Knowledge Graph Reasoning Method and System Based on Spatiotemporal Relationships

Abstract

This invention provides a knowledge graph reasoning method and parsing system based on spatiotemporal relationships, belonging to the field of knowledge graph technology, and solves the problem that existing technologies cannot support complex spatiotemporal knowledge reasoning. The method includes the following steps: uniformly dividing the target geographic space into an M×N×L grid matrix; for each unit grid in the grid matrix, establishing a set including its spatiotemporal information and entity information as the knowledge graph data model of that unit grid; receiving the input spatiotemporal relationship problem, determining the judgment rules related to the problem, and then establishing a judgment model for verifying the correctness of rule matching for the spatiotemporal information and entity information of each unit grid; verifying the knowledge graph data model of each unit grid according to the above judgment model, obtaining all unit grids that conform to the above judgment rules, and outputting the region formed by all obtained unit grids or the entity objects within the region as the reasoning result.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph technology, and in particular to a knowledge graph reasoning method and system based on spatiotemporal relationships. Background Technology

[0002] Spatiotemporal knowledge graphs are extensions of knowledge graphs in time and space. They are structured spatiotemporal information knowledge bases that describe concepts, entities, attributes, and their interrelationships within the geographic spatiotemporal information domain, forming a network-like knowledge structure. The core of spatiotemporal knowledge graphs is the effective organization of geographic spatiotemporal information and knowledge, forming a spatiotemporal knowledge network. Then, through spatiotemporal computation models and natural semantic models, they enable the storage, retrieval, reasoning, computation, and management of spatiotemporal information and knowledge, thereby achieving functions such as semantic search, spatiotemporal computation, spatiotemporal knowledge recommendation, and association analysis.

[0003] Most existing spatiotemporal knowledge graphs adopt entity-relationship-based data representation, that is, as semantic knowledge graphs, they have stable storage management and reasoning application capabilities for entity and concept nodes and logical relationship knowledge, but they do not support the representation of geographic spatiotemporal nodes (including spatiotemporal entities and time-space data), and do not have efficient spatiotemporal storage, retrieval, reasoning, computation and management capabilities.

[0004] Existing semantic network search can only perform simple spatiotemporal relationship reasoning, and its spatiotemporal computation accuracy is far lower than that required for spatiotemporal question answering. Due to its weak spatiotemporal computation capabilities, semantic networks are designed for spatiotemporal scenarios and cannot support complex business needs such as spatiotemporal knowledge reasoning. Summary of the Invention

[0005] Based on the above analysis, the embodiments of the present invention aim to provide a knowledge graph reasoning method and system based on spatiotemporal relationships, so as to solve the problem that the existing technology is difficult to support complex spatiotemporal knowledge reasoning.

[0006] On one hand, embodiments of the present invention provide a knowledge graph reasoning method based on spatiotemporal relationships, comprising the following steps:

[0007] Determine the target geographic space and uniformly divide the target geographic space into a three-dimensional grid matrix of M×N×L;

[0008] For each cell in the 3D mesh matrix, a set of spatiotemporal information and entity information is established as a knowledge graph data model for that cell; the entity information includes entity objects and their number.

[0009] The system receives an input spatiotemporal relationship problem, determines the judgment rules related to the problem, and then establishes a judgment model based on the judgment rules to verify the correctness of rule matching for the spatiotemporal information and entity information of each unit grid.

[0010] The knowledge graph data model of each unit grid is verified based on the above judgment model. All unit grids in the target geospace whose spatiotemporal information and entity information conform to the above judgment rules are obtained. The region formed by all the obtained unit grids or the entity objects within the region are output as the reasoning result.

[0011] The beneficial effects of the above technical solution are as follows: Traditional semantic networks can only perform simple spatiotemporal relationship reasoning, and their spatiotemporal computation accuracy is far lower than that required for spatiotemporal question answering. Due to their weak spatiotemporal computation capabilities, semantic networks are unable to support complex spatiotemporal reasoning (spatiotemporal knowledge representation, spatiotemporal judgment, etc.) business needs in spatiotemporal scenarios. The above technical solution utilizes grid algebra computation to achieve knowledge judgment for spatiotemporal grid knowledge graphs. Specifically, based on the received spatiotemporal relationship question, judgment rules are determined, and each unit grid cell in the target geographic space is judged using a (spatiotemporal) knowledge graph data model, resulting in a spatiotemporal domain graph of the judgment result (i.e., the region composed of all unit grid cells that conform to the above judgment rules at a certain moment).

