nodes and graph edges;

[0193] The graph dynamic update module 54 is used to perform storage update processing on the graph substructure to obtain a carbon source-sink spatiotemporal dynamic knowledge graph.

[0194] Furthermore, such as Figure 8 As shown, based on the above-mentioned method and system for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 8 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0195] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a spatiotemporal dynamic knowledge graph construction program 40 for carbon sources and sinks, which can be executed by the processor 10 to implement the spatiotemporal dynamic knowledge graph construction method for carbon sources and sinks in this application.

[0196] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing the spatiotemporal dynamic knowledge graph construction method of the carbon source sink.

[0197] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface.

[0198] In one embodiment, the steps of the method for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks are implemented when the processor 10 executes the spatiotemporal dynamic knowledge graph construction program 40 of the memory 20.

[0199] In summary, this invention provides a method, system, and terminal for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks. The method includes: acquiring multi-source sensing data and preprocessing the multi-source sensing data to obtain preprocessed multi-source sensing data; performing entity modeling processing on the preprocessed multi-source sensing data to obtain carbon source and sink entities, and performing feature modeling and relation modeling processing on the carbon source and sink entities to obtain carbon source and sink entity features and carbon source and sink entity relations; performing graph mapping processing on the carbon source and sink entity features and carbon source and sink entity relations to obtain graph nodes and graph edges, and constructing a graph substructure based on the carbon source and sink entities, the graph nodes, and the graph edges; and performing storage update processing on the graph substructure to obtain a spatiotemporal dynamic knowledge graph of carbon sources and sinks. This invention acquires multi-source sensing data, constructs carbon source and sink entity features and relationships based on the multi-source sensing data, and then constructs a graph substructure based on the carbon source and sink entity features and relationships. Furthermore, it introduces a time evolution mechanism to form a spatiotemporal dynamic knowledge graph of carbon sources and sinks, which can effectively improve the estimation accuracy of carbon sources and sinks.

[0200] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.

[0201] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.

[0202] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks, characterized in that, The spatiotemporal dynamic knowledge graph construction method for carbon sources and sinks includes: Acquire multi-source sensing data and preprocess the multi-source sensing data to obtain preprocessed multi-source sensing data; The preprocessed multi-source sensing data is subjected to entity modeling to obtain carbon source sink entities. The carbon source sink entities are then subjected to feature modeling and relation modeling to obtain carbon source sink entity features and carbon source sink entity relations. The step of performing entity modeling on the preprocessed multi-source sensing data to obtain carbon source sink entities, and then performing feature modeling and relation modeling on the carbon source sink entities to obtain carbon source sink entity features and carbon source sink entity relationships, specifically includes: The preprocessed multi-source sensing data is judged according to entity function to obtain carbon source sink entities, wherein the carbon source sink entities include carbon source entities and carbon sink entities; Carbon source entities are classified into industrial emission facilities, transportation systems and agricultural activity units according to emission type. Carbon sink entities are classified into forest ecosystems, wetland systems, and marine carbon sink areas based on their ecological functions and spatial affiliation. The carbon source sink entity is subjected to feature modeling processing to obtain carbon source sink entity features, wherein the carbon source sink entity features include semantic features, spatial features, temporal features and attribute features; Relationship modeling is performed on the carbon source and sink entities to obtain carbon source and sink entity relationships, wherein the carbon source and sink entity relationships include conceptual relationships, spatial relationships, temporal relationships, and attribute relationships; The step of performing feature modeling on the carbon source and sink entities to obtain carbon source and sink entity features specifically includes: The semantic features are obtained by performing entity name normalization, entity type code encoding, and entity identifier code construction on the carbon source sink entity. The spatial features are obtained by extracting latitude and longitude coordinates and determining spatial representation methods for the carbon source and sink entities. The carbon source and sink entities are subjected to time series table construction and time topology construction to obtain the time features; The carbon source and sink entities are subjected to entity attribute assignment processing to obtain the attribute features; The step of performing relational modeling on the carbon source and sink entities to obtain the carbon source and sink entity relationships specifically includes: The concept relationship is obtained by performing hierarchical structure identification and semantic role parsing on the carbon source and sink entities. The spatial relationships of the carbon source and sink entities are obtained by using GIS topology analysis methods to identify geospatial relationships and determine spatial conditions. The time relationship is obtained by performing time-series evolution path extraction on the carbon source and sink entities. The attribute relationships are obtained by performing attribute feature extraction and attribute similarity calculation on the carbon source and sink entities. The carbon source and sink entity features and carbon source and sink entity relationships are processed by graph mapping to obtain graph nodes and graph edges, and a graph substructure is constructed based on the carbon source and sink entities, the graph nodes and graph edges; The storage and update process of the graph substructure is performed to obtain a spatiotemporal dynamic knowledge graph of carbon sources and sinks.

