Spatial data dynamic association method based on edge collaboration and incremental learning

Abstract

This invention relates to the field of spatial data information retrieval and distributed database technology, specifically a dynamic spatial data association method based on edge collaboration and incremental learning. It achieves distributed acquisition and standardized preprocessing of multi-source heterogeneous spatial data through an edge node cluster, constructs edge-collaborative hierarchical storage and a two-layer distributed spatial index, extracts multi-dimensional spatial features based on a hybrid deep learning network and constructs a distributed feature vector library, completes dynamic spatial data association matching and association graph construction under spatiotemporal constraints, iterates and optimizes the association model and feature library through edge-center collaborative incremental learning, realizes edge-collaborative association retrieval and result output, and completes full-link verification and closed-loop optimization of association results. This invention improves the accuracy and retrieval efficiency of spatial data association matching, fully utilizes edge node resources, adapts to the incremental update requirements of spatial data in distributed edge scenarios, achieves closed-loop optimization of the entire association process, and has good scenario adaptability.

Description

Technical Field

[0001] This invention relates to the field of spatial data information retrieval and distributed database technology, and in particular to a method for dynamic association of spatial data based on edge collaboration and incremental learning. Background Technology

[0002] With the continuous development of geographic information acquisition technology, remote sensing observation technology, and IoT positioning technology, the sources of spatial data are constantly enriching, and the data scale is experiencing explosive growth. Data types cover multiple categories such as geospatial vectors, remote sensing images, point trajectories, and attribute labels, and are widely distributed across multiple levels, including edge acquisition nodes, regional data centers, and cloud storage platforms. Dynamic correlation and matching of spatial data is the core foundation for applications such as spatial information retrieval, geographic scene analysis, and spatiotemporal pattern mining, directly determining the depth and breadth of spatial data applications. Traditional centralized spatial data processing models require transmitting all edge-acquired data back to the central node for processing and correlation analysis, resulting in high data transmission overhead, high processing response latency, and low utilization of edge node resources. This approach is no longer suitable for the real-time processing and dynamic correlation needs of spatial data in scenarios with widely distributed edge node clusters.

[0003] Existing spatial data association technologies have significant shortcomings in multi-source data fusion and association matching. Most existing technologies only perform simple association matching based on spatial geometric distance, failing to fully integrate the topological relationships, semantic attributes, and temporal variation characteristics of spatial data to achieve multi-dimensional comprehensive association. This easily leads to problems such as insufficient matching accuracy and one-sided association relationships. Furthermore, existing technologies have limited standardized processing capabilities for multi-source heterogeneous spatial data, unable to achieve unified processing and accurate spatiotemporal alignment of spatial data from different spatial coordinate systems, data formats, and acquisition frequencies. The lack of a unified data foundation for association matching between spatial data from different sources further reduces the accuracy and reliability of the association results.

[0004] Existing spatial data association technologies also suffer from several shortcomings in terms of adaptability to distributed edge scenarios and dynamic model updates. Most existing technologies employ centralized storage and indexing architectures, lacking collaborative storage and access mechanisms between edge nodes. This results in low efficiency for spatial data association retrieval across edge nodes and fails to fully utilize the local computing power and storage resources of edge nodes for distributed association processing. Furthermore, existing spatial data association models are mostly statically trained, requiring full retraining for new spatial data updates, leading to high computational overhead. This makes them unsuitable for the limited computing resources of edge nodes and fails to meet the dynamic association requirements arising from continuous incremental updates of spatial data. In addition, existing technologies lack end-to-end verification and closed-loop optimization mechanisms for spatial data association results. They cannot continuously optimize the association model, index structure, and collaborative mechanisms based on verification feedback, and the topological consistency and spatiotemporal logical rationality of the association results cannot be consistently guaranteed, making it difficult to adapt to the long-term application requirements of dynamic spatial data association in wide-area distributed scenarios. Summary of the Invention

[0005] This invention provides a method for dynamic association of spatial data based on edge collaboration and incremental learning to solve the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A method for dynamic spatial data association based on edge collaboration and incremental learning includes the following steps: The distributed acquisition and standardized preprocessing steps for multi-source heterogeneous spatial data are used to complete the distributed acquisition of multi-source heterogeneous spatial data such as geospatial vectors, remote sensing images, point trajectories, and attribute labels through edge node clusters. The standardized preprocessing steps include data format normalization, coordinate system unification, noise filtering, and missing value completion, generating standardized spatial data units with unique spatiotemporal labels. The steps of edge-collaborative spatial data hierarchical storage and distributed index construction are used to complete the cold and hot hierarchical distributed storage of spatial data based on the geographical coverage and computing resources of edge nodes, construct a two-layer distributed spatial index structure based on the spatiotemporal dimension, and establish a data synchronization and collaborative access mechanism between edge nodes. The steps of multi-dimensional spatial feature deep extraction and feature vector library construction are used to complete the multi-dimensional deep extraction of spatial geometric features, semantic features, temporal features, and topological relationship features of standardized spatial data based on a hybrid deep learning network, generate high-dimensional spatial feature vectors of unified dimension, and construct a distributed spatial feature vector library in collaboration between edge nodes and central nodes. The steps for dynamic association matching and association graph construction of spatial data under spatiotemporal constraints are used to complete multi-dimensional dynamic association matching of spatial data units based on spatial distance thresholds, topological relationship constraints, semantic similarity matching, and temporal correlation analysis, construct a weighted dynamic association graph of spatial data, and store it in a distributed spatial database. The dynamic update of the association model and the iterative optimization of the feature library based on incremental learning are used to collect new spatial data and association matching results. The incremental update of the spatial feature extraction network and the association matching model is completed through the edge-center collaborative incremental learning framework, and the iterative optimization of the distributed feature vector library and spatial index structure is completed simultaneously. The edge collaborative spatial data association retrieval and result fusion output steps are used to receive spatial data retrieval requests, complete cross-edge node collaborative association retrieval based on dynamic association graphs, and complete the fusion, sorting, deduplication and structured output of multi-node retrieval results. The spatial data association result end-to-end verification and consistency closed-loop optimization steps are used to complete the accuracy verification, topology consistency verification, and spatiotemporal logic verification of the association results. Based on the verification results, the association model, index structure, and collaborative access mechanism are optimized in a closed loop.

