Method for constructing three-dimensional semantic knowledge graph of geospatial entity spatio-temporal relationship

Abstract

This invention discloses a method for constructing a three-dimensional semantic knowledge graph of spatiotemporal relationships of geographic entities, belonging to the field of knowledge graph construction technology. The method includes: acquiring a unified benchmark geographic entity flow; selecting a set of mutually overlapping geographic entity pairs based on XY-axis projection overlap detection and reading sensor accuracy settings; determining the vertical orientation relationship based on the bounding box vertical extremum; querying a preset layer thickness mapping table to determine the coarse-grained sectioning layer thickness; calculating the coarse-grained vertical coverage and traversal depth interval in the Z-axis direction and defining them as coarse-grained vertical indices; comparing these with a first threshold library to generate preliminary vertical relationship status labels; for ambiguous or enveloped entity pairs, determining the local depth range based on the coarse-grained vertical indices; resampling with a fixed fine-grained sectioning layer thickness; revising the threshold to generate a second threshold library for re-judgment; and forming a revised graph version. This invention can eliminate vertical semantic ambiguity and improve the accuracy of determining the vertical relationship of geographic entities.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph construction technology, and more specifically, to a method for constructing a three-dimensional semantic knowledge graph of spatiotemporal relationships of geographic entities. Background Technology

[0002] With the advancement of smart city construction, urban management departments are gradually integrating multi-source spatiotemporal data, such as 3D point clouds of the ground, underground facility detection profiles, and building information, and are attempting to unify these data using knowledge graphs. However, existing methods mostly follow the topological approach of 2D GIS or are geared towards single-scenario BIM pipeline systems.

[0003] In pursuit of real-time performance, synchronous lightweight processing often cuts above-ground and underground entities into sections with a fixed layer thickness in the same process and directly applies 2D overlap judgment rules. This strategy ignores the differences in sensor accuracy and elevation coordinate reference, resulting in distortion of the vertical extrema of the bounding box.

[0004] Vertical semantic ambiguity frequently occurs in complex working conditions such as pipelines under bridges and building pile foundations. The vertical positional relationship is often misjudged as vertical envelope or vertical ambiguity due to improper setting of coarse-grained vertical coverage threshold. The erroneous relationship is written into downstream systems such as planning and operation and maintenance along the data chain, causing inaccurate collision analysis and construction risks. There is a lack of dynamic correction mechanism. Most existing threshold libraries are statically encoded and it is difficult to adjust them in real time based on historical error statistics tables. Once extreme mapping errors occur, the thresholds are prone to overall drift, further amplifying semantic noise. The spatiotemporal evolution dimension is missing. Most two-dimensional or isolated maps do not introduce a unified time index tree and spatial index tree, making it impossible to continuously track the evolution of pipeline relocation, building additions, etc., and making it difficult to support the full life cycle management of urban lifelines.

[0005] To address the above problems, this invention proposes a solution. Summary of the Invention

[0006] The purpose of this invention is to provide a method for constructing a three-dimensional semantic knowledge graph of spatiotemporal relationships of geographic entities. This method addresses the problem that existing knowledge graph construction methods employ a lightweight vertical topology construction strategy that simultaneously processes above-ground and underground data, which frequently leads to vertical semantic ambiguity in the determination of pipeline relationships and continues to spread along the data chain to various downstream business systems.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A method for constructing a three-dimensional semantic knowledge graph of spatiotemporal relationships of geographic entities includes the following: obtaining a unified benchmark geographic entity flow; the unified benchmark geographic entity flow contains a set of geographic entities with geographic entity identifiers, spatial coordinates, bounding box vertical extrema, and timestamps;

[0009] The XY-axis projection overlap detection is used to select a set of mutually overlapping geographic entity pairs and read the sensor accuracy of each geographic entity; the vertical positional relationship of the mutually overlapping geographic entity pairs is determined based on the bounding box vertical extremum.

[0010] According to the sensor accuracy level, the preset layer thickness mapping table is queried to determine the coarse-grained cutting layer thickness. For geographical entities that cover each other, adaptive coarse-grained vertical sampling is performed in the Z-axis direction to calculate the coarse-grained vertical coverage and traversal depth range. The results are compared with the threshold values ​​of each boundary parameter in the preset first threshold library to generate preliminary geographical entity vertical relationship status labels.

[0011] When the initial determination of the vertical relationship status is ambiguous or enclosed, the local depth range is determined based on the vertical extreme value of the bounding box and the traversal depth range. Within the local depth range, the layer thickness is resampled with a preset fixed fine-grained cutting thickness. The threshold of each boundary parameter is revised by combining historical error statistics to generate a second threshold library for re-determination. The final geographic entity vertical relationship status label and confidence level are then output.

[0012] Relationships with confidence levels meeting the preset release threshold are written into the graph database to form revised graph versions. Historical versions are managed through version chains, while low-confidence relationships are pushed for manual review.

[0013] As a further aspect of the present invention: the acquisition logic of the unified benchmark geographic entity stream is as follows: a time alignment algorithm and a spatial coordinate unification algorithm are executed on the ground 3D point cloud data, underground facility detection profile data, and building information data acquired within the urban area to generate a multi-source geographic entity data stream; geometric abstraction is performed on the multi-source geographic entity data stream, that is, the ground 3D point cloud data is clustered into independent geographic entities; spatial interpolation reconstruction is performed on the underground facility detection profile data to identify and reconstruct the 3D entity outline of the underground facilities; geometric geographic entities are directly extracted from the building information data, and a geographic entity identifier is assigned to each geographic entity.

[0014] Mapping accuracy based on the original data of each geographic entity and preset accuracy threshold and Compare;

[0015] When the surveying accuracy is less than or equal to the preset accuracy threshold At that time, the mapping accuracy of the corresponding geographic entity will be classified as high accuracy.

[0016] When the surveying accuracy is less than or equal to the preset accuracy threshold If the accuracy is low, the mapping accuracy of the corresponding geographic entity will be classified as medium accuracy; otherwise, the mapping accuracy of the corresponding geographic entity will be classified as low accuracy.