[0012] Based on a further improvement to the above method, after the step of uniformly dividing the target geographic space into a three-dimensional grid matrix of M×N×L, and before the step of establishing a set including its spatiotemporal information and entity information for each grid cell in the three-dimensional grid matrix, the method further includes the following steps:

[0013] Grid coding is performed on the spatiotemporal and entity information in the target geospace to obtain grid coding results that characterize the spatial location information of entity objects and the subdivision level to which the entity objects belong.

[0014] Furthermore, for each unit grid cell in the three-dimensional grid matrix, a set including its spatiotemporal information and entity information is established based on the grid encoding results, serving as a knowledge graph data model for that unit grid cell; and...

[0015] Based on the aforementioned judgment rules, a judgment model is established according to the grid encoding results to verify the correctness of rule matching for the spatiotemporal information and entity information of each grid cell.

[0016] Furthermore, spatiotemporal information includes specific time and spatial coordinates; entity information also includes the entity's attributes, events, and the influencing factors of each event; and,

[0017] Entity objects include at least one of pedestrians, ships, and vehicles;

[0018] The attributes of an entity object further include the remaining lifetime of the entity object, whether it is passable, and the degree of association between entity objects;

[0019] The inference results include restricted areas and drivable areas;

[0020] The method also includes the step of integrating the spatiotemporal information and entity information of grids that meet the decision rules into a grid spatiotemporal knowledge graph through the triple description framework of the grid spatiotemporal knowledge graph.

[0021] Furthermore, for entities that are ships or vehicles, the determination model includes:

[0022]

[0023] in,

[0024]

[0025] In the formula, k is the propagation constant under the wind field diffusion scenario, and C t Let C be the three-dimensional coordinates of any element mesh at time t. ap t Let Attr(a) be the location of the p-th wind field center at time t, where p = 1, ..., n, and n is the total number of wind field centers. p t ) represents the wind field intensity at the p-th wind field center at time t, Distance() represents the distance, and value(C t ) represents the three-dimensional coordinates C of the element mesh. t Wind power at the location, windpower(value(C) t )) represents the three-dimensional coordinates C of the element mesh. t The wind level at the location, P is the set value, result(C) t The result is 1, indicating that the navigation is suitable and 0 indicates that the navigation is not suitable.

[0026] Furthermore, after uniformly dividing the target geographic space into an M×N×L three-dimensional grid matrix, each grid cell in the three-dimensional grid matrix is ​​assigned an independent ID value for positioning; and,

[0027] The ID value of the grid located in row i and column j is:

[0028] ID(i,j)=N(i-1)+j,

[0029] In the formula, i = 1, ..., M, j = 1, ..., N.

[0030] Furthermore, the step of establishing a set of spatiotemporal information and entity information for each unit grid cell in the three-dimensional grid matrix, as a knowledge graph data model for that unit grid cell, further includes:

[0031] Collect and organize spatiotemporal data;

[0032] Semantic analysis and knowledge extraction are performed on the spatiotemporal data to obtain all data in the spatiotemporal data that are related to the target geographic space and entity objects, and to generate individual knowledge sets of entity objects;

[0033] Based on the above individual knowledge set, determine the number of entity objects, the trajectory of each entity object, the attributes of the entity objects, the events, and the influencing factors of each event;

[0034] Based on the trajectory of each entity object, the spatiotemporal related information of each unit grid in the three-dimensional grid matrix and each entity object is further determined, thereby determining the spatiotemporal information and entity information of each unit grid. The set of spatiotemporal information and entity information of the unit grid is used as the knowledge graph data model of the unit grid.

[0035] Furthermore, the step of performing semantic analysis and knowledge extraction on the spatiotemporal data to obtain all data in the spatiotemporal data that are related to the target geographic space and entity objects further includes:

[0036] To identify entity objects from plain text data and obtain spatial information related to the entity objects, information on the degree of association between entity objects, and historical event information;

[0037] The remote sensing imagery is used to identify the target geospatial space, and to obtain the spatial information of road conditions and objects contained in the target geospatial space. At the same time, the attribute information of road conditions and objects is added, including road width and object size.

[0038] Perform image recognition on video or image data to detect whether any events of interest have occurred. If so, generate corresponding event semantics and then determine the influencing factors of the event based on the generated event template.

[0039] Based on all the information obtained above and the influencing factors of the event, an individual knowledge set of the entity object is generated and displayed in the form of a knowledge graph.

[0040] Furthermore, it also includes the following steps to complete the function of improving the knowledge graph data model of each existing grid cell based on the newly acquired spatiotemporal data:

[0041] Receive newly acquired spatiotemporal data at set intervals;

[0042] The knowledge graph data model of each existing grid cell is fused with the newly acquired spatiotemporal data to update the knowledge graph data model of each grid cell.