[0007] Furthermore, it also includes steps for comprehensive quantitative calculation of multi-dimensional spatial data association similarity and adaptive adjustment of matching thresholds. This is used to construct a comprehensive spatial data association similarity calculation model. Combining multi-dimensional parameters such as spatial geometric distance, topological relationship fit, semantic similarity, and temporal relevance, it calculates the comprehensive association similarity between spatial data units. Based on similarity distribution characteristics, it completes the adaptive dynamic adjustment of the association matching threshold. The expression for comprehensive spatial data association similarity calculation is as follows: ; This represents the comprehensive correlation similarity between two spatial data units, with a value ranging from 0 to 1. , , , These are the weight coefficients for spatial geometric distance, topological compatibility, semantic similarity, and temporal relevance, respectively, and the sum of all coefficients is 1. This represents the normalized numerical value of spatial geometric distance similarity. This represents the normalized numerical value of the topological fit. This is a normalized numerical value for semantic feature similarity. This represents the normalized values ​​for time-series correlation. The time decay coefficient, , These are the acquisition timestamps for the two spatial data units, The maximum effective time span for spatial data association. This represents the spatial topological deviation value between two spatial data units. This represents the maximum permissible threshold for spatial topological deviation.

[0008] Furthermore, it also includes dynamic scheduling and load balancing steps for the computing power and storage resources of the edge node cluster. This is used to collect the computing power utilization rate, storage remaining amount, network bandwidth, and data access frequency of the edge nodes in real time, construct an edge node resource load assessment system, and complete the dynamic scheduling of spatial data storage distribution, computing task allocation, and collaborative access links based on the assessment results, so as to achieve balanced resource allocation and load balancing of the edge node cluster.

[0009] Furthermore, the incremental learning process employs an edge-center collaborative incremental sample selection and model update mechanism. Based on incremental sample information entropy, feature differences, and correlation contribution, a sample value quantification evaluation model is constructed to automate the selection of high-value incremental samples. Through a collaborative mode of local incremental pre-training at edge nodes and global model aggregation and updating at center nodes, incremental updates of the spatial feature extraction network and the correlation matching model are achieved. The incremental sample value quantification evaluation expression is as follows: ; The score is the overall value rating for the incremental samples, ranging from 0 to 1. , , These are the weighting coefficients for information entropy, feature difference, and correlation contribution, respectively, and the sum of all coefficients is 1. The information entropy normalized value for incremental samples. This represents the normalized numerical value of the feature differences between the incremental sample and the existing feature library. This represents the normalized numerical value of the contribution of incremental samples to the association matching model. The influence coefficient of the model update frequency. This represents the number of historical incremental updates for the model. This is the normalized value of the association matching error rate corresponding to the incremental samples. This is the maximum allowable threshold for the association matching error rate.

[0010] Furthermore, in the standardization preprocessing of multi-source heterogeneous spatial data, an adaptive transformation of spatial coordinates and a spatiotemporal alignment mechanism for multi-source data are adopted to support the automatic identification and unified transformation of spatial coordinate systems. Based on spatiotemporal labels, the time and spatial dimensions of multi-source spatial data are accurately aligned. At the same time, differentiated preprocessing strategies are adopted for vector spatial data, raster remote sensing image data, and point trajectory data to complete the data format normalization and quality verification, and to remove abnormal data units that do not conform to spatial topological logic.

[0011] Furthermore, in the process of edge-collaborative spatial data hierarchical storage and distributed index construction, a hierarchical storage strategy is adopted, which involves local storage of hot data at the edge and archive storage of cold data in the center. Based on the access frequency, update frequency and geographical coverage of spatial data, the automatic division and dynamic migration of hot and cold data are completed. At the same time, a two-layer distributed spatial index structure based on Geohash encoding and R-tree is constructed. An R-tree index of local spatial data is built at the edge nodes, and a global distributed index based on Geohash encoding is built at the center nodes. An incremental update mechanism for the index is established, and the collaborative synchronization of the index between edge nodes is completed synchronously.

