Methods and apparatus for integrated analysis of spatial big data and knowledge graphs for natural resources

Abstract

This invention discloses a method and apparatus for comprehensive analysis of spatial big data and knowledge graphs for natural resources, belonging to the interdisciplinary field of natural resource management and artificial intelligence. The method includes: acquiring multi-source data and preprocessing it to form a standardized dataset; constructing a natural resource domain knowledge graph containing land parcel entities (with vector boundary coordinate sequences), simplifying coordinates and automatically discovering spatial relationships using the Douglas-Puk algorithm; utilizing a hybrid reasoning engine integrating symbolic reasoning, uncertainty reasoning, and spatiotemporal reasoning, outputting conclusion confidence levels through a weighted fusion formula; performing scenario-driven deduction and outputting conclusions and influence chains; and dynamically adjusting the knowledge graph confidence level and reasoning rule weights based on feedback. This invention solves the problems of insufficient spatial and semantic fusion and single reasoning modality in existing technologies, improving the scientific rigor, foresight, and adaptability of natural resource analysis, and providing reliable decision support for scenarios such as resource planning and conflict detection.

Description

Technical Field

[0001] This invention relates to the field of comprehensive analysis technology of natural resources, and in particular to a method and apparatus for comprehensive analysis of spatial big data and knowledge graphs for natural resources. Background Technology

[0002] Natural resource management involves various resource elements such as land, minerals, forests, and water, and its data is characterized by multi-source heterogeneity, spatiotemporal dynamics, and complex relationships. With the development of big data and artificial intelligence technologies, knowledge graph technology, due to its powerful semantic association and logical expression capabilities, is widely used to build a "knowledge brain" in the field of natural resources, integrating fragmented data to form a structured knowledge system. Meanwhile, Large Language Models (LLM) have demonstrated powerful capabilities in information extraction and semantic understanding of unstructured texts (such as policy documents and survey reports), becoming an important tool for constructing high-quality domain knowledge graphs.

[0003] Current technologies for applying knowledge graphs to in-depth analysis and forward-looking decision support for natural resources still face a series of key bottlenecks. First, there is insufficient integration of spatial and semantic knowledge. Traditional knowledge graphs focus on semantic relationships between entities (such as "belongs to" and "located in"), but lack native and accurate representation and computational capabilities for spatial topological relationships (such as adjacency, inclusion, intersection, and buffer zone effects) and geometric features (such as plot shape and boundaries) that are crucial for natural resource management. Existing methods typically store spatial information simply as entity attributes, failing to support complex spatial queries and spatial reasoning. This limits the ability of knowledge graphs to solve core spatial problems such as "planning conflict detection" and "ecological red line encroachment analysis." Second, the reasoning modality is singular, making it difficult to handle complex uncertainties. Existing systems mostly employ deterministic reasoning based on symbolic logic or probabilistic reasoning based on statistical models, resulting in a disconnect between the two. Natural resource issues are rife with incomplete data, fuzzy rules, and random factors (such as meteorological disasters and market fluctuations). A single reasoning modality cannot simultaneously guarantee logical rigor and quantitative assessment of uncertainty, leading to biased conclusions or insufficient confidence.

[0004] Therefore, in view of this, the inventors have researched and improved the existing technology and its shortcomings, and provided a new solution for intelligent analysis and decision support of natural resources that integrates spatial big data and knowledge graph for comprehensive analysis of natural resources. This solution can deeply integrate spatial and semantic knowledge, integrate multimodal reasoning, support interactive scenario inference, and has self-optimization capabilities. Summary of the Invention