[0017] And write the corresponding accuracy file to the sensor accuracy file field;

[0018] Calculate the 3D bounding box for each geographic entity and obtain the vertical extreme value data of the bounding box, including the highest and lowest values ​​of the bounding box;

[0019] Using geographic entity identifiers, spatial coordinates, bounding box vertical extrema, and timestamps as index keys, spatial index trees and temporal index trees are constructed respectively, and a unified benchmark geographic entity flow containing the above data is output.

[0020] As a further solution of the present invention: the selection of mutually overlapping geographic entity pairs based on XY axis projection overlap detection specifically involves: extracting newly added and updated geographic entities within a preset time window through a time index tree, forming a geographic entity set, and performing timestamp validity verification.

[0021] For each geographic entity in the verified geographic entity set, its geographic grid hash encoding sequence in the spatial index tree is read; a first-stage coarse filtering is performed using the geographic grid hash prefix matching rule; if the first N characters of the geographic grid hash strings of two geographic entities are the same, they are determined to be a potentially spatially overlapping geographic entity pair; and written into the candidate geographic entity pair set; XY axis projection overlap detection is performed on the candidate geographic entity pairs in the candidate geographic entity pair set:

[0022] The overlapping area is calculated by projecting the geometry of candidate geographic entities onto the XY plane.

[0023] When the ratio of the overlapping area to the projected area of ​​the upper-layer geographic entity is greater than or equal to the preset projection overlap threshold, it is marked as a pair of geographic entities that cover each other and the sensor accuracy is recorded; otherwise, it is excluded.

[0024] As a further aspect of the present invention: the determination of the top-bottom orientation relationship of mutually overlapping geographic entity pairs based on the vertical extrema of the bounding box includes: for each mutually overlapping geographic entity pair in the set of mutually overlapping geographic entity pairs... ,in M represents the number of mutually overlapping geographic entity pairs; compare their bounding box vertical extremes; denote the highest value of the bounding box of geographic entity A as . minimum value of the bounding box The highest value of the bounding box of geographic entity B is minimum value of the bounding box If a comparison reveals Greater than If the value is greater than or equal to the value of B, it means that geographic entity A is higher than geographic entity B. In this case, geographic entity A is marked as the upper geographic entity and geographic entity B is marked as the lower geographic entity; otherwise, they are swapped.

[0025] As a further aspect of the present invention: the preset layer thickness mapping table is generated as follows: according to the sensor accuracy level, for each sensor accuracy level, the historical mapping error is first multiplied by a preset error amplification factor, compared with the industry recommended minimum layer thickness, and the larger value is taken as the candidate layer thickness, and then rounded up according to the preset layer thickness rounding step; the sensor accuracy level identifier and the final layer thickness are written into the mapping table to form the preset layer thickness mapping table; for each pair of mutually covered entities, the corresponding sensor accuracy level field is read, and the preset layer thickness mapping table is queried, and the larger section layer thickness value is selected as the coarse-grained section layer thickness for the initial vertical sampling of the pair of mutually covered geographical entities.

[0026] As a further aspect of the present invention: the calculation of coarse-grained vertical coverage and traversal depth interval specifically involves: after determining the vertical relationship of each pair of mutually covering geographic entities and the initial coarse-grained section thickness of the vertical sampling, the intersection elevation interval defined by the highest value of the bounding box of the lower geographic entity and the lowest value of the bounding box of the upper geographic entity; generating a horizontal slice sequence at equal intervals according to the coarse-grained section thickness in the Z-axis direction; performing the following operations on each slice: calculating the intersection area of ​​the horizontal projected polygons of the mutually covering geographic entity pairs; accumulating the intersection area to obtain the cumulative value of the overlapping surface; accumulating the projected area of ​​the upper geographic entity to obtain the cumulative value of the upper entity's projected surface;

[0027] The coarse-grained vertical coverage is the ratio of the accumulated value of the overlapping surface to the accumulated value of the projected surface of the upper entity; the traversal depth interval is defined by its lower and upper boundaries.

[0028] The value of the difference between the highest value of the bounding box of the lower geographic entity and the lowest value of the bounding box of the upper geographic entity is defined as the lower bound of the crossing depth interval; the value of the difference between the lowest value of the bounding box of the lower geographic entity and the highest value of the bounding box of the upper geographic entity is defined as the upper bound of the crossing depth interval; the crossing depth interval is encapsulated by the lower bound and the upper bound of the crossing depth interval.

[0029] The coarse-grained vertical coverage and traversal depth range are combined to form a coarse-grained vertical index.

[0030] As a further aspect of the present invention: the threshold values ​​of each boundary parameter in the preset first threshold library include: an upper boundary threshold composed of an upper coverage threshold and a lower depth threshold; and a lower boundary threshold composed of a lower coverage threshold and a lower depth threshold.

[0031] Wherein: the coverage upper limit threshold is determined based on the minimum vertical overlap ratio set by industry safety standards; the depth upper limit threshold is the industry minimum safe crossing depth threshold plus a preset equipment error;

[0032] The coverage lower limit threshold is determined based on the minimum overlap ratio obtained from historical collision case statistics; the depth lower limit threshold is the minimum envelope threshold of historical collision cases plus a preset safety factor;

[0033] And it always satisfies the following conditions: the upper limit of depth is less than the lower limit of depth, and the upper limit of coverage is greater than the lower limit of coverage.

[0034] As a further aspect of the present invention: the generation of preliminary geographic entity hierarchical relationship status labels includes:

[0035] If the coarse-grained vertical coverage is less than the coverage upper limit threshold and the upper bound of the traversal depth interval is less than the depth upper limit threshold, it is marked as an upper and lower safety label.

[0036] If the coarse-grained vertical coverage is greater than or equal to the lower coverage threshold or the lower boundary of the traversal depth interval is greater than or equal to the lower depth threshold, it is marked as an upper or lower envelope label; otherwise, it is marked as an upper or lower ambiguous label.