[0012] Furthermore, in the process of multi-dimensional spatial feature extraction, a hybrid deep learning architecture that integrates convolutional neural networks and graph neural networks is adopted. The spatial geometric and texture features of the raster spatial data are extracted through two-dimensional convolutional neural networks, the topological relationship features and semantic features of the vector spatial data are extracted through graph convolutional neural networks, and the temporal change features of the spatial data are extracted through temporal convolutional networks. The features extracted from multiple branches are fused and spliced ​​to generate a high-dimensional spatial feature vector of the same dimension, and the feature vector is normalized at the same time.

[0013] Furthermore, in the process of dynamic association matching and association graph construction of spatial data under spatiotemporal constraints, spatial data units are used as graph nodes, and the comprehensive association similarity between data units is used as edge weights to construct a directed weighted association graph with spatiotemporal attributes. At the same time, based on the update frequency of spatial data and the temporal changes of association relationships, the association graph is dynamically and incrementally updated, invalid association relationships are eliminated, valid association matching results are added, and a bidirectional mapping relationship between the association graph and the distributed spatial database is established, supporting fast retrieval of spatial data and traversal of association relationships based on the association graph.

[0014] Furthermore, in the edge-collaborative spatial data association retrieval process, a collaborative retrieval mechanism is adopted that combines the edge-distributed decomposition of retrieval tasks with the fusion of results at the center. Upon receiving a spatial data retrieval request, the retrieval task is decomposed to edge nodes with corresponding geographical coverage based on the spatial range, semantic conditions, and temporal constraints of the retrieval request. Each edge node completes local retrieval and matching of associated data based on local association maps and spatial indexes, and sends the retrieval results back to the central node to complete fusion sorting, duplicate data removal, and association relationship completion. Finally, a complete set of spatial data retrieval results and association relationships is output.

[0015] Furthermore, in the process of full-link verification and consistency closed-loop optimization of spatial data association results, a three-level verification system is established, which includes spatial topology consistency verification, semantic logic rationality verification, and spatiotemporal dimension correlation verification. Full-dimensional verification is performed on the association matching results, and an accuracy score for the association results is generated. Based on the score results and the abnormal associations identified during the verification process, the feature weights of the spatial feature extraction network, the dimensional parameters of the association matching model, and the structure of the distributed spatial index are optimized in reverse. At the same time, the collaborative access strategy between edge nodes is updated, forming a closed-loop optimization of the entire process of dynamic spatial data association.

[0016] Compared with existing technologies, the beneficial effects of this invention are: This invention achieves unified processing and precise spatiotemporal alignment of spatial data of different types, coordinate systems, and acquisition frequencies through a distributed acquisition and standardized preprocessing mechanism for multi-source heterogeneous spatial data. It can complete the format normalization and quality verification of multi-source spatial data, improve the standardization and quality of spatial data, and provide a unified and reliable data foundation for subsequent dynamic association and matching of spatial data.

[0017] This invention achieves distributed storage of spatial data and collaborative access between edge nodes through an edge-coordinated hot and cold tiered storage strategy and a two-layer distributed spatial index structure. It can make full use of the local storage resources of edge nodes, reduce cross-node data transmission overhead, and the two-layer distributed index structure can significantly improve the execution efficiency of spatial data retrieval and association matching, adapting to the spatial data access needs in wide-area distributed edge scenarios.

[0018] This invention achieves multi-dimensional deep extraction of spatial geometric features, topological relationship features, semantic features, and temporal features of spatial data by integrating a hybrid deep learning architecture of convolutional neural networks and graph neural networks. It can comprehensively capture the multi-dimensional attribute features of spatial data and generate high-dimensional feature vectors of unified dimensions, providing comprehensive feature support for spatial data association matching and improving the dimensional richness and accuracy of dynamic association matching of spatial data.

[0019] This invention enables incremental updates of spatial feature extraction networks and association matching models through an edge-center collaborative incremental learning framework. It can complete lightweight model updates based on high-value incremental samples, adapting to the association requirements of newly added spatial data without performing full retraining, significantly reducing the computational overhead of model updates, adapting to the limited computing resources of edge nodes, and improving the generalization ability and scenario adaptability of association models.

[0020] This invention achieves a structured representation and dynamic incremental update of spatial data relationships through a directed weighted dynamic association graph with spatiotemporal attributes. Combined with an edge collaborative retrieval mechanism, it can complete the collaborative association retrieval of spatial data across edge nodes, realize the fusion and optimization of multi-node retrieval results, and improve the comprehensiveness and response speed of spatial data retrieval.

[0021] This invention achieves full-dimensional verification of spatial data association results through a three-level verification system and a closed-loop optimization mechanism. Based on the verification results, it can reverse-optimize the spatial feature extraction network, association matching model, distributed index structure, and edge collaborative access strategy, forming a closed-loop optimization of the entire process of dynamic spatial data association. This continuously improves the accuracy, topological consistency, and spatiotemporal logic rationality of dynamic spatial data association, and can adapt to the application requirements of continuous incremental updates of spatial data in distributed edge scenarios for a long time. Attached Figure Description

[0022] Figure 1 Here is the overall flowchart of the method for dynamic association of spatial data; Figure 2 Logical diagram for standardized preprocessing and edge-layered storage of multi-source data; Figure 3 Flowchart for hybrid deep learning feature extraction and feature library construction; Figure 4 Flowchart for spatiotemporal constraint association matching and dynamic graph construction; Figure 5 A flowchart for collaborative incremental learning and closed-loop optimization iteration. Detailed Implementation

[0023] 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.