[0005] To address the problems mentioned in the background section, this invention provides a method and apparatus for comprehensive analysis of spatial big data and knowledge graphs for natural resources.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for comprehensive analysis of spatial big data and knowledge graphs for natural resources, comprising the following steps: S1: Acquire and preprocess spatial big data of natural resources, socio-economic auxiliary data and unstructured archival data to form standardized spatiotemporal datasets and structured archival metadata; S2: Based on the output of S1, extract entities, attributes, and relationships to construct a knowledge graph in the natural resources domain; the entities in the knowledge graph include land parcel entities with vector boundary coordinate sequences; S3: Utilize the constructed knowledge graph to perform reasoning computation through a hybrid reasoning engine that integrates symbolic reasoning, uncertainty reasoning, and spatiotemporal reasoning; S4: Receive scenario conditions input by the user or cases parsed from historical archives, and based on the knowledge graph and the hybrid reasoning engine, perform simulation deduction and output the deduction conclusion and the chain of influence; S5: Based on the feedback and application verification of the inference results, dynamically adjust the confidence level of the facts in the knowledge graph and the weight parameters of the rules in the hybrid inference engine; The hybrid reasoning engine in S3 generates conclusions based on a weighted fusion of symbolic reasoning results and uncertain reasoning results. The fusion calculation process is as follows: CF final (C)=α·I(CS)+(1-α)·P(C|E); Among them, CF final (C) indicates the final confidence level of conclusion C; I(CS) represents the symbolic reasoning function, which takes the value of 1 when there is at least one rule that satisfies all the preconditions in the current knowledge graph, and 0 otherwise. The preconditions include factual conditions and spatiotemporal constraints. P(C|E) represents the conditional probability of conclusion C calculated based on evidence E using an uncertainty reasoning model; α is the fusion coefficient, used to balance the contributions of the two inference modes, α∈[0,1].

[0007] Preferably, the knowledge graph construction in S2 includes constructing a sequence of boundary coordinates of land parcel entity vectors, P'={p1,p2,…,p...}. n Geometric feature-preserving simplification is performed, specifically using the Douglas-Puk algorithm for recursive compression: Starting from the first point P of the line segment start With the final point P end Construct a baseline segment and calculate all intermediate points P. i The perpendicular distance d to the line segment i ; Determine the maximum distance d max =MAX(d i ) and its corresponding point P index ; If d max If it exceeds the preset tolerance threshold ε, then P will be... index Mark them as key points, and then perform operations on line segments [P] respectively. start ,P index ] and [P index ,P end The above process is executed recursively; Otherwise, discard all intermediate points and keep only P. start With P end ; Merge all key points and start and end points marked during the recursion process, arrange them in the original order, and generate a simplified coordinate point set P' in sequence, preserving the original geometric and topological features.

[0008] The method for further determining the preset tolerance threshold ε includes any one or more combinations of the following: Determined based on the conversion relationship between the scale accuracy of the original spatial data and the actual accuracy; Based on the target business scenario, the threshold is matched and obtained from the pre-set tolerance threshold configuration table corresponding to different analysis granularities; Calculate the global or local geometric feature metric values ​​of the coordinate sequence, and determine the tolerance threshold ε using its preset ratio; Through experimental verification, the ε value that optimizes the number of simplified coordinates under the preset geometric fidelity constraint was selected.

[0009] Preferably, the construction of the knowledge graph in S2 further includes automatically discovering spatial relationships between entities based on the simplified coordinate point set P': For two land parcel entities A and B, construct polygons using their key coordinate point sets, and calculate their areas Area(A) and Area(B), as well as the area of ​​the intersection region Area(A∩B). Calculate the intersection-union ratio (IoU) of two entities: IoU(A,B) = Area(A∩B) / [Area(A) + Area(B) - Area(A∩B)]; If IoU(A,B)≥θ, where θ is a preset spatial alignment threshold, a semantic relationship edge from entity A to entity B is created in the knowledge graph. The relationship type is "spatial overlap". The edge is given the attribute {'IoU':IoU(A,B)} to accurately represent the degree of spatial overlap between the land parcel entities A and B in the knowledge graph. At least one rule in the hybrid inference engine contains threshold judgment logic for the quantified attribute 'IoU' in its condition part. The uncertainty inference model uses the IoU(A,B) value as input evidence to calculate the conditional probability P(C|E).