[0043] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0044] 1. A knowledge determination method based on a grid-based spatiotemporal knowledge graph (a knowledge graph data model for each grid cell) is adopted. This method can, for data, phenomena, and events (problems) with spatiotemporal attributes, obtain the required spatiotemporal conclusions (the region comprised of all grid cells conforming to the determination rules, or the entity objects within that region) based on specific knowledge definitions (determination models, logical reasoning using specified determination rules, and algebraic calculations of spatiotemporal grid encoding) using specific knowledge definitions. The spatiotemporal conclusions can be presented as a spatiotemporal representation graph.

[0045] 2. Through the judgment calculation of the judgment model, the relevant nodes or relationships are judged according to the input logic. The judgment result is "TRUE" or "FALSE" for each entity object or related spatiotemporal grid according to the input logic, or a specific count value. The information can be integrated based on the calculation results to obtain conclusions, spatiotemporal grid diagrams or other visualization results.

[0046] 3. In spatiotemporal question-answering scenarios, the judgment result is a spatiotemporal domain that meets the judgment conditions, which can be presented in the form of a spatiotemporal domain graph. Spatiotemporal data and entity data can be organized using the data organization model of a grid spatiotemporal knowledge graph to form a grid spatiotemporal knowledge graph. Different judgment rules can be managed by a rule base. Rules in the rule base can be applied to the grid spatiotemporal knowledge graph through knowledge judgment to obtain a spatiotemporal domain representation grid graph of the judgment result.

[0047] On the other hand, embodiments of the present invention provide a knowledge graph reasoning system based on spatiotemporal relationships, including:

[0048] The knowledge graph data model construction module is used to determine the target geospace and uniformly divide the target geospace into a three-dimensional grid matrix of M×N×L. For each grid cell in the three-dimensional grid matrix, a set of spatiotemporal information and entity information is established as the knowledge graph data model of that grid cell. The entity information includes entity objects and their number.

[0049] The data reasoning module receives input spatiotemporal relationship questions, identifies judgment rules related to the questions, and then establishes a judgment model based on the judgment rules to verify the correctness of conditional matching of spatiotemporal information and entity information of each unit grid. The module then verifies the knowledge graph data model of each unit grid according to the judgment model, and obtains the region composed of all unit grids in the target geospace whose spatiotemporal information and entity information conform to the above judgment rules, or the entity objects within the region, as the reasoning result output.

[0050] The summary section is provided to present the chosen concepts in a simplified form, which will be further described in the detailed description below. The summary section is not intended to identify essential or essential features of the invention, nor is it intended to limit the scope of the invention. Attached Figure Description

[0051] The above and other objects, features and advantages of the present invention will become more apparent from the more detailed description of exemplary embodiments of the invention in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same parts.

[0052] Figure 1 A schematic diagram of the steps of the knowledge graph reasoning method in Example 1 is shown;

[0053] Figure 2 A schematic diagram illustrating the principle of the knowledge graph reasoning method in Example 2 is shown.

[0054] Figure 3 An example of the spatiotemporal knowledge graph knowledge determination process for any grid in Example 2 is shown. Detailed Implementation

[0055] Embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0056] The term "comprising" and its variations as used herein signify open inclusion, i.e., "including but not limited to". Unless otherwise stated, the term "or" means "and / or". The term "based on" means "at least partially based on". The terms "one example embodiment" and "one embodiment" mean "at least one example embodiment". The term "another embodiment" means "at least one additional embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0057] Example 1

[0058] One embodiment of the present invention discloses a knowledge graph reasoning method based on spatiotemporal relationships, such as... Figure 1 As shown, it includes the following steps:

[0059] S1. Determine the target geographic space and uniformly divide the target geographic space into a three-dimensional grid matrix of M×N×L;

[0060] S2. For each cell in the 3D mesh matrix, establish a set including its spatiotemporal information and entity information as the knowledge graph data model of that cell (referred to as the mesh knowledge graph data model); entity information includes entity objects and their number;

[0061] S3. Receive the spatiotemporal relationship problem input, determine the judgment rules (system built-in rule base) related to the problem, and then establish a judgment model based on the judgment rules to verify the correctness of rule matching for the spatiotemporal information and entity information of each unit grid.

[0062] S4. Verify the knowledge graph data model of each unit grid according to the above judgment model, obtain all unit grids in the target geospace whose spatiotemporal information and entity information conform to the above judgment rules, and output the region formed by all obtained unit grids or the entity objects within the region as the reasoning result.

[0063] To illustrate the application scenarios of the above reasoning method, examples are provided below.