[0024] Reference Figures 1 to 5 A spatial data dynamic association method based on edge collaboration and incremental learning includes the following steps: The distributed acquisition and standardized preprocessing steps for multi-source heterogeneous spatial data are used to complete the distributed acquisition of multi-source heterogeneous spatial data such as geospatial vectors, remote sensing images, point trajectories, and attribute labels through edge node clusters. The standardized preprocessing steps include data format normalization, coordinate system unification, noise filtering, and missing value completion, generating standardized spatial data units with unique spatiotemporal labels. The steps of edge-collaborative spatial data hierarchical storage and distributed index construction are used to complete the cold and hot hierarchical distributed storage of spatial data based on the geographical coverage and computing resources of edge nodes, construct a two-layer distributed spatial index structure based on the spatiotemporal dimension, and establish a data synchronization and collaborative access mechanism between edge nodes. The steps of multi-dimensional spatial feature deep extraction and feature vector library construction are used to complete the multi-dimensional deep extraction of spatial geometric features, semantic features, temporal features, and topological relationship features of standardized spatial data based on a hybrid deep learning network, generate high-dimensional spatial feature vectors of unified dimension, and construct a distributed spatial feature vector library in collaboration between edge nodes and central nodes. The steps for dynamic association matching and association graph construction of spatial data under spatiotemporal constraints are used to complete multi-dimensional dynamic association matching of spatial data units based on spatial distance thresholds, topological relationship constraints, semantic similarity matching, and temporal correlation analysis, construct a weighted dynamic association graph of spatial data, and store it in a distributed spatial database. The dynamic update of the association model and the iterative optimization of the feature library based on incremental learning are used to collect new spatial data and association matching results. The incremental update of the spatial feature extraction network and the association matching model is completed through the edge-center collaborative incremental learning framework, and the iterative optimization of the distributed feature vector library and spatial index structure is completed simultaneously. The edge collaborative spatial data association retrieval and result fusion output steps are used to receive spatial data retrieval requests, complete cross-edge node collaborative association retrieval based on dynamic association graphs, and complete the fusion, sorting, deduplication and structured output of multi-node retrieval results. The spatial data association result end-to-end verification and consistency closed-loop optimization steps are used to complete the accuracy verification, topology consistency verification, and spatiotemporal logic verification of the association results. Based on the verification results, the association model, index structure, and collaborative access mechanism are optimized in a closed loop.

[0025] This invention also includes steps for comprehensive quantitative calculation of multi-dimensional spatial data association similarity and adaptive adjustment of matching thresholds. These steps are used to construct a comprehensive spatial data association similarity calculation model. By combining multi-dimensional parameters such as spatial geometric distance, topological relationship fit, semantic similarity, and temporal relevance, the comprehensive association similarity between spatial data units is calculated. Based on similarity distribution characteristics, the association matching threshold is adaptively and dynamically adjusted. The expression for the comprehensive spatial data association similarity calculation is as follows: ; This represents the comprehensive correlation similarity between two spatial data units, with a value ranging from 0 to 1, and is dimensionless. , , , These are the weight coefficients for spatial geometric distance, topological fit, semantic similarity, and temporal relevance, respectively. The sum of all coefficients is 1, and they are dimensionless. This represents the normalized numerical value of spatial geometric distance similarity, which is dimensionless. This is a dimensionless, normalized value for the topological fit. The normalized numerical value of semantic feature similarity is dimensionless. The time-series correlation normalized values ​​are dimensionless. The time decay coefficient is dimensionless. , These are the acquisition timestamps for the two spatial data units, in seconds. The maximum effective time span for spatial data association, in seconds. The spatial topological deviation between two spatial data units is dimensionless. The maximum permissible threshold for spatial topological deviation is dimensionless. Through multi-dimensional feature weighted fusion and dual constraints of time decay and topological deviation correction, it achieves end-to-end accurate quantification of the correlation similarity between spatial data units, providing an objective quantitative basis for dynamic correlation matching. At the same time, the adaptively adjusted weight coefficients can adapt to the correlation matching needs of different types of spatial data, improving the accuracy of correlation matching and scenario adaptability.

[0026] This invention also includes a dynamic scheduling and load balancing step for the computing power and storage resources of the edge node cluster. This step is used to collect the computing power utilization rate, remaining storage capacity, network bandwidth, and data access frequency of the edge nodes in real time, construct an edge node resource load assessment system, and complete the dynamic scheduling of spatial data storage distribution, computing task allocation, and collaborative access links based on the assessment results, thereby achieving balanced resource allocation and load balancing of the edge node cluster.