[0010] Preferably, the rules used by the hybrid inference engine in S3 are formally represented using spatiotemporally enhanced triples, and their general structure is as follows: IF <condition>THEN <conclusion>WITHweight=W R &ST_Constraint(); in, <condition>and <conclusion>A basic fact triple consisting of an entity, an attribute, and a value, or a logical combination thereof. W R These are the weight parameters for the rule; ST_Constraint() is a spatiotemporal constraint function, which includes at least one of the following forms of computation: WithinBuffer(e1,e2,r) determines whether the geometric center of entity e1 is located within a buffer with radius r centered at e2. TimeAfter(t1,t2) determines whether time point t1 is after time point t2; Contains(Region,e) determines whether the geometric boundary of the administrative division entity Region completely contains the geometric boundary of the land parcel entity e. During rule retrieval, the hybrid reasoning engine constructs a multi-dimensional inverted index based on conclusion type, entity type, and spatiotemporal constraint fingerprint, and uses the spatial index generated by S2 to pre-filter candidate entities, accelerating the variable binding process.

[0011] Furthermore, the hybrid inference engine is configured to perform the following steps when executing rule retrieval and variable binding: (a) Construct a multi-dimensional inverted index based on the conclusion type of the rule, the type of the entity involved, and the type of the spatiotemporal constraint function; (b) After receiving the reasoning target or triggering fact, the multi-dimensional inverted index is used for initial screening to obtain a set of candidate rules; (c) For the rules in the candidate rule set, the candidate entity set corresponding to the entity variables in the rule conditions is pre-filtered using the spatial index generated in step S2; (d) Perform precise variable binding and condition validation in the pre-filtered candidate entity set.

[0012] Preferably, the scenario-driven deduction step in S4 specifically includes: S4.1: Define the initial state of the knowledge graph as KG t The system receives a set I of intervention operations input by the user, the intervention operations including those related to KG. t Modifying entity attribute values, adding or deleting entities or relationships; S4.2: Apply intervention to generate the deduced knowledge graph state KG t+1 =Apply(KG t ,I); S4.3: KG t and KG t+1 The inputs are respectively fed into the hybrid inference engine, inference calculations are performed, and the inference result fact set F is obtained. t and F t+1 ; S4.4: Calculate and output the inferred impact difference set ΔF=F t+1 \F t It also visualizes the state transition path, the impact propagation chain triggered by intervention I, and the complete traceability link from the conclusion to the original data source.

[0013] Preferably, the system closed-loop optimization step in S5 includes adaptive adjustment of rule weights, specifically: For a rule R that is triggered and used to derive a conclusion, its weight parameter W is updated according to the feedback signal δ of that conclusion. R The updated formula is: ; Where η is the learning rate, which controls the step size of weight updates; the feedback signal δ∈[0,1], when the inference conclusion is confirmed by the user or highly consistent with the actual situation, δ approaches 1; when the conclusion is corrected or inconsistent with the actual situation, δ approaches 0.

[0014] Secondly, this invention provides a natural resource intelligent analysis and decision-making system for implementing the above-described method. The system includes: a data fusion and governance module, a knowledge graph construction and management module, a multimodal hybrid reasoning engine module, a scenario inference and decision support module, and a closed-loop optimization module. Each module corresponds one-to-one with the steps of the above method and works collaboratively.

[0015] Thirdly, the present invention provides an electronic device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described in the first aspect.

[0016] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in the first aspect.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention effectively solves four core challenges in natural resource management: fragmented knowledge representation, single reasoning modality, static decision support, and rigid system intelligence. Through "geometric refinement and automatic spatial relationship discovery" technology, it achieves for the first time the native fusion of semantic knowledge and spatial topological relationships in a knowledge graph, enabling the system to possess accurate spatial analysis and reasoning capabilities. The innovatively proposed multimodal hybrid reasoning engine and confidence fusion formula deeply combine the rigor of symbolic logic with the flexibility of probabilistic reasoning, significantly improving the reliability of conclusions in complex and uncertain scenarios. Its scenario-driven deduction function constructs an interactive "digital sandbox," supporting "hypothesis analysis" simulations of policies and planning schemes, intuitively revealing the chain reactions of decisions, and realizing a paradigm shift from passive analysis to proactive prediction. Finally, it introduces a feedback-based closed-loop optimization mechanism, enabling the system to continuously learn from practice, dynamically optimize knowledge quality and reasoning strategies, and become a constantly evolving "growth-oriented" intelligent agent. This solution forms a complete closed loop from data perception to decision optimization, greatly enhancing the scientific rigor, foresight, and adaptability of natural resource management. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a method for comprehensive analysis of spatial big data and knowledge graphs for natural resources according to the present invention. Detailed Implementation

[0020] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. 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.