[0037] As a further aspect of the present invention: the determination of the local depth range based on the vertical extreme value of the bounding box and the crossing depth interval includes: obtaining the vertical extreme value of the bounding box and the upper and lower bounds of the crossing depth interval for mutually overlapping geographical entities marked as ambiguous or enveloped in the preliminary geographical entity vertical relationship status labels; at this time, there are mutually overlapping geographical entities marked as ambiguous or enveloped whose corresponding lower bound of the crossing depth interval is negative; the lower bound of the local depth range is equal to the lowest value of the upper entity bounding box minus a preset safety margin; the upper bound of the local depth range is equal to the highest value of the lower entity bounding box plus the preset safety margin.

[0038] The lower bound and the upper bound of the local depth range constitute the local depth range.

[0039] As a further aspect of the present invention: take half of the minimum cutting layer thickness of all sensor accuracy fields and round it up to obtain a fixed fine-grained cutting layer thickness; resample mutually overlapping geographic entity pairs with ambiguity or envelope within a local depth range to obtain a fine vertical index; subtract the fine vertical index from the archived coarse-grained vertical index to obtain a coarse-fine difference value including coverage difference and depth difference; call the mean and variance of the historical error statistics table to obtain its corresponding standard deviation; when any dimension of the coarse-fine difference value exceeds one standard deviation, expand the corresponding boundary parameter threshold in the first threshold library outward by a preset micro-step to form a second threshold library; otherwise, keep the original threshold.

[0040] The beneficial effects of this invention are:

[0041] (1) This invention introduces a two-level coarse-fine cross-sectional sampling with sensor accuracy adaptation and a dynamic threshold library iteration driven by historical error on a unified spatiotemporal reference: First, the coarse-grained layer thickness is determined adaptively based on the accuracy level to obtain the initial vertical relationship. Then, the threshold is revised incrementally with fine-grained resampling and linkage error statistics in the local depth range, thereby significantly reducing the vertical semantic ambiguity in the vertical position determination and calculating the quantitative confidence for each relationship. If the difference between coarse and fine layers exceeds one standard deviation, the threshold is automatically expanded to ensure that the determination result is both in line with the real-time error and avoids boundary drift caused by a single anomaly, and outputs a stable high-confidence label. The high-confidence edge is solidified through the version chain and released to the outside, while the low-confidence edge enters the error interception layer to wait for manual review. A dynamic closed loop is realized from data input to threshold correction to version sedimentation to feedback and re-input, so that the three-dimensional semantic knowledge graph has the characteristics of real-time updating, transparent evolution, high reliability and traceability, and continuously provides accurate and stable vertical topological relationship data for downstream businesses such as pipeline protection, risk warning and spatial planning. Attached Figure Description

[0042] The invention will now be further described with reference to the accompanying drawings.

[0043] Figure 1 This is a schematic diagram of the overall method flow for constructing a three-dimensional semantic knowledge graph of spatiotemporal relationships of geographic entities according to the present invention;

[0044] Figure 2 This is a flowchart illustrating step two of the method for constructing a three-dimensional semantic knowledge graph of spatiotemporal relationships of geographic entities according to the present invention.

[0045] Figure 3 This is a flowchart illustrating step three of the method for constructing a three-dimensional semantic knowledge graph of spatiotemporal relationships of geographic entities in this invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Example 1:

[0048] Please see Figure 1 As shown, this invention provides a method for constructing a three-dimensional semantic knowledge graph of spatiotemporal relationships of geographic entities, comprising the following steps:

[0049] Step 1: Collect and time-align urban ground 3D point cloud data, underground facility detection profile data, and building information data to form a multimodal temporal geographic entity flow; perform spatial normalization and elevation conversion on the multimodal temporal geographic entity flow to generate a geometric basis set and calculate a 3D bounding box on each geographic entity in the geometric basis set, and record the vertical extrema of the geographic entity bounding box; construct a spatial index tree and a temporal index tree respectively using geographic entity identifier, spatial coordinates, bounding box vertical extrema, and timestamp as index keys, and output a unified benchmark geographic entity flow;

[0050] The ground 3D point cloud data, underground facility detection profile data and building information data acquired within the urban area are accessed and preliminarily processed respectively. Time alignment algorithm and spatial coordinate unification algorithm are applied to the three types of data respectively to generate multi-source geographic entity data streams.

[0051] The time alignment algorithm includes: using the Global Navigation Satellite System (GNSS) Pulse Second (PPS) signal as a high-precision time reference to calibrate the device clock corresponding to the ground 3D point cloud data; using the Network Time Protocol (NTP) to calibrate the built-in clock of the underground facility detection profile data acquisition device in real time; using the GNSS time axis as a reference, calculating the time difference between the device clock and the GNSS clock of the underground facility detection profile data in real time, and adjusting the difference in real time according to a preset threshold until the difference stabilizes within the set threshold in multiple consecutive time windows, thus completing the time alignment between the data sources;

[0052] The specific spatial coordinate unification algorithm includes: performing spatial coordinate transformation on the ground 3D point cloud data and the underground facility detection profile data respectively, and unifying the data to the national unified coordinate system and the 1985 National Elevation Datum; converting the latitude, longitude and ellipsoidal height coordinates of the ground 3D point cloud data to the national unified coordinate system and the 1985 National Elevation Datum coordinate system through a seven-parameter transformation algorithm; mapping the relative coordinates of the underground facility detection data to the unified national coordinate system through a two-dimensional affine transformation and linear interpolation algorithm; performing corresponding spatial coordinate transformation on the building information data, and manually verifying the transformation accuracy to ensure data accuracy;

[0053] Furthermore, geometric abstraction is performed on the multi-source geographic entity data stream to cluster the ground 3D point cloud data into independent geographic entity features and identify independent geometric geographic entities; spatial interpolation reconstruction is performed on the underground exploration data to identify and reconstruct the 3D entity outline of underground facilities; geometric geographic entities are directly extracted from building information data and a geographic entity identifier is assigned to each geometric geographic entity.