[0027] In this invention, an edge-center collaborative incremental sample selection and model update mechanism is adopted during incremental learning. A sample value quantification evaluation model is constructed based on incremental sample information entropy, feature differences, and correlation contribution, completing the automated selection of high-value incremental samples. Through a collaborative mode of local incremental pre-training at edge nodes and global model aggregation and updating at center nodes, incremental updates of the spatial feature extraction network and the correlation matching model are achieved. The incremental sample value quantification evaluation expression is as follows: ; This is a comprehensive value score for incremental samples, ranging from 0 to 1, and is dimensionless. , , These are the weighting coefficients for information entropy, feature difference, and correlation contribution, respectively. The sum of all coefficients is 1, and they are dimensionless. The information entropy of the incremental samples is a normalized value, dimensionless. This is a dimensionless, normalized numerical representation of the feature differences between the incremental sample and the existing feature library. This represents the normalized, dimensionless contribution of incremental samples to the association matching model. The frequency influence coefficient is used to update the model; it is dimensionless. This represents the number of historical incremental updates to the model, expressed in times. This is the normalized numerical value of the association matching error rate corresponding to the incremental samples, which is dimensionless. The maximum permissible threshold for association matching error rate is dimensionless. By weighting the sample information entropy, feature differences, and association contribution in multiple dimensions, combined with the constraints of model update frequency and error rate, the value of incremental samples can be accurately quantified and evaluated. This can effectively screen out high-value incremental samples for incremental model training, eliminate redundant and noisy samples, reduce the computational cost of incremental learning, and improve the association matching accuracy and generalization ability of the model after incremental update.

[0028] In the standardization preprocessing of multi-source heterogeneous spatial data in this invention, an adaptive transformation of spatial coordinates and a spatiotemporal alignment mechanism for multi-source data are adopted. It supports automatic identification and unified transformation of multiple spatial coordinate systems such as WGS84, CGCS2000, and UTM. Based on spatiotemporal labels, it completes the accurate alignment of the time and spatial dimensions of multi-source spatial data. At the same time, it adopts differentiated preprocessing strategies for vector spatial data, raster remote sensing image data, and point trajectory data to complete the data format normalization and quality verification, and remove abnormal data units that do not conform to spatial topological logic.

[0029] In the process of edge-collaborative spatial data hierarchical storage and distributed index construction, this invention adopts a hierarchical storage strategy of storing hot data locally at the edge and archiving cold data in the center. Based on the access frequency, update frequency and geographical coverage of spatial data, it completes the automatic division and dynamic migration of hot and cold data. At the same time, it constructs a two-layer distributed spatial index structure based on the combination of Geohash encoding and R-tree. The R-tree index of local spatial data is built at the edge nodes, and a global distributed index based on Geohash encoding is built at the center nodes. An incremental update mechanism for the index is established, and the collaborative synchronization of the index between edge nodes is completed synchronously.

[0030] In the process of multi-dimensional spatial feature deep extraction in this invention, a hybrid deep learning architecture that integrates convolutional neural networks and graph neural networks is adopted. The spatial geometric and texture features of raster spatial data are extracted by two-dimensional convolutional neural networks, the topological and semantic features of vector spatial data are extracted by graph convolutional neural networks, and the temporal variation features of spatial data are extracted by temporal convolutional networks. The features extracted from multiple branches are fused and spliced ​​to generate a high-dimensional spatial feature vector of the same dimension. At the same time, the feature vector is normalized to ensure the consistency of the feature vector matching in the distributed feature library.

[0031] In the process of dynamic association matching and association graph construction of spatial data under spatiotemporal constraints, this invention uses spatial data units as graph nodes and the comprehensive association similarity between data units as edge weights to construct a directed weighted association graph with spatiotemporal attributes. At the same time, based on the update frequency of spatial data and the temporal changes of association relationships, the association graph is dynamically and incrementally updated, invalid association relationships are eliminated, valid association matching results are added, and a bidirectional mapping relationship between the association graph and the distributed spatial database is established, supporting fast retrieval of spatial data and traversal of association relationships based on the association graph.

[0032] In the edge-collaborative spatial data association retrieval process of this invention, a collaborative retrieval mechanism of edge-distributed decomposition of retrieval tasks and result center fusion is adopted. After receiving a spatial data retrieval request, the retrieval task is decomposed to edge nodes with corresponding geographical coverage based on the spatial range, semantic conditions and temporal constraints of the retrieval request. Each edge node completes local retrieval and matching of associated data based on local association maps and spatial indexes, and sends the retrieval results back to the central node to complete fusion sorting, duplicate data removal and association relationship completion, and finally outputs a complete set of spatial data retrieval results and association relationships.

[0033] In the process of full-link verification and consistency closed-loop optimization of spatial data association results in this invention, a three-level verification system is established, which includes spatial topology consistency verification, semantic logic rationality verification, and spatiotemporal dimension correlation verification. Full-dimensional verification is performed on the association matching results, and an accuracy score of the association results is generated. Based on the score results and the abnormal association relationships identified during the verification process, the feature weights of the spatial feature extraction network, the dimensional parameters of the association matching model, and the structure of the distributed spatial index are optimized in reverse. At the same time, the collaborative access strategy between edge nodes is updated, forming a closed-loop optimization of the entire process of dynamic spatial data association.