[0064] Example 1: When receiving a spatiotemporal question such as "In a certain sea area, at a certain time, which ships will be in danger of wind and need to leave as soon as possible", the system will call the known rule related to the question, "In the scenario of ocean navigation, when the wind force is above level 8, the XX type of ship should not sail". The question can be answered by the spatiotemporal knowledge graph (the knowledge graph data model of each unit grid).

[0065] Example 2 illustrates how, upon receiving a spatiotemporal question such as "At a certain time, what are the no-navigation zones for a certain vessel in a given sea area?", the system invokes relevant known rules: "Due to vessel B's condition limitations, vessel B cannot enter sea areas with operational meteorological restrictions of Level II or higher. Furthermore, it is known that operational meteorological restriction Level II refers to a restricted navigation area with a significant wave height not exceeding 2.0m and a wind force not exceeding force 6 (Beaufort scale)." Through the knowledge graph data model of each grid cell (meteorological monitoring data such as wind force and significant wave height within the navigation area, included in the entity information), the system can determine in real time the areas satisfying the aforementioned known rules, thus obtaining the no-navigation zones.

[0066] In the above case, a spatiotemporal knowledge graph of the grid was constructed using meaningful wave height maps and Beaufort scale wind power maps collected at fixed time intervals within the study area. The criterion is that sea areas with meaningful wave heights not exceeding 2.0m and wind speeds not exceeding force 6 are navigable. The calculation output is the region comprised of all navigable cell grids for ship B at time t.

[0067] The reasoning method of spatiotemporal knowledge graphs is a way to obtain spatiotemporal conclusions (spatiotemporal domain graphs) based on specific knowledge definitions for data, phenomena, and events with spatiotemporal attributes. Spatiotemporal conclusions are often presented in the form of spatiotemporal representation graphs.

[0068] Compared to existing technologies, traditional semantic networks can only perform simple spatiotemporal relationship reasoning, and their spatiotemporal computation accuracy is far lower than that required for spatiotemporal question answering. Due to their weak spatiotemporal computation capabilities, semantic networks struggle to support complex spatiotemporal reasoning (spatiotemporal knowledge representation, spatiotemporal judgment, etc.) business requirements in spatiotemporal scenarios. The technical solution provided in this embodiment utilizes grid algebra computation to achieve knowledge judgment for spatiotemporal grid knowledge graphs. Specifically, based on the received spatiotemporal relationship question, judgment rules are determined, and each grid cell in the target geographic space is judged using a (spatiotemporal) knowledge graph data model, resulting in a spatiotemporal domain graph of the judgment result (i.e., the region composed of all grid cells that conform to the above judgment rules at a certain moment).

[0069] Example 2

[0070] Based on Example 1, the method is improved by including the following steps after step S1 and before step S2:

[0071] S20. Perform grid coding on the spatiotemporal information and entity information in the target geospace to obtain grid coding results used to characterize the spatial location information of entity objects and the subdivision level to which the entity objects belong.

[0072] Specifically, in step S20, the grid coding method can be generated by a partitioning coding method, such as the GeoSOT grid partitioning scheme or the Beidou grid code scheme.

[0073] For example, the GeoSOT grid partitioning scheme can be adopted. The core idea of ​​GeoSOT grid partitioning is based on the principle of Earth partitioning. By dividing the Earth's surface into dimensions, it explores and constructs a dedicated grid suitable for spatial information or data organization. On the one hand, this grid can have good scale aggregation and location correlation with existing major spatial information organization grids on Earth; on the other hand, a location identification system more suitable for spatial information organization can be built on this grid, providing a reference basis for the geographic grid identification and consistent indexing of global spatial information locations, thereby solving the problem of unified location organization and regional correlation scheduling of spatial information. Furthermore, it can quickly determine the directional relationship and distance between two grids and is easy to calculate. In the embodiment, the current geographic location information of a spatial object can be processed to determine the grid code where the spatial object is currently located, and the grid code represents the current geographic location of the spatial object.

[0074] For example, the BeiDou grid code scheme can also be adopted. BeiDou grid code technology is a multi-scale, discrete global geographic grid location coding technology suitable for navigation and positioning services, developed based on GeoSOT (Geographic Space Occlusion Theory). It is a new extended location code output by the BeiDou latitude and longitude location code and an important assignment system for a new geospatial location framework. The BeiDou grid code technology system, on the one hand, provides precise coding and expression of information across the entire global space, assigning a calculable and easily searchable globally unique spatial identifier to every inch of space, including land, sea, air, space, electricity, and underground; on the other hand, it is also a big data organization framework based on space, performing three-dimensional grid data modeling, and combining time coding to assign spatiotemporal data attributes to everything in geospatial space, realizing the interconnection of all data based on spatiotemporal coding as the primary index. The BeiDou grid is formulated according to the GeoSOT global grid. This embodiment can utilize the GeoSOT coding rules to divide multiple grid levels, with each grid level having the same grid size. The specific level of the coding is not limited and can include M layers, each layer including N levels. As for the grid size, it can be determined when determining the grid code to define the minimum positioning range. For example, if the preset coding level is 26 levels and the preset coding range is 1m*1m, it can be determined that the coding needs to be applied to a specific 1 square meter coding range.