[0054] Furthermore, the mapping accuracy of reading the original data of each geographic entity. The specific values ​​are obtained from historical calibration, and the mapping accuracy of the original data of each geographic entity is compared with the preset accuracy threshold. and Compare;

[0055] when At that time, the mapping accuracy of the corresponding geographic entity will be classified as high accuracy.

[0056] when If the accuracy is low, the mapping accuracy of the corresponding geographic entity will be classified as medium accuracy; otherwise, the mapping accuracy of the corresponding geographic entity will be classified as low accuracy.

[0057] This leads to the sensor accuracy profile field, where the high accuracy profile is marked. Medium precision marking Low-precision marking And write it into the entity attribute;

[0058] For each of the aforementioned geometric geographic entities, a 3D bounding box is calculated. The minimum bounding rectangle algorithm is used to determine the extreme coordinates of each geographic entity in the vertical direction, thus obtaining the vertical extreme value data of the bounding box, including the highest value of the bounding box. minimum value of the bounding box ;

[0059] Using the geographic entity identifier, spatial coordinates, bounding box vertical extremum, and timestamp of the geometric geographic entity as index keys, a spatial index tree and a time index tree are constructed respectively to achieve fast data retrieval and management;

[0060] Ultimately, the device outputs a unified benchmark geographic entity stream through a unified interface, which is a multimodal data stream that has undergone time alignment, spatial coordinate unification, geometric abstraction, bounding box calculation, and index construction. This stream is then used for subsequent vertical topological relationship determination and knowledge graph construction.

[0061] Through the above implementation methods, the embodiments of the present invention achieve precise unification of different data sources on spatiotemporal references and standardized extraction of geometric features, providing a solid data foundation for the accurate determination of the vertical relationship between geographic entities in the next step, and significantly improving the accuracy and efficiency of subsequent analysis and processing.

[0062] like Figure 2 As shown, step two: extract data from the baseline geographic entity stream according to a set time window; perform XY projection overlap detection to filter mutually overlapping geographic entity pairs; on each mutually overlapping geographic entity pair, use a coarse-grained cutting layer thickness that matches the original sensor accuracy to perform vertical sampling, and obtain the coarse-grained vertical coverage and traversal depth interval as coarse-grained vertical indices; call a preset first threshold library, compare the two coarse-grained vertical indices with the preset first threshold library, and generate preliminary geographic entity vertical relationship status labels;

[0063] The preset time window Δt is extracted by the time index tree, which is 1 hour by default in this embodiment; and the time index tree is called to extract the geographic entities that are added or updated within the current preset time window Δt from the unified benchmark geographic entity stream, so as to obtain a set of geographic entities and perform a validity check on the timestamp of each geographic entity in the set of geographic entities to ensure that it meets the time alignment requirements of step one.

[0064] For each geographic entity in the set of verified geographic entities, its spatial grid code in the spatial index tree is read to obtain the geographic grid hash code sequence; the geographic grid hash prefix matching rule is used to perform the first stage of coarse filtering to obtain a set of candidate geographic entity pairs.

[0065] It should be noted that the geographic grid hash prefix matching rule encodes the two-dimensional coordinates of each geographic entity into a geographic grid hash string. The longer the string, the smaller the grid size. If two geographic entities are identical in the first N positions (N=6 characters by default), they will fall into the same 64×64 meter grid, which can be considered as potential overlap. Candidate pairs are quickly screened out using this rule, and then precise area superposition is performed, which significantly reduces the computational load compared to direct pairwise calculation. The prefix length N of the geographic grid hash can be adjusted according to application requirements. For example, to expand the candidate set and ensure no missed detections, the prefix length can be shortened to use a larger grid. To reduce false detections, the prefix can be lengthened to improve matching accuracy. Through the above coarse filtering rule based on geographic grid hash, the computational load of calculating each pair of entities is significantly reduced, achieving efficient locking of potential hierarchical entity pairs.

[0066] The XY-axis projection area superposition operation is performed on the candidate geographic entity pair set to calculate the projection overlap area, which determines the actual degree of overlap of the geographic entity pair on the horizontal plane. The specific method is as follows: the geometry of each pair of candidate geographic entities A and B is projected onto the same horizontal plane, i.e., the XY plane, and the area of ​​their projection overlap area is calculated; at the same time, the projection area of ​​the upper-level geographic entity is obtained, tentatively designated as A; then the proportion of the projection overlap area to the projection area of ​​the upper-level geographic entity is calculated and compared with a preset projection overlap threshold; this projection overlap threshold is a percentage value set by professionals in the relevant field based on experience.

[0067] If the ratio of the projected overlap area of ​​each geographic entity pair to the projected area of ​​the upper-layer geographic entity is greater than or equal to the preset projection overlap threshold, then it is marked as a set of mutually overlapping geographic entity pairs, and the sensor accuracy file is recorded for each mutually overlapping geographic entity pair; otherwise, it is excluded.

[0068] For each mutually overlapping geographic entity pair in the set of mutually overlapping geographic entity pairs ,in M represents the number of mutually overlapping geographic entity pairs; compare their bounding box vertical extremes; denote the highest value of the bounding box of geographic entity A as . minimum value of the bounding box The highest value of the bounding box of geographic entity B is minimum value of the bounding box If a comparison reveals Greater than If A is above B, it means that geographic entity A is above geographic entity B, i.e., A is above B and B is below. In this case, A is marked as the upper geographic entity and B is marked as the lower geographic entity; otherwise, they are swapped.

[0069] For example, a building A and an underground pipeline B constitute a geographical entity pair. Typically, the lowest point of the building's foundation, i.e., the depth of the foundation, is higher than the lowest point where the pipeline is buried. Therefore... Greater than Therefore, it can be determined that the building is on top and the pipeline is below; for example, if an elevated bridge A and a ground building B overlap each other, and the height of the bottom of the bridge is higher than the height of the roof of the building, then A (the bridge) is the upper entity and B is the lower entity.