2. The method for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks according to claim 1, characterized in that, The multi-source sensing data includes structured data and unstructured text corpus data; the preprocessed multi-source sensing data includes target structured data and target text corpus data; The acquisition of multi-source sensing data and the preprocessing of the multi-source sensing data to obtain preprocessed multi-source sensing data specifically include: The structured data is acquired from the multi-source sensing data, wherein the structured data includes remote sensing image data, ground sensing monitoring data, meteorological and carbon flux statistics data, and socio-economic activity data; The structured data is processed by data completion, standardization, outlier handling, and deduplication to obtain the target structured data. The unstructured text corpus data in the multi-source perception data is obtained, and entity information extraction processing is performed on the unstructured text corpus data to obtain the target text corpus data.

3. The method for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks according to claim 1, characterized in that, The step of performing graph mapping processing on the carbon source and sink entity features and carbon source and sink entity relationships to obtain graph nodes and graph edges, and constructing a graph substructure based on the carbon source and sink entities, the graph nodes, and the graph edges, specifically includes: The semantic features, spatial features, temporal features, and attribute features are subjected to graph mapping processing to obtain multiple graph nodes; The conceptual relationships, spatial relationships, temporal relationships, and attribute relationships are subjected to graph mapping processing to obtain multiple graph edges; The graph is constructed based on the carbon source and sink entities, multiple graph nodes, and multiple graph edges to obtain a graph substructure.

4. The method for constructing a spatiotemporal dynamic knowledge graph of carbon sources and sinks according to claim 1, characterized in that, The process of storing and updating the substructure of the graph to obtain a spatiotemporal dynamic knowledge graph of carbon sources and sinks specifically includes: When new carbon source and sink data is received, a state node chain is constructed based on the new carbon source and sink data; The spatiotemporal structure and attribute structure in the graph substructure are determined, and the graph substructure is three-dimensionally linked with the state node chain according to the spatiotemporal structure and the attribute structure to obtain the initial carbon source sink knowledge graph; The initial carbon source and sink knowledge graph is subjected to dynamic incremental writing and dynamic tracking processing to obtain a spatiotemporal dynamic knowledge graph of carbon sources and sinks.

5. A spatiotemporal dynamic knowledge graph construction system for carbon sources and sinks, characterized in that, The spatiotemporal dynamic knowledge graph construction system for carbon sources and sinks is used to implement the spatiotemporal dynamic knowledge graph construction method for carbon sources and sinks as described in any one of claims 1-4, wherein the spatiotemporal dynamic knowledge graph construction system for carbon sources and sinks comprises: The data preprocessing module is used to acquire multi-source sensing data and preprocess the multi-source sensing data to obtain preprocessed multi-source sensing data. The entity feature and relationship extraction module is used to perform entity modeling processing on the preprocessed multi-source sensing data to obtain carbon source sink entities, and to perform feature modeling processing and relationship modeling processing on the carbon source sink entities to obtain carbon source sink entity features and carbon source sink entity relationships. The initial graph construction module is used to perform graph mapping processing on the carbon source and sink entity features and carbon source and sink entity relationships to obtain graph nodes and graph edges, and to construct graph substructures based on the carbon source and sink entities, the graph nodes and graph edges; The graph dynamic update module is used to store and update the graph substructure to obtain a carbon source-sink spatiotemporal dynamic knowledge graph.

6. A terminal, characterized in that, The terminal includes: a memory, a processor, and a spatiotemporal dynamic knowledge graph construction program for carbon sources and sinks stored in the memory and executable on the processor. When the spatiotemporal dynamic knowledge graph construction program for carbon sources and sinks is executed by the processor, it implements the steps of the spatiotemporal dynamic knowledge graph construction method for carbon sources and sinks as described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a spatiotemporal dynamic knowledge graph construction program for carbon sources and sinks. When the spatiotemporal dynamic knowledge graph construction program for carbon sources and sinks is executed by a processor, it implements the steps of the spatiotemporal dynamic knowledge graph construction method for carbon sources and sinks as described in any one of claims 1-4.

Images