[0021] Example 1

[0022] This invention provides a method for comprehensive analysis of spatial big data and knowledge graphs for natural resources, comprising the following steps: S1. Acquire and preprocess multi-source data to form a standardized spatiotemporal dataset with structured traceability metadata and structured archive metadata. The multi-source data includes unstructured archives, spatial big data, and socio-economic auxiliary data. In this embodiment, the system acquires the following three types of data: Unstructured archives: 500 policy documents related to natural resource management, 300 survey reports, and 200 historical records from 2010 to 2023; Spatial Big Data: Land Use Status Map of 2023 (10-meter resolution), Sentinel-2 Satellite Imagery (acquired in October 2023), and DEM Data (30-meter resolution). Socioeconomic supplementary data: 2023 census data, GDP data of each administrative region, and output data of each industry.

[0023] S1.1 Establish a multi-source data preprocessing model to perform text semantic embedding processing on unstructured archives: After the archive text is segmented into sentences, it is mapped into pre-trained semantic vectors. The semantic association blocks are divided by density clustering algorithm (DBSCAN) to generate a set of semantic blocks with topic labels.

[0024] S1.2 Extract core topic keywords for each semantic block, calculate the topic similarity (Jaccard coefficient) and semantic vector cosine similarity of adjacent semantic blocks, and weight them to obtain a comprehensive similarity and form a semantic block association matrix.

[0025] S1.3 When the overall similarity of adjacent semantic blocks is greater than a preset threshold (≥0.7), the semantic blocks are merged to generate a sub-document, and the topic tags, overall similarity threshold and the range of semantic blocks covered by the sub-document are recorded.

[0026] S1.4 Generate a semantic feature fingerprint for each sub-document (based on the merging of the hash value of the topic keyword and the average hash value of the semantic vector), associate the semantic anchor point of the sub-document (the position of the first sentence of the sub-document + the core topic word), the page number and chapter identifier of the source archive, generate structured source traceability metadata, and form a "sub-document-semantic anchor point-feature fingerprint-source archive" traceability chain.

[0027] S1.5 Perform coordinate unification and spatial texture feature enhancement processing on spatial big data, and perform temporal normalization (including multi-period data) and feature standardization processing on socio-economic auxiliary data to integrate and form a standardized spatiotemporal dataset.

[0028] S2: Based on the standardized spatiotemporal dataset, extract entities, attributes and relationships in the natural resources field, and construct a knowledge graph that integrates spatiotemporal constraints and multimodal features. The entities in the knowledge graph include land parcel entities with vector boundary coordinate sequences. In this embodiment, 1000 land parcel entities are processed, and their vector boundary coordinate sequence contains 1000 points.

[0029] S2.1, For the entity vector boundary coordinate sequence P=p1,p2,…,p n The Douglas-Puk algorithm is used for recursive compression to generate a simplified set of key coordinate points P'.

[0030] S2.2 Extract the area value, coordinate system type, and certificate issuance date attribute of the land parcel entity, convert the certificate issuance date into a timestamp format, and bind it with the coordinate system type to form a spatial-temporal dual constraint.

[0031] S2.3 Extract multimodal features from remote sensing images and CAD drawings, and convert them into standardized feature vectors as additional attributes of land parcel entities.

[0032] S2.4. Construct a boundary polygon based on the key coordinate point set P', and calculate the intersection-union ratio (IoU) (A,B) of the two land parcel entities through a boundary contour optimization and verification mechanism. The IoU calculation formula is: ; in, Let Area(A) and Area(B) be the areas of the overlapping regions, respectively, and let them be the envelope areas of the two entity outlines. ( When the spatial alignment threshold is set to 0.3 in this embodiment, a "spatial overlap" relationship edge is created and an intersection-union ratio attribute is attached.

[0033] S2.4.1 Construct the boundary polygons of the two land parcel entities using the key coordinate point set.