[0075] Preferably, in step S2, for each unit grid in the three-dimensional grid matrix, a set including its spatiotemporal information and entity information is established based on the grid encoding result, which serves as the knowledge graph data model of that unit grid.

[0076] Preferably, in step S3, based on the judgment rules, a judgment model is established according to the grid coding results to verify the correctness of rule matching for the spatiotemporal information and entity information of each grid cell.

[0077] Preferably, the aforementioned spatiotemporal information includes specific time and spatial location coordinates. The spatiotemporal information is multi-source heterogeneous spatiotemporal data, including vector data, imagery, etc.

[0078] Entity information includes entity objects and their number, entity object attributes, events, and the influencing factors of each event.

[0079] The entity object further includes at least one of pedestrians, ships, and vehicles.

[0080] The attributes of an entity object further include its remaining lifetime, whether it is passable, and the degree of association between entity objects.

[0081] For example, a "cargo ship" is an entity with the attribute "remaining lifespan of 25-30 years", and it has different sub-entities such as "dry bulk carrier", "liquid cargo carrier", "crude oil carrier", "general cargo ship" and "container ship".

[0082] Preferably, spatiotemporal information and entity information can be integrated into a grid spatiotemporal knowledge graph using a triplet description framework, such as... Figure 2 As shown, a triple consists of three parts: (Subject, Predicate, Object). For example, Zhejiang University being located in Hangzhou can be simply represented by a triple. A triple represents a statement of a logical fact in the objective world. These triples, connected end to end, form a graph describing the relationships between everything.

[0083] The rule base is a collection of various spatiotemporal or logical rules. For example, "The meteorological restriction for the operation of ground effect vehicles I is that the meaningful wave height in the navigation area does not exceed 3.0m and the wind force does not exceed level 7" is a rule.

[0084] Preferably, the reasoning results include restricted areas and drivable areas, which can be represented by a spatiotemporal domain diagram.

[0085] Preferably, for the entity object being a ship or vehicle, the determination model includes:

[0086]

[0087] in,

[0088]

[0089] In the formula, k is the propagation constant under the wind field diffusion scenario, and C t Let C be the three-dimensional coordinates of any element mesh at time t. ap t Let Attr(a) be the location of the p-th wind field center at time t, where p = 1, ..., n, and n is the total number of wind field centers. p t ) represents the wind field intensity at the p-th wind field center at time t, Distance() represents the distance, and value(C t ) represents the three-dimensional coordinates C of the element mesh. t Wind power at the location, windpower(value(C) t )) represents the three-dimensional coordinates C of the element mesh. t The wind level at the location, P is the set value, result(C) t The result is 1, indicating that the navigation is suitable and 0 indicates that the navigation is not suitable.

[0090] The derivation process is as follows: the target geographic space (interest space domain) is C, the time domain is T, time t∈T, and the grid set is {C1}. t ,…,C MN t}, where MN is the total number of element mesh cells, and for each element mesh cell, the three-dimensional coordinates C t The coordinates C of the unit mesh can be calculated. t Wind force at the location:

[0091]

[0092] C t The value may be affected by multiple factors. i t ∈{a1 t ,…,a n t The influence of}. If we assume that the influencing factors are independent, then we have,

[0093]

[0094] At time t, when influencing factor a i t When the interactions between them are independent, a i t For the three-dimensional coordinates C of the element mesh t The effects at point a are cumulative, and when a i t When related to space, a i t For grid C t The influence and a i t and C t spatiotemporal location, a i t The attribute (Attr(a) i t )) and others Related factors (θ(a) i t ))related.

[0095] For example, given the two wind field centers a1 at time t. t and a2 t The spatial location of wind field center 1 at time t is C. a1 t The wind field intensity is Attr(a1) t The same applies to wind field center 2. The two wind field centers are related to grid C. t The wind force also causes a cumulative effect, and the degree of the effect is inversely weighted by distance. Therefore, grid C... t The wind force is:

[0096]

[0097] The judgment condition "When the wind force is above level 8, XX type of vessel is not suitable for navigation" applies to grid C. t The function to determine whether ship number XX can sail is:

[0098]

[0099] For the decision grid diagram studied in the time domain and the spatial domain, it can be represented as:

[0100]

[0101]

[0102] In the formula, τ is an indicator function, and ° is an operator that identifies whether β is within a set range. If it is within the set range, the value of τ is 1, otherwise it is 0.