[0034] Example 1

[0035] Implementation of Dynamic Spatial Data Association Based on Edge Collaboration and Incremental Learning in Wide-Area Natural Resource Monitoring Scenarios

[0036] This embodiment is applied to a provincial-level wide-area natural resource monitoring scenario. The scenario deploys 32 edge monitoring nodes across the province, covering five major geographical types: mountains, forests, water areas, cultivated land, and construction land. Each edge node corresponds to a fixed geographical monitoring grid and possesses local data acquisition, computing power processing, and storage capabilities. The central node is deployed in the provincial natural resource data center, responsible for global model management, data archiving, and collaborative scheduling. Spatial data sources within the scenario include five major categories: geospatial vector topographic data, multispectral remote sensing image data, vegetation sampling point location trajectory data, land use attribute label data, and hydrological monitoring spatial data. Data acquisition frequencies cover daily, weekly, and monthly levels, exhibiting characteristics of wide data distribution, heterogeneous data types, large differences in update frequency, and high centralized processing and transmission overhead.

[0037] In this embodiment, the distributed acquisition and standardized preprocessing of multi-source heterogeneous spatial data are first implemented. A cluster of 32 edge nodes completes the distributed acquisition of multi-source heterogeneous spatial data within the corresponding monitoring grid. Differentiated acquisition strategies are adopted for different data types. Edge nodes perform high-frequency acquisition of remote sensing imagery and point trajectory data daily, and update vector terrain and attribute label data weekly. The preprocessing stage employs an adaptive spatial coordinate transformation and multi-source data spatiotemporal alignment mechanism. It automatically identifies and performs unified transformations of multiple spatial coordinate systems, including WGS84, CGCS2000, and UTM. All data is mapped to the CGCS2000 national geodetic coordinate system. Unique spatiotemporal labels are generated based on the acquisition timestamp and geographic grid encoding, achieving precise alignment of the time and spatial dimensions of the multi-source data. Simultaneously, topology verification and format normalization are performed for vector spatial data; radiometric calibration, geometric correction, and noise filtering are performed for raster remote sensing imagery data; drift point removal and missing value completion are performed for point trajectory data; and structured mapping and outlier filtering are performed for attribute label data. Finally, standardized spatial data units with unique spatiotemporal labels are generated.

[0038] In the edge-collaborative spatial data hierarchical storage and distributed index construction steps, based on the geographical coverage and computing power and storage resource configuration of edge nodes, a hierarchical storage strategy is adopted, with hot data stored locally at the edge and cold data archived in the central data center. Automatic hot and cold data partitioning and dynamic migration are completed based on the access frequency, update frequency, and geographical coverage of spatial data. Monitoring data accessed frequently within the last three months is classified as hot data and stored locally on the corresponding edge nodes, while historical archived data older than one year is classified as cold data and stored on the central node. Simultaneously, a two-layer distributed spatial index structure based on Geohash encoding and R-trees is constructed. An R-tree index is built for local spatial data at each edge node to achieve fast spatial retrieval of local data. A global distributed index based on Geohash encoding is built at the central node, establishing a mapping relationship between the geographic grid and edge nodes. An incremental update mechanism for the index is set up, synchronously updating the corresponding index entries when data is added or changed. Furthermore, a high-frequency data synchronization mechanism between adjacent edge nodes and a cross-regional data collaborative access mechanism are established.

[0039] In the multi-dimensional spatial feature deep extraction and feature vector library construction steps, a hybrid deep learning architecture integrating convolutional neural networks and graph neural networks is adopted. The spatial geometric and texture features of raster remote sensing image data are extracted through two-dimensional convolutional neural networks, the topological relationship features and semantic features of vector terrain and land use data are extracted through graph convolutional neural networks, and the temporal change features of vegetation points and land use change data are extracted through temporal convolutional neural networks. The features extracted from multiple branches are fused and stitched together to generate a unified 256-dimensional high-dimensional spatial feature vector. The feature vector is normalized to ensure the consistency of matching of feature vectors in the distributed feature library. Finally, a distributed spatial feature vector library is constructed that coordinates the local feature library of edge nodes and the global feature library of central nodes.

[0040] In the process of dynamic association matching and association graph construction of spatial data under spatiotemporal constraints, multi-dimensional dynamic association matching of spatial data units is completed based on spatial distance thresholds, topological relationship constraints, semantic similarity matching, and temporal correlation analysis. Using spatial data units as graph nodes and the comprehensive association similarity between data units as edge weights, a directed weighted association graph with spatiotemporal attributes is constructed. Dynamic incremental updates of the association graph are completed based on the update frequency of spatial data. A full graph update is performed monthly, and an incremental update is performed weekly. Invalid association relationships are removed, and valid association matching results are added. A bidirectional mapping relationship between the association graph and the distributed spatial database is established, supporting fast retrieval of spatial data and traversal of association relationships based on the association graph.

[0041] In the dynamic update of the association model and iterative optimization of the feature library based on incremental learning, an edge-center collaborative incremental sample screening and model update mechanism is adopted. Each edge node collects new spatial data and association matching results. Based on the incremental sample information entropy, feature differences, and association contribution, high-value incremental samples are automatically screened. Through the collaborative mode of local incremental pre-training at edge nodes and global model aggregation and update at center nodes, the spatial feature extraction network and association matching model are updated incrementally, and the distributed feature vector library and spatial index structure are iteratively optimized simultaneously.