[0034] S2.4.2 Traverse the key coordinate points of the boundary polygon, identify the outermost boundary point set and connect them to form the initial contour envelope.

[0035] S2.4.3 Calculate the error rate between the area of ​​the initial contour envelope and the area of ​​the original boundary polygon. If the error rate is less than the preset contour simplification tolerance threshold (2%), the initial contour envelope is deemed valid; otherwise, the original boundary polygon is used as the final contour envelope.

[0036] S2.4.4 Calculate the area of ​​the overlapping region and the area of ​​the outlines of the two entities based on the effective contour envelope, and substitute them into the cross-union ratio calculation formula to calculate the cross-union ratio.

[0037] S3. Using the knowledge graph as input, perform reasoning calculations through a hybrid reasoning engine that integrates symbolic reasoning, uncertainty reasoning, and spatiotemporal reasoning. The hybrid reasoning engine employs optimized spatial overlap determination logic. In this embodiment, the rules of the hybrid inference engine are formally represented using spatiotemporal enhanced triples, with the following structure: IF <condition>THEN <conclusion>WITHweight=W R &ST_Constraint(); in, <condition>and <conclusion>A basic fact triple consisting of an entity, an attribute, and a value, or a logical combination thereof. W R These are the weight parameters for the rule; ST_Constraint() is a spatiotemporal constraint function, which includes at least one of the following forms of computation: WithinBuffer(e1,e2,r) determines whether the geometric center of entity e1 is within the buffer of entity e2 with radius r. TimeAfter(t1,t2) determines whether time point t1 is after time point t2; The function `Contains(Region,e)` determines whether the geometric boundary of the administrative division entity `Region` contains the geometric boundary of the land parcel entity `e`.

[0038] For example, the rule "IF(Plot A is located in an ecological protection zone) AND(Plot A type = industrial land) THEN(Plot A has development risks) WITHweight=0.8&ST_Constraint(Contains(ecological protection zone, plot A))".

[0039] The hybrid reasoning engine generates conclusions based on a weighted fusion of symbolic reasoning results and uncertain reasoning results. The fusion calculation process is as follows: CF final (C)=α·I(CS)+(1-α)·P(C|E); Among them, CF final (C) indicates the final confidence level of conclusion C; I(CS) represents the symbolic reasoning function, which takes the value of 1 when there is at least one rule that satisfies all the preconditions in the current knowledge graph, and 0 otherwise. The preconditions include factual conditions and spatiotemporal constraints. P(C|E) represents the conditional probability of conclusion C calculated based on evidence E using an uncertainty reasoning model; α is the fusion coefficient, used to balance the contributions of the two inference modes, α∈[0,1].

[0040] S4. Receive scenario conditions or historical cases, perform simulation deduction based on the knowledge graph and hybrid reasoning engine, and output the deduction conclusion, the chain of influence and the tracing link. In this embodiment, the user inputs the scenario condition: "Add an industrial development zone in area A," and the following steps are executed: S4.1 Define the initial state of the knowledge graph as KG_t, and receive a set of intervention operations I input by the user. The intervention operations include modifying the attribute values ​​of entities in KG_t, and adding or deleting entities or relations.

[0041] S4.2 Applying intervention to generate the deduced knowledge graph state KG_{t+1}=A(I,KG t ).

[0042] S4.3 Input KG_t and KG_{t+1} into the hybrid inference engine respectively, perform inference calculation, and obtain the inference result fact set F_t and F_{t+1}.

[0043] S4.4 Calculate and output the inferred impact difference set It also visualizes the state transition from KG_t to KG_{t+1} and the inference chain triggered by intervention I, while linking structured source metadata to show the archive source, sub-document semantic anchors and feature fingerprint reliability scores corresponding to each difference fact, forming a full-link visualization of "conclusion-inference chain-data source".

[0044] S5. Based on the feedback of the inference results, structured source metadata and historical cases, dynamically adjust the entity confidence, fact confidence and rule weight parameters in the knowledge graph and the hybrid reasoning engine. In this embodiment, the system performs closed-loop optimization based on the feedback of the deduction results: S51. Adjust entity confidence based on the reliability score of structured traceability metadata. The calculation formula is as follows: ; Among them, Score trace To score the reliability of traceability metadata, As a feedback signal for the conclusion, .