[0070] Based on the preset error amplification factor, the initial value of which is given by the developers in conjunction with historical surveying error experience; the industry-recommended minimum layer thickness is set according to current standards such as the "Urban Pipeline Detection Technical Standard"; the preset layer thickness rounding step is a fixed value of 5cm for easy rasterization storage; and the layer thickness mapping table version number is automatically assigned by the system when it is first generated, and is automatically extended with each subsequent upgrade; then, the sensor accuracy range field is retrieved in sequence to obtain the high accuracy range, medium accuracy range, and low accuracy range, and the upper limit of surveying error for each range is retrieved; for each sensor accuracy range, the historical measurement error is first... The plotting error is multiplied by a preset error amplification factor and compared with the industry-recommended minimum layer thickness. The larger value is taken as the candidate layer thickness. Then, the layer thickness is rounded up by the preset layer thickness rounding step. If the candidate value is not divisible by the preset layer thickness rounding step, the layer thickness is increased by centimeters until it is divisible, ensuring that all layer thicknesses fall into a unified preset layer thickness arithmetic sequence. After rounding, the precision file identifier - final layer thickness is written into the layer thickness mapping table, and the generation date and layer thickness mapping table version number are recorded. The layer thickness mapping table is output with the unified reference stream. Subsequent steps can determine the sectioning layer thickness by querying the layer thickness mapping table based on the entity, without the need for repeated calculations.

[0071] For each pair of mutually covered entities, read the corresponding sensor accuracy file field, query the preset layer thickness mapping table, and select the larger section layer thickness value as the coarse-grained section layer thickness for the first vertical sampling of the pair of mutually covered geographical entities.

[0072] After determining the vertical relationship between each pair of overlapping geographic entities and the initial coarse-grained section thickness for vertical sampling, a sequence of horizontal slices is generated along the Z-axis at equal intervals according to the corresponding coarse-grained section thickness, starting from and ending at the intersection elevation interval defined by the highest value of the lower geographic entity's bounding box and the lowest value of the upper geographic entity's bounding box. For each slice, the horizontal projection polygons of the overlapping geographic entity pairs are calculated, and the intersection area is determined. The accumulated intersection area yields the accumulated value of the overlapping surface, and the accumulated projection area of ​​the upper geographic entity yields the accumulated value of the upper entity's projection surface. The accumulated values ​​of the overlapping surface and the upper entity's projection surface are then combined. The ratio of the values ​​is recorded as the coarse-grained vertical coverage. The highest value of the bounding box of the lower geographic entity and the lowest value of the bounding box of the upper geographic entity are read, and an elevation difference calculation is performed to obtain the minimum net distance between the closest points of mutually covered geographic entity pairs in the vertical direction. Conversely, the maximum net distance between the farthest points of mutually covered geographic entity pairs in the vertical direction can be obtained in the same way. The minimum net distance is recorded as the lower bound of the crossing depth interval, and the maximum net distance is recorded as the upper bound of the crossing depth interval. The crossing depth interval is encapsulated in the form of the lower bound and the upper bound. Finally, the coarse-grained vertical coverage and the crossing depth interval are summarized to obtain the coarse-grained vertical index.

[0073] Retrieve a preset first threshold library, which includes: an upper limit boundary threshold composed of an upper coverage threshold and an upper depth threshold; and a lower limit boundary threshold composed of a lower coverage threshold and a lower depth threshold.

[0074] Wherein: the coverage upper limit threshold is determined based on the minimum vertical overlap ratio set by industry safety standards; the depth upper limit threshold is the industry minimum safe crossing depth threshold plus a preset equipment error;

[0075] The coverage lower limit threshold is determined based on the minimum overlap ratio obtained from historical collision case statistics; the depth lower limit threshold is the minimum envelope threshold of historical collision cases plus a preset safety factor;

[0076] And it always satisfies the following conditions: the upper limit of depth is less than the lower limit of depth, and the upper limit of coverage is greater than the lower limit of coverage.

[0077] The coarse-grained vertical coverage and the upper and lower bounds of the crossing depth interval obtained above are compared with the preset first threshold library to achieve the initial marking of the vertical relationship status of geographic entities.

[0078] If the coarse-grained vertical coverage is lower than the coverage upper limit threshold and the upper limit of the traversal depth interval is lower than the depth upper limit threshold, then the two geographical entities that cover each other are sufficiently separated vertically, and the labels can be directly assigned vertically for safety.

[0079] If the coarse-grained vertical coverage is not lower than the lower coverage threshold or the lower boundary of the traversal depth interval is not lower than the lower depth threshold, then it is determined that the upper geographic entity has significantly enclosed the lower geographic entity, and the upper and lower envelopes are marked.

[0080] The remaining cases falling between the two sets of boundaries are classified as vertical ambiguity.

[0081] For mutually overlapping geographical entity pairs that are determined to be secure in terms of hierarchical relationship, the hierarchical relationship is clear and unambiguous, and they can be directly entered into the knowledge graph as a hierarchical relationship edge; for mutually overlapping geographical entity pairs that are determined to be ambiguous or in an envelope state, further analysis is required.

[0082] like Figure 3 As shown, step three involves: triggering a re-analysis for geographic entity pairs marked as ambiguous or enveloped in the initial geographic entity top-bottom relationship status labels; performing a combined calculation based on the vertical extreme value of the geographic entity bounding box and the traversal depth range to obtain a local depth range; resampling vertical data within the local depth range using a fixed fine-grained cutting layer thickness; incrementally revising the first threshold library by combining fine-grained vertical indices and historical error statistics to generate a second threshold library; re-determining the geographic entity top-bottom relationship using the second threshold library to obtain the final geographic entity top-bottom relationship status labels and confidence levels; and marking entity pairs that do not meet the threshold determination conditions as requiring manual review.