[0042] In the edge-collaborative spatial data association retrieval and result fusion output step, a collaborative retrieval mechanism of edge-distributed decomposition of retrieval tasks and result center fusion is adopted. After receiving a spatial data retrieval request, the retrieval task is decomposed to edge nodes with corresponding geographical coverage based on the spatial range, semantic conditions and temporal constraints of the retrieval request. Each edge node completes local retrieval and matching of associated data based on local association maps and spatial indexes, and sends the retrieval results back to the central node to complete fusion sorting, duplicate data removal and association relationship completion. Finally, a complete set of spatial data retrieval results and association relationships is output.

[0043] In the process of full-link verification and consistency closed-loop optimization of spatial data association results, a three-level verification system is established, which includes spatial topology consistency verification, semantic logic rationality verification, and spatiotemporal dimension correlation verification. Full-dimensional verification is performed on the association matching results, and an accuracy score for the association results is generated. Based on the score results and the abnormal associations identified during the verification process, the feature weights of the spatial feature extraction network, the dimensional parameters of the association matching model, and the structure of the distributed spatial index are optimized in reverse. At the same time, the collaborative access strategy between edge nodes is updated, forming a closed-loop optimization of the entire process of dynamic spatial data association.

[0044] Table 1. Comparison of core performance between the method of this invention and traditional spatial data processing methods.

[0045] Table 1 shows the data from the statistical results of three consecutive months of actual operation in this embodiment. Traditional centralized spatial data processing methods require the full amount of edge data to be transmitted back to the central node for processing and correlation analysis. This results in high retrieval response latency, large data transmission bandwidth consumption, and low utilization of edge resources, making them unsuitable for the real-time processing needs of wide-area distributed edge scenarios. Traditional single-dimensional spatial correlation methods only perform correlation matching based on spatial geometric distance, failing to fully integrate multi-dimensional spatial features, resulting in limited correlation matching accuracy. Furthermore, model updates require full retraining, leading to high computational overhead. The method of this invention, through an edge-cooperative distributed architecture, fully utilizes the local resources of edge nodes, significantly reducing data transmission overhead and retrieval response latency. At the same time, the multi-dimensional feature fusion correlation matching mechanism significantly improves the accuracy of correlation results, and the edge-center collaborative incremental learning mode greatly reduces the computational overhead of model updates, making it fully adaptable to the dynamic spatial data correlation needs of wide-area natural resource monitoring scenarios.

[0046] Example 2

[0047] Implementation of Dynamic Spatial Data Association Based on Edge Collaboration and Incremental Learning in Urban Traffic Network Spatiotemporal Data Scenarios

[0048] This embodiment is applied to a smart traffic management scenario in a provincial capital city. The city has deployed edge control nodes in eight main urban areas, covering the city's core road network, expressways, main and secondary roads, and branch roads of all levels. Each edge node corresponds to the geographical area of ​​the main urban area and has the capability for local traffic data collection, real-time processing, and storage. The central node is deployed in the city's traffic management center, responsible for global traffic model management, data archiving, and cross-regional collaborative scheduling. Spatial data sources within the scenario include five major categories: urban road network vector data, traffic checkpoint remote sensing image data, vehicle location trajectory data, traffic facility attribute label data, and traffic event spatial data. Data collection frequencies cover three levels: second-level, minute-level, and hourly-level. The scenario is characterized by high real-time requirements, rapid update frequency, strong heterogeneity of multi-source data, and frequent cross-regional correlation needs.

[0049] In this embodiment, the distributed acquisition and standardized preprocessing of multi-source heterogeneous spatial data are first implemented. This involves the distributed acquisition of multi-source heterogeneous spatial data within the corresponding area through a cluster of eight edge nodes in the main urban area. Differentiated acquisition strategies are adopted for different data types. Edge nodes complete vehicle location trajectory data acquisition within seconds, traffic checkpoint remote sensing image data acquisition within minutes, and road network vector and traffic facility attribute data update acquisition within hours. The preprocessing stage employs an adaptive spatial coordinate transformation and multi-source data spatiotemporal alignment mechanism. It automatically identifies the spatial coordinate systems of data from different sources and performs a unified transformation. All data is mapped to a dual coordinate system of the city's independent plane coordinate system and the CGCS2000 national geodetic coordinate system. Unique spatiotemporal labels are generated based on the acquisition timestamp and the city's road network grid code, achieving microsecond-level precise alignment of the time and spatial dimensions of the multi-source data. Simultaneously, topological connectivity verification and format normalization are performed on vector road network data; image dehazing, geometric correction, and noise filtering are performed on raster remote sensing image data; map matching, drift point removal, and missing value completion are performed on vehicle location trajectory data; and structured mapping and outlier filtering are performed on traffic facility attribute data, ultimately generating standardized spatial data units with unique spatiotemporal labels.