[0103] Preferably, after uniformly dividing the target geographic space into an M×N grid matrix, each cell in the grid matrix is ​​assigned an independent ID value for location purposes. Furthermore, the ID value of the cell located in the i-th row and j-th column is:

[0104] ID(i,j)=N(i-1)+j,

[0105] In the formula, i = 1, ..., M, j = 1, ..., N.

[0106] Preferably, step S2 further includes:

[0107] S21. Collect and organize spatiotemporal data; the spatiotemporal data includes at least one of the following: plain text data, remote sensing images, video or image data, Terrain datasets, web page data, map data, POI data, and sensor data.

[0108] S22. Perform semantic analysis and knowledge extraction on the spatiotemporal data to obtain all data in the spatiotemporal data that are related to the target geographic space and entity objects, and generate an individual knowledge set of the entity objects;

[0109] S23. Based on the above individual knowledge set, determine the number of entity objects, the trajectory of each entity object, the attributes of the entity objects, the events, and the influencing factors of each event;

[0110] S24. Based on the running trajectory of each entity object, further determine the spatiotemporal related information of each unit grid body in the three-dimensional grid matrix and each entity object, and then determine the spatiotemporal information and entity information of each unit grid body. The set of spatiotemporal information and entity information of the unit grid body is used as the knowledge graph data model of the unit grid body.

[0111] Preferably, step S22, which involves performing semantic analysis and knowledge extraction on the spatiotemporal data to obtain all data in the spatiotemporal data that are related to the target geographic space and entity objects, further includes:

[0112] S221. Recognize entity objects in plain text data and obtain spatial information related to entity objects, information on the degree of association between entity objects, and historical event information;

[0113] S222. Identify the target geospatial space in the remote sensing image, obtain the spatial information of road conditions and objects contained in the target geospatial space, and add attribute information of road conditions and objects, wherein the attribute information is road width and object size;

[0114] S223. Perform image recognition on video or image data to detect whether any events of interest have occurred. If so, generate corresponding event semantics and then determine the influencing factors of the event based on the generated event template.

[0115] S224. Based on all the information obtained above and the influencing factors of the event, generate an individual knowledge set of the entity object and display it in the form of a knowledge graph.

[0116] Preferably, the method further includes the following steps to complete the function of improving the knowledge graph data model of each original grid cell based on the newly acquired spatiotemporal data:

[0117] S5. Receive newly acquired spatiotemporal data at set intervals;

[0118] S6. Integrate the existing knowledge graph data model of each grid cell with the newly acquired spatiotemporal data to update the knowledge graph data model of each grid cell.

[0119] When implementing, such as Figure 3 As shown, node B points to node C. a t1 and C b t2 Let represent the grid where ship B is located at time t1 and t2. Through implicit relations, we know that C... p t3 and C q t3 If ship B has two equivalent passable grids at the next time point t3, then according to the judgment rule (passable in sea areas with a meaningful wave height not exceeding 2.0m and a wind force not exceeding level 6), then C... q t3 The wind force was level 5, with a wave height of 0.5m, while C p t3 The wind force is level 6, with a significant wave height of 2.5m. The calculation will return: C.q t3 It is a passable grid, and C p t3 It is an impassable grid.

[0120] Compared with existing technologies, the knowledge graph reasoning method based on spatiotemporal relationships provided in this embodiment has the following beneficial effects:

[0121] 1. A knowledge determination method based on a grid-based spatiotemporal knowledge graph (a knowledge graph data model for each grid cell) is adopted. This method can, for data, phenomena, and events (problems) with spatiotemporal attributes, obtain the required spatiotemporal conclusions (the region comprised of all grid cells conforming to the determination rules, or the entity objects within that region) based on specific knowledge definitions (determination models, logical reasoning using specified determination rules, and algebraic calculations of spatiotemporal grid encoding) using specific knowledge definitions. The spatiotemporal conclusions can be presented as a spatiotemporal representation graph.

[0122] 2. Through the judgment calculation of the judgment model, the relevant nodes or relationships are judged according to the input logic. The judgment result is "TRUE" or "FALSE" for each entity object or related spatiotemporal grid according to the input logic, or a specific count value. The information can be integrated based on the calculation results to obtain conclusions, spatiotemporal grid diagrams or other visualization results.