[0083] The system retrieves the vertical extreme values ​​and upper and lower bounds of the crossing depth intervals for mutually overlapping geographic entities marked as ambiguous or enveloped in the initial geographic entity vertical relationship status labels. At this point, if the lower bound of the crossing depth interval for a mutually overlapping geographic entity marked as ambiguous or enveloped is negative, it indicates that the upper entity has been inserted into the height domain of the lower entity. The logic for obtaining its corresponding local depth range is as follows:

[0084] The lower bound of the local depth range is taken as the lowest value of the bounding box of the geographic entity above, and a preset safety margin is reserved below it; this preset safety margin is taken as half the absolute value of the lower bound of the depth range, and is used to compensate for coarse sampling error.

[0085] The upper limit of the local depth range is taken as the highest value of the bounding box of the geographic entity below, and another preset safety margin is reserved upwards; the logic for obtaining the preset safety margin in this invention is based on the absolute value ratio of the lower limit of the depth range, to ensure that the top of the envelope is completely covered.

[0086] If the lower bound of the traversal depth range is detected to be non-negative and the minimum value of the bounding box of the upper geographic entity is higher than the maximum value of the bounding box of the lower geographic entity, then the vertical separation of the overlapping geographic entity pair is sufficient, this pair retains the safety label and exits this process; the final generated local depth range only accounts for a small segment of the overall height, and subsequent fine-grained cutting is only performed within this range.

[0087] For pairs of geographical entities that potentially overlap within a local depth range, vertical data for that local segment is resampled with a fixed fine-grained section thickness; the fixed fine-grained section thickness is obtained by rounding down to half of the minimum section thickness in all sensor accuracy profile fields.

[0088] Simultaneously, the built-in historical error statistics table is obtained, which includes the difference between coarse and fine granular indicators in previous batches; the deviation between the results of manual review and confirmation and the results of automatic judgment; the difference between the depth measured on-site and the system prediction; and new entries are written after each round of the process is completed, and the mean and variance are calculated periodically to obtain the latest error distribution.

[0089] Furthermore, based on a fixed fine-grained cutting layer thickness, ambiguous or enveloped geographic entity pairs are first fine-grained cut within a local depth range using a fixed fine-grained cutting layer thickness to obtain fine vertical indices; then, the fine vertical indices are compared with the coarse vertical indices archived in the previous step for the same geographic entity pair to obtain coarse-fine difference values ​​that include coverage difference and depth difference.

[0090] The standard deviation is obtained by linking the coarse and fine differences with the mean and variance of the historical error statistics table. This standard deviation is then used to adjust the thresholds for each boundary parameter in the first threshold library. Incremental revisions are implemented according to the principle that if the coarse or fine difference exceeds one standard deviation from the historical average, the corresponding boundary is expanded outward by a preset micro-step, thus obtaining the second threshold library; otherwise, the original threshold is maintained. This revision logic ensures that threshold adjustments are both based on real-time errors and constrained by historical statistics, avoiding boundary drift caused by a single anomaly. As the second threshold library is updated, the process of determining the vertical relationship in step two is re-executed, and the final geographical entity vertical relationship status label is output. The distance of each coarse-grained vertical index value from each boundary parameter threshold is recorded and defined as its corresponding confidence score. If the label is still ambiguous or enveloped and the confidence score is lower than the preset release threshold, the entity pair is marked as requiring manual review and entered into the error interception queue. Other high-confidence score results are directly written into the knowledge graph revision version, thus completing the closed loop of fine-grained correction and dynamic threshold update.

[0091] Step 4: Write the final geographic entity hierarchical relationship status labels into the graph database to form the initial map version; update the initial map with the high-confidence hierarchical relationship status obtained after threshold correction to form a revised map version; link the revised version with the initial version through version linking relationships; for geographic entity pairs that are still ambiguous or enveloped after correction and whose confidence is lower than the preset threshold, perform error interception; automatically push the intercepted low-confidence geographic entity pairs to the manual review queue; when a new round of multi-source data is accessed, automatically re-trigger the entire process from Step 2 to Step 4 to form a dynamic closed loop;

[0092] Based on the status labels and confidence scores of the corresponding final geographic entity top-bottom relationships for each mutually overlapping geographic entity, an initial map version is first created in the graph database. Specifically, using the geographic entity ID in the unified benchmark geographic entity flow as a node, attributes such as location coordinates, timestamps, and sensor accuracy fields are written. Top-bottom safe, top-bottom envelope, and top-bottom ambiguous label relationships are written as directed edges, and the confidence scores are stored synchronously. Subsequently, based on a preset release threshold, top-bottom safe or top-bottom envelope label relationships with sufficient confidence scores are extracted, a filtering and reinforcement process is performed, and a revised map version is generated. Edges with low confidence scores or those that are still ambiguous are marked as unconfirmed and automatically pushed to the manual review queue.

[0093] Simultaneously, a version chain is established within the graph database, binding the initial graph version with the revised graph version, recording differences and their causes, and enabling traceability of the source of the judgment. The external API only exposes edges with confirmation markers in the revised graph version, ensuring that all data output from the knowledge graph to downstream businesses is highly reliable, forming an error interception layer. This error interception layer can explicitly mark such data as requiring manual review, prohibiting the API interface from publishing such low-confidence or uncertain data to downstream business systems, ensuring that error data exists only within the internal knowledge graph and is not accessed externally.

[0094] When a new batch of multi-source data enters the unified benchmark geographic entity flow triggering time window Δt, the scheduler immediately restarts steps two and three, recalculates the vertical relationships of newly added or changed geographic entities, and updates edge attributes based on the threshold correction results; after completion, a new revised map version is generated and automatically written into the version chain; at the same time, the differences between fine-grained judgment and review feedback are written into the historical error statistics table, driving the rolling calibration of the threshold library and layer thickness mapping table, further improving the judgment accuracy of the next round; thus forming a dynamic closed loop from data input to judgment output to threshold correction to version accumulation to feedback and re-input;

[0095] With the help of the aforementioned version management, error interception and self-learning mechanisms, the spatiotemporal three-dimensional semantic knowledge graph of geographic entities constructed by this invention has the characteristics of real-time updating, transparent evolution and high reliability. It can continuously present the vertical topological relationship of various facilities in urban space and provide stable and reliable data support for applications such as visualized operation and maintenance, risk warning and spatial planning.