[0045] S52. Update the inference rule weights based on the feedback signal. The calculation formula is as follows: ; in, For learning rate, .

[0046] S53. Link and store the simulation process, initial state, intervention measures, conclusions, feedback signals and source metadata to the case library.

[0047] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.< / conclusion> < / condition> < / conclusion> < / condition> < / conclusion> < / condition> < / conclusion> < / condition>

Claims

1. A method for comprehensive analysis of spatial big data and knowledge graphs for natural resources, characterized in that, Includes the following steps: S1. Acquire and preprocess multi-source data to form a standardized spatiotemporal dataset with structured traceability metadata and structured archive metadata. The multi-source data includes unstructured archives, spatial big data, and socio-economic auxiliary data. S2. Based on the standardized spatiotemporal dataset, extract entities, attributes and relationships in the natural resources field, and construct a knowledge graph that integrates spatiotemporal constraints and multimodal features. The land parcel entities in the knowledge graph contain a geometrically simplified sequence of vector boundary coordinates. S3. Using the knowledge graph as input, perform reasoning calculations through a hybrid reasoning engine that integrates symbolic reasoning, uncertainty reasoning, and spatiotemporal reasoning. The hybrid reasoning engine employs optimized spatial overlap determination logic. S4. Receive user-defined scenario conditions or historical cases as intervention input, perform simulation and deduction based on the current knowledge graph state, and output the deduction conclusion, the impact propagation chain, and the complete data traceability chain; S5. Based on the feedback signal δ of the inference results, the structured traceability metadata and the historical case library, dynamically update the entity confidence and fact confidence in the knowledge graph, and adaptively adjust the rule weight parameters in the hybrid inference engine to form a continuously optimized closed-loop system. The hybrid reasoning engine in S3 generates conclusions based on a weighted fusion of symbolic reasoning results and uncertain reasoning results. The fusion calculation process is as follows: CF final (C) = a - I(CS) + (1 - a) - P(C|E); where CF final (C) denotes the final confidence of conclusion C. I(CS) represents the symbolic reasoning function, which takes the value of 1 when there is at least one rule that satisfies all the preconditions in the current knowledge graph, and 0 otherwise. The preconditions include factual conditions and spatiotemporal constraints. P(C|E) represents the conditional probability of conclusion C calculated based on evidence E using an uncertainty reasoning model; α is the fusion coefficient, used to balance the contributions of the two inference modes, α∈[0,1].

2. The method for comprehensive analysis of spatial big data and knowledge graphs for natural resources according to claim 1, characterized in that, The constructing knowledge graph in S2 includes geometric feature preserving simplification on the parcel entity vector boundary coordinate sequence P' = {p1, p2, …, p n} specifically by using the Douglas-Peucker algorithm for recursive compression: With the first point P of the line segment start and the end point P end construct the reference line segment and calculate the perpendicular distance d i of all intermediate points P i to the line segment determine the maximum distance d max = MAX(d i ) and its corresponding point P index ; If d max is greater than a preset tolerance threshold ε, then P index is marked as a key point, and the above process is recursively performed on line segments [P start , P index ] and [P index , P end ] respectively; Otherwise, discard all intermediate points, keeping only P start With P end ; Merge all key points and start and end points marked during the recursion process, arrange them in the original order, and generate a simplified coordinate point set P' in sequence, preserving the original geometric and topological features.

3. The method for comprehensive analysis of spatial big data and knowledge graphs for natural resources according to claim 2, characterized in that, The construction of the knowledge graph in S2 also includes automatically discovering spatial relationships between entities based on the simplified coordinate point set P': For two land parcel entities A and B, construct polygons using their key coordinate point sets, and calculate their areas Area(A) and Area(B), as well as the area of ​​the intersection region Area(A∩B). Calculate the intersection-union ratio (IoU) of two entities: IoU(A,B) = Area(A∩B) / [Area(A) + Area(B) - Area(A∩B)]; If IoU(A,B)≥θ, where θ is a preset spatial alignment threshold, then a semantic relation edge from entity A to entity B is created in the knowledge graph, with the relation type being "spatial overlap", and the attribute {'IoU':IoU(A,B)} is attached to this edge.