[0123] 3. In spatiotemporal question-answering scenarios, the judgment result is a spatiotemporal domain that meets the judgment conditions, which can be presented in the form of a spatiotemporal domain graph. Spatiotemporal data and entity data can be organized using the data organization model of a grid spatiotemporal knowledge graph to form a grid spatiotemporal knowledge graph. Different judgment rules can be managed by a rule base. Rules in the rule base can be applied to the grid spatiotemporal knowledge graph through knowledge judgment to obtain a spatiotemporal domain representation grid graph of the judgment result.

[0124] Example 3

[0125] The present invention also discloses a knowledge graph reasoning system based on spatiotemporal relationships. The program corresponding to the knowledge graph reasoning method based on spatiotemporal relationships described in embodiment 1 or 2 includes a knowledge graph data model construction module and a data reasoning module connected in sequence.

[0126] The knowledge graph data model construction module is used to determine the target geospace and uniformly divide the target geospace into an M×N grid matrix; and for each unit grid in the three-dimensional grid matrix, to establish a set including its spatiotemporal information and entity information as the knowledge graph data model of that unit grid; wherein, the entity information includes entity objects and their number.

[0127] The data reasoning module is used to receive input spatiotemporal relationship questions, identify judgment rules related to the questions, and then establish a judgment model based on the judgment rules to verify the correctness of conditional matching of spatiotemporal information and entity information of each unit grid; and to verify the knowledge graph data model of each unit grid according to the judgment model, and obtain the region composed of all unit grids in the target geospace whose spatiotemporal information and entity information conform to the above judgment rules or the entity objects in the region, as the reasoning result output.

[0128] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical applications, or improvements to the prior art of the embodiments, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A knowledge graph reasoning method based on spatiotemporal relationships, characterized in that, Including the following steps: Determine the target geographic space and uniformly divide the target geographic space into a three-dimensional grid matrix of M×N×L; For each cell in the 3D mesh matrix, a set of spatiotemporal information and entity information is established as a knowledge graph data model for that cell; the entity information includes entity objects and their number; the entity objects include at least one of pedestrians, ships, and vehicles. The system receives an input spatiotemporal relationship problem, determines the judgment rules related to the problem, and then establishes a judgment model based on the judgment rules to verify the correctness of rule matching for the spatiotemporal information and entity information of each unit grid. For entities that are ships or vehicles, the determination model includes: ， in, In the formula, k is the propagation constant under the wind field diffusion scenario, and C t Let C be the three-dimensional coordinates of any element mesh at time t. ap t Let Attr(a) be the location of the p-th wind field center at time t, where p = 1, ..., n, and n is the total number of wind field centers. p t Let ) represent the wind field intensity at the p-th wind field center at time t, Distance() represent the distance, and value(C) represent the wind field intensity at the p-th wind field center. t ) represents the three-dimensional coordinates C of the element mesh. t Wind power at the location, windpower(value(C) t )) represents the three-dimensional coordinates C of the element mesh. t The wind level at the location, P is the set value, result(C) t The result is 1, indicating that the sea is suitable for navigation, and 0, indicating that the sea is not suitable for navigation. The knowledge graph data model of each unit grid is verified based on the above judgment model. All unit grids in the target geospace whose spatiotemporal information and entity information conform to the above judgment rules are obtained. The region formed by all the obtained unit grids or the entity objects within the region are output as the reasoning result.

2. The knowledge graph reasoning method based on spatiotemporal relationships according to claim 1, characterized in that, After the step of uniformly dividing the target geographic space into a three-dimensional grid matrix of M×N×L, and before the step of establishing a set including its spatiotemporal information and entity information for each cell in the three-dimensional grid matrix, the method further includes the following steps: Grid coding is performed on the spatiotemporal and entity information in the target geospace to obtain grid coding results that characterize the spatial location information of entity objects and the subdivision level to which the entity objects belong.

3. The knowledge graph reasoning method based on spatiotemporal relationships according to claim 2, characterized in that, For each cell in the three-dimensional mesh matrix, a set including its spatiotemporal information and entity information is established based on the mesh encoding results, serving as a knowledge graph data model for that cell; and... Based on the aforementioned judgment rules, a judgment model is established according to the grid encoding results to verify the correctness of rule matching for the spatiotemporal information and entity information of each grid cell.

4. The knowledge graph reasoning method based on spatiotemporal relationships according to any one of claims 1-3, characterized in that, Spatiotemporal information includes specific time and spatial coordinates; entity information also includes the entity's attributes, events, and the influencing factors of each event; and, The attributes of an entity object further include the remaining lifetime of the entity object, whether it is passable, and the degree of association between entity objects; The inference results include restricted areas and drivable areas; The method also includes the step of integrating the spatiotemporal information and entity information of grids that meet the decision rules into a grid spatiotemporal knowledge graph through the triple description framework of the grid spatiotemporal knowledge graph.