[0096] In this embodiment, a unified benchmark geographic entity flow is acquired, comprising a set of geographic entities including geographic entity identifiers, spatial coordinates, bounding box vertical extrema, and timestamps. Based on XY-axis projection overlap detection, a set of mutually overlapping geographic entity pairs is selected, and sensor accuracy settings are read. The vertical orientation relationship is determined based on the bounding box vertical extrema. A layer thickness mapping table is queried according to the sensor accuracy settings to determine the coarse-grained sectioning layer thickness. Adaptive coarse-grained vertical sampling is performed in the Z-axis direction to calculate the coarse-grained vertical coverage and traversal depth range, which is compared with a first threshold library to generate preliminary vertical relationship status labels. For ambiguous or enveloped entity pairs, a local depth range is determined based on the bounding box vertical extrema and traversal depth range. Resampling is performed with a fixed fine-grained sectioning layer thickness, and a second threshold library is generated by combining historical error statistics to re-determine the relationship. The final status label and confidence level are output. High-confidence relationships are written into the graph database to form a revised map version and managed through a version chain, while low-confidence relationships are pushed for manual review. This invention eliminates vertical semantic ambiguity and improves the accuracy of determining the vertical relationship of geographic entities through a dual coarse-grained and fine-grained determination mechanism and a dynamic threshold calibration mechanism.

[0097] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0098] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0099] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0100] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0101] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0102] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for constructing a three-dimensional semantic knowledge graph of spatio-temporal relations of geographical entities, characterized in that, This includes the following: obtaining a unified benchmark geographic entity flow; the unified benchmark geographic entity flow contains a set of geographic entities with geographic entity identifiers, spatial coordinates, bounding box vertical extrema, and timestamps; The XY-axis projection overlap detection is used to select a set of mutually overlapping geographic entity pairs and read the sensor accuracy of each geographic entity; the vertical positional relationship of the mutually overlapping geographic entity pairs is determined based on the bounding box vertical extremum. According to the sensor accuracy setting, a preset layer thickness mapping table is consulted to determine the coarse-grained cutting layer thickness. For geographical entities that overlap, adaptive coarse-grained vertical sampling is performed in the Z-axis direction to calculate the coarse-grained vertical coverage and crossing depth range. This is then compared with the threshold values ​​of each boundary parameter in a preset first threshold library to generate preliminary geographical entity vertical relationship status labels. The threshold values ​​of each boundary parameter in the preset first threshold library include: an upper boundary threshold composed of an upper coverage threshold and an upper depth threshold; and a lower boundary threshold composed of a lower coverage threshold and a lower depth threshold. The upper coverage threshold is determined based on the minimum vertical overlap ratio set by industry safety standards; the upper depth threshold is the industry minimum safe crossing depth threshold plus a preset equipment error; the lower coverage threshold is determined based on the minimum overlap ratio obtained from historical collision case statistics; and the lower depth threshold is the minimum envelope threshold of historical collision cases plus a preset safety factor. Furthermore, the upper depth threshold is always less than the lower depth threshold, and the upper coverage threshold is greater than the lower coverage threshold. When the initial determination of the vertical relationship status is ambiguous or enclosed, the local depth range is determined based on the vertical extreme value of the bounding box and the traversal depth range. Within the local depth range, the layer thickness is resampled with a preset fixed fine-grained cutting layer. The threshold of each boundary parameter is revised by combining historical error statistics to generate a second threshold library for re-determination. The final geographic entity vertical relationship status label and confidence score are then output. Relationships with confidence levels meeting the preset release threshold are written into the graph database to form revised graph versions. Historical versions are managed through version chains, while low-confidence relationships are pushed for manual review.

2. The method for constructing a three-dimensional semantic knowledge graph of spatiotemporal relationships of geographic entities according to claim 1, characterized in that, The logic for acquiring the unified benchmark geographic entity stream is as follows: A time alignment algorithm and a spatial coordinate unification algorithm are executed on the ground 3D point cloud data, underground facility detection profile data, and building information data acquired within the urban area to generate a multi-source geographic entity data stream; geometric abstraction is then performed on the multi-source geographic entity data stream, that is, the ground 3D point cloud data is clustered into independent geographic entities; Spatial interpolation reconstruction is performed on the underground facility detection profile data to identify and reconstruct the three-dimensional solid outline of the underground facility; Geometric geographic entities are extracted directly from building information data, and a geographic entity identifier is assigned to each geographic entity. Mapping accuracy based on the original data of each geographic entity and preset accuracy threshold and Compare; When the surveying accuracy is less than or equal to the preset accuracy threshold At that time, the mapping accuracy of the corresponding geographic entity will be classified as high accuracy. When the surveying accuracy is less than or equal to the preset accuracy threshold If the accuracy is low, the mapping accuracy of the corresponding geographic entity will be classified as medium accuracy; otherwise, the mapping accuracy of the corresponding geographic entity will be classified as low accuracy. And write the corresponding accuracy file to the sensor accuracy file field; Calculate the 3D bounding box for each geographic entity and obtain the vertical extreme value data of the bounding box, including the highest and lowest values ​​of the bounding box; Using geographic entity identifiers, spatial coordinates, bounding box vertical extrema, and timestamps as index keys, spatial index trees and temporal index trees are constructed respectively, and a unified benchmark geographic entity flow containing the above data is output.

3. The method for constructing a three-dimensional semantic knowledge graph of spatiotemporal relationships of geographic entities according to claim 1, characterized in that, The selection of mutually overlapping geographic entity pairs based on XY axis projection overlap detection specifically involves: extracting updated geographic entities within a preset time window through a time index tree, forming a geographic entity set, and performing timestamp validity verification. For each geographic entity in the set of verified geographic entities, read its geographic grid hash code sequence in the spatial index tree; The first stage of coarse filtering is performed using the geographic grid hash prefix matching rule. If the first N characters of the geographic grid hash strings of two geographic entities are the same, they are determined to be a potentially spatially overlapping geographic entity pair. And write it into the candidate geographic entity pair set; Performing XY axis projection overlap detection on candidate geographic entity pairs in the candidate geographic entity pair set includes: projecting the geometry of the candidate geographic entity onto the XY plane and calculating the overlap area; when the overlap area is greater than or equal to the projection area of ​​the upper geographic entity, it is marked as a mutually overlapping geographic entity pair and the sensor accuracy is recorded; otherwise, it is excluded.