4. The method for comprehensive analysis of spatial big data and knowledge graphs for natural resources according to claim 1, characterized in that, The rules used by the hybrid inference engine in S3 are formalized using spatiotemporally enhanced triples, and their general structure is as follows: IF <condition>THEN <conclusion>WITH weight = W R & ST_Constraint();< / conclusion> < / condition> in, <condition>and <conclusion> A basic fact triple consisting of an entity, an attribute, and a value, or a logical combination thereof.< / conclusion> < / condition> W R is a weight parameter for the rule; ST_Constraint() is a spatiotemporal constraint function, which includes at least one of the following forms of computation: WithinBuffer(e1,e2,r) determines whether the geometric center of entity e1 is located within a buffer with radius r centered at e2. TimeAfter(t1,t2) determines whether time point t1 is after time point t2; The function `Contains(Region,e)` determines whether the geometric boundaries of the administrative division entity `Region` completely contain the geometric boundaries of the land parcel entity `e`. During rule retrieval, the hybrid reasoning engine constructs a multi-dimensional inverted index based on conclusion type, entity type, and spatiotemporal constraint fingerprint, and uses the spatial index generated by S2 to pre-filter candidate entities, accelerating the variable binding process.

5. The method for comprehensive analysis of spatial big data and knowledge graphs for natural resources according to claim 1, characterized in that, The scenario-driven deduction steps in S4 specifically include: S4.1: define the knowledge graph state at the initial moment as KG t , receive a set of intervention operations I input by the user, which includes modification of entity attribute values in KG t , addition or deletion of entities or relationships S4.2: Apply intervention to generate the deduced knowledge graph state KG t+1 =Apply(KG t ,I); S4.3: KG t and KG t+1 The inputs are respectively fed into the hybrid inference engine, inference calculations are performed, and the inference result fact set F is obtained. t and F t+1 ; S4.4: Calculate and output the inferred impact difference set ΔF=F t+1 \F t It also visualizes the state transition path, the impact propagation chain triggered by intervention I, and the complete traceability link from the conclusion to the original data source.

6. The method for comprehensive analysis of spatial big data and knowledge graphs for natural resources according to claim 1, characterized in that, The system closed-loop optimization step in S5 includes adaptive adjustment of rule weights, specifically: For a rule R that is triggered and used to derive a conclusion, its weight parameter W is updated according to the feedback signal δ of that conclusion. R The updated formula is: ； in, η is the learning rate, which controls the step size for weight updates; The feedback signal δ∈[0,1] approaches 1 when the reasoning conclusion is confirmed by the user or is highly consistent with the actual situation; and approaches 0 when the conclusion is corrected or is inconsistent with the actual situation.

7. A system for implementing the comprehensive analysis method of spatial big data and knowledge graph for natural resources as described in any one of claims 1 to 6, characterized in that, include: The data fusion and governance module is used to execute step S1. The knowledge graph construction and management module is used to execute the S2 step and includes a geometric simplification unit and a spatial relationship discovery unit; The multimodal hybrid inference engine module is used to execute the S3 step and realize the confidence fusion calculation and spatiotemporal rule management; The scenario simulation and decision support module is used to execute the S4 step and realize the state difference calculation and visualization simulation; The closed-loop optimization module is used to execute the S5 step and realize the adaptive update of the rule weights.

8. The system according to claim 7, characterized in that, The multimodal hybrid inference engine module includes a rule editor for users to define and configure business rules. These business rules include a spatiotemporal constraint function ST_Constraint() and a weight parameter W. R The structure.

9. The system according to claim 7, characterized in that, The scenario simulation and decision support module includes a case library management unit, used to store each simulation process and initial state KG. t The intervention (I), conclusion (ΔF), and final feedback (δ) are stored as a structured case in a knowledge graph, forming a searchable and reusable historical experience database.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements method S3 as described in any one of claims 1 to 6: using the constructed knowledge graph, reasoning computation is performed through a hybrid reasoning engine that integrates symbolic reasoning, uncertainty reasoning, and spatiotemporal reasoning.

Images