5. The knowledge graph reasoning method based on spatiotemporal relationships according to claim 1, characterized in that, After uniformly dividing the target geographic space into an M×N×L three-dimensional grid matrix, each cell in the three-dimensional grid matrix is ​​assigned an independent ID value for positioning; and, The ID value of the grid located in row i and column j is: ID(i,j)=N(i-1)+j, In the formula, i=1,......,M, j=1,......,N.

6. The knowledge graph reasoning method based on spatiotemporal relationships according to any one of claims 1-3, characterized in that, The step of establishing a set of spatiotemporal information and entity information for each unit grid cell in the three-dimensional grid matrix as a knowledge graph data model for that unit grid cell further includes: Collect and organize spatiotemporal data; Semantic analysis and knowledge extraction are performed on the spatiotemporal data to obtain all data in the spatiotemporal data that are related to the target geographic space and entity objects, and to generate individual knowledge sets of entity objects; Based on the above individual knowledge set, determine the number of entity objects, the trajectory of each entity object, the attributes of the entity objects, the events, and the influencing factors of each event; Based on the trajectory of each entity object, the spatiotemporal related information of each unit grid in the three-dimensional grid matrix and each entity object is further determined, thereby determining the spatiotemporal information and entity information of each unit grid. The set of spatiotemporal information and entity information of the unit grid is used as the knowledge graph data model of the unit grid.

7. The knowledge graph reasoning method based on spatiotemporal relationships according to claim 6, characterized in that, The step of performing semantic analysis and knowledge extraction on the spatiotemporal data to obtain all data in the spatiotemporal data that are related to the target geographic space and entity objects further includes: To identify entity objects from plain text data and obtain spatial information related to the entity objects, information on the degree of association between entity objects, and historical event information; The remote sensing imagery is used to identify the target geospatial space, and to obtain the spatial information of road conditions and objects contained in the target geospatial space. At the same time, the attribute information of road conditions and objects is added, including road width and object size. Perform image recognition on video or image data to detect whether any events of interest have occurred. If so, generate corresponding event semantics and then determine the influencing factors of the event based on the generated event template. Based on all the information obtained above and the influencing factors of the event, an individual knowledge set of the entity object is generated and displayed in the form of a knowledge graph.

8. The knowledge graph reasoning method based on spatiotemporal relationships according to any one of claims 1, 2, 3, 5, and 7, characterized in that, It also includes the following steps to complete the function of improving the knowledge graph data model of each existing grid cell based on the newly acquired spatiotemporal data: Receive newly acquired spatiotemporal data at set intervals; The knowledge graph data model of each existing grid cell is fused with the newly acquired spatiotemporal data to update the knowledge graph data model of each grid cell.

9. A knowledge graph reasoning system based on spatiotemporal relationships, characterized in that, include: The knowledge graph data model construction module is used to determine the target geographic space and uniformly divide the target geographic space into a three-dimensional grid matrix of M×N×L. For each cell in the 3D mesh matrix, a set of spatiotemporal information and entity information is established as a knowledge graph data model for that cell; the entity information includes entity objects and their number; the entity objects include at least one of pedestrians, ships, and vehicles. The data reasoning module is used to receive input spatiotemporal relationship problems, identify judgment rules related to the problems, and then establish a judgment model based on the judgment rules to verify the correctness of condition matching of spatiotemporal information and entity information of each unit grid. For entities that are ships or vehicles, the determination model includes: ， in, In the formula, k is the propagation constant under the wind field diffusion scenario, and C t Let C be the three-dimensional coordinates of any element mesh at time t. ap t Let Attr(a) be the location of the p-th wind field center at time t, where p = 1, ..., n, and n is the total number of wind field centers. p t Let ) represent the wind field intensity at the p-th wind field center at time t, Distance() represent the distance, and value(C) represent the wind field intensity at the p-th wind field center. t ) represents the three-dimensional coordinates C of the element mesh. t Wind power at the location, windpower(value(C) t )) represents the three-dimensional coordinates C of the element mesh. t The wind level at the location, P is the set value, result(C) t The result is 1, indicating that the sea is suitable for navigation, and 0, indicating that the sea is not suitable for navigation. The knowledge graph data model of each unit grid is verified according to the judgment model to obtain the region or entity objects within the target geospace whose spatiotemporal information and entity information conform to the above judgment rules, and the reasoning result is output.

Images