4. The method for constructing a three-dimensional semantic knowledge graph of spatiotemporal relationships of geographic entities according to claim 1, characterized in that, Determining the top-to-bottom orientation of mutually overlapping geographic entity pairs based on bounding box vertical extrema includes: for each mutually overlapping geographic entity pair {A,B} in the set of mutually overlapping geographic entity pairs, where... M represents the number of mutually overlapping geographic entity pairs; compare their bounding box vertical extremes; denote the highest value of the bounding box of geographic entity A as... minimum value of the bounding box The highest value of the bounding box of geographic entity B is minimum value of the bounding box If a comparison reveals Greater than If the value is greater than or equal to the value of B, it means that geographic entity A is higher than geographic entity B. In this case, geographic entity A is marked as the upper geographic entity and geographic entity B is marked as the lower geographic entity; otherwise, they are swapped.

5. The method for constructing a three-dimensional semantic knowledge graph of spatiotemporal relationships of geographic entities according to claim 1, characterized in that, The preset layer thickness mapping table is generated as follows: according to the sensor accuracy level, for each sensor accuracy level, first multiply the historical mapping error by the preset error amplification factor, compare it with the industry recommended minimum layer thickness and take the larger value as the candidate layer thickness, and then round up according to the preset layer thickness. The sensor accuracy profile identifier and the final layer thickness are written into a mapping table to form a preset layer thickness mapping table; for each pair of mutually covered entities, the corresponding sensor accuracy profile field is read, and the preset layer thickness mapping table is queried to select the larger section layer thickness value as the coarse-grained section layer thickness for the initial vertical sampling of the pair of mutually covered geographical entities.

6. The method for constructing a three-dimensional semantic knowledge graph of spatiotemporal relationships of geographic entities according to claim 4, characterized in that, The calculation of coarse-grained vertical coverage and traversal depth interval is specifically as follows: After determining the vertical relationship of each pair of mutually covering geographic entities and the initial coarse-grained section thickness of the vertical sampling, the intersection elevation interval is defined by the highest value of the bounding box of the lower geographic entity and the lowest value of the bounding box of the upper geographic entity; a sequence of horizontal slices is generated at equal intervals along the Z-axis according to the coarse-grained section thickness; for each slice, the following operations are performed: calculate the intersection area of ​​the horizontal projected polygons of the mutually covering geographic entity pairs; accumulate the intersection area to obtain the accumulated value of the overlapping surface; accumulate the projected area of ​​the upper geographic entity to obtain the accumulated value of the upper entity's projected surface; the coarse-grained vertical coverage is the ratio of the accumulated value of the overlapping surface to the accumulated value of the upper entity's projected surface; the traversal depth interval consists of a lower boundary and an upper boundary, and its generation logic is as follows: the value of the highest value of the bounding box of the lower geographic entity minus the lowest value of the bounding box of the upper geographic entity is defined as the lower boundary of the traversal depth interval; the value of the lowest value of the bounding box of the lower geographic entity minus the highest value of the bounding box of the upper geographic entity is defined as the upper boundary of the traversal depth interval; the lower boundary and the upper boundary of the traversal depth interval are used to encapsulate the traversal depth interval. The coarse-grained vertical coverage and traversal depth range are combined to form a coarse-grained vertical index.

7. The method for constructing a three-dimensional semantic knowledge graph of spatiotemporal relationships of geographic entities according to claim 1, characterized in that, Generate preliminary geographic entity hierarchical relationship status labels, including: If the coarse-grained vertical coverage is less than the coverage upper limit threshold and the upper bound of the traversal depth interval is less than the depth upper limit threshold, it is marked as an upper and lower safety label. If the coarse-grained vertical coverage is greater than or equal to the lower coverage threshold or the lower boundary of the traversal depth interval is greater than or equal to the lower depth threshold, it is marked as an upper or lower envelope label; otherwise, it is marked as an upper or lower ambiguous label.

8. The method for constructing a three-dimensional semantic knowledge graph of spatiotemporal relationships of geographic entities according to claim 1, characterized in that, Determining the local depth range based on the vertical extreme value of the bounding box and the crossing depth interval includes: obtaining the vertical extreme value of the bounding box and the upper and lower bounds of the crossing depth interval for mutually overlapping geographical entities marked as ambiguous or enveloped in the preliminary geographical entity vertical relationship status labels; at this time, there are mutually overlapping geographical entities marked as ambiguous or enveloped whose corresponding lower bound of the crossing depth interval is negative; the lower bound of the local depth range is equal to the lowest value of the upper entity's bounding box minus a preset safety margin; the upper bound of the local depth range is equal to the highest value of the lower entity's bounding box plus a preset safety margin. The lower bound and the upper bound of the local depth range constitute the local depth range.

9. The method for constructing a three-dimensional semantic knowledge graph of spatiotemporal relationships of geographic entities according to claim 1, characterized in that, Generating a second threshold library includes: taking half of the minimum cutting layer thickness of all sensor accuracy file fields and rounding it up to a fixed fine-grained cutting layer thickness; resampling mutually overlapping geographic entity pairs with ambiguous or enveloping features within a local depth range to obtain fine vertical indices; subtracting the fine vertical indices from the archived coarse-grained vertical indices to obtain coarse-fine difference values ​​that include coverage difference and depth difference; calling the mean and variance of the historical error statistics table to obtain the corresponding standard deviation; when any dimension of the coarse-fine difference value exceeds one standard deviation, expanding the corresponding boundary parameter threshold in the first threshold library outward by a preset micro-step to form a second threshold library; otherwise, maintaining the original threshold.

Images