[0050] In the edge-collaborative spatial data hierarchical storage and distributed index construction steps, based on the geographical coverage and computing power and storage resource configuration of edge nodes, a hierarchical storage strategy is adopted, with hot data stored locally at the edge and cold data archived in the central data center. Automatic hot and cold data partitioning and dynamic migration are completed based on the access frequency, update frequency, and geographical coverage of spatial data. Real-time traffic data within 72 hours is classified as hot data and stored locally on the corresponding edge nodes, while historical traffic archived data exceeding 6 months is classified as cold data and stored on the central node. Simultaneously, a two-layer distributed spatial index structure based on Geohash encoding and R-trees is constructed. An R-tree index for local traffic spatial data is built at each edge node, enabling millisecond-level spatial retrieval of local data. A global distributed index based on urban road network Geohash encoding is built at the central node, establishing a mapping relationship between the road network grid and edge nodes. A real-time incremental update mechanism for the index is set up, synchronously updating the corresponding index entries when data is added or changed. Furthermore, a real-time traffic data synchronization mechanism between adjacent edge nodes and a cross-regional traffic data collaborative access mechanism are established.

[0051] In the multi-dimensional spatial feature deep extraction and feature vector library construction steps, a hybrid deep learning architecture integrating convolutional neural networks and graph neural networks is adopted. Two-dimensional convolutional neural networks are used to extract spatial geometric features and road texture features from traffic checkpoint remote sensing images. Graph convolutional neural networks are used to extract topological connectivity features and traffic semantic features from road network vector data. Temporal convolutional neural networks are used to extract temporal variation features from vehicle trajectory and traffic flow data. The features extracted from multiple branches are fused and spliced ​​to generate a unified 256-dimensional high-dimensional spatial feature vector. Normalization processing is performed on the feature vector to ensure the consistency of feature vector matching in the distributed feature library. Finally, a distributed spatial feature vector library is constructed that coordinates the local feature library of edge nodes and the global feature library of central nodes.

[0052] In the steps of dynamic association matching and association graph construction of spatial data under spatiotemporal constraints, multi-dimensional dynamic association matching of traffic spatial data units is completed based on spatial distance thresholds, topological relationship constraints, semantic similarity matching, and temporal correlation analysis. Using traffic spatial data units as graph nodes and the comprehensive association similarity between data units as edge weights, a directed weighted traffic association graph with spatiotemporal attributes is constructed. Dynamic incremental updates of the association graph are completed based on the real-time update frequency of traffic data. A full graph update is performed every hour, and an incremental update is performed every minute. Invalid associations are removed, and valid association matching results are added. A bidirectional mapping relationship between the association graph and the distributed traffic spatial database is established, supporting fast retrieval of traffic spatial data and traversal of associations based on the association graph.

[0053] In the dynamic update of the association model and iterative optimization of the feature library based on incremental learning, an edge-center collaborative incremental sample screening and model update mechanism is adopted. Each edge node collects new traffic spatial data and association matching results. Based on the incremental sample information entropy, feature differences, and association contribution, high-value incremental samples are automatically screened. Through the collaborative mode of local incremental pre-training at edge nodes and global model aggregation and update at center nodes, the incremental update of the spatial feature extraction network and the traffic association matching model is completed, and the iterative optimization of the distributed feature vector library and spatial index structure is completed simultaneously.

[0054] In the edge-collaborative spatial data association retrieval and result fusion output step, a collaborative retrieval mechanism of edge-distributed decomposition of retrieval tasks and result center fusion is adopted. After receiving a traffic spatial data retrieval request, the retrieval task is decomposed to edge nodes with corresponding geographical coverage based on the spatial range, semantic conditions, and temporal constraints of the retrieval request. Each edge node completes local retrieval and related data matching based on local association maps and spatial indexes, and sends the retrieval results back to the central node to complete fusion sorting, duplicate data removal, and relationship completion. Finally, a complete set of traffic spatial data retrieval results and related relationships is output.

[0055] In the process of full-link verification and consistency closed-loop optimization of spatial data association results, a three-level verification system is established, which includes spatial topology consistency verification, semantic logic rationality verification, and spatiotemporal dimension correlation verification. Full-dimensional verification is performed on traffic association matching results, and an accuracy score for the association results is generated. Based on the score results and the abnormal associations identified during the verification process, the feature weights of the spatial feature extraction network, the dimensional parameters of the traffic association matching model, and the structure of the distributed spatial index are optimized in reverse. At the same time, the collaborative access strategy between edge nodes is updated, forming a closed-loop optimization of the entire process of dynamic association of traffic spatial data.

[0056] Table 2. Statistical analysis of the association and matching effects of the method of the present invention on different types of spatial data.

[0057] Table 2 shows that the method of this invention can achieve high accuracy and recall rates for association matching for different types of spatial data. Among them, the matching effect of road network vector data and point trajectory data is the best, which is due to the fact that the hybrid deep learning architecture can fully capture the unique features of different types of data and realize the comprehensive extraction and fusion of multi-dimensional features. At the same time, compared with the traditional centralized retrieval mode, the edge collaborative retrieval mechanism of this invention has achieved a significant improvement in retrieval response speed for various types of spatial data. It can fully adapt to the real-time requirements of data retrieval in urban traffic scenarios, and can comprehensively cover the dynamic association needs of multiple types of spatial data in traffic scenarios, with good scenario adaptability and data compatibility.

[0058] Reference Figure 1 This diagram illustrates the complete business loop of the system from raw data input to final consistency optimization. The process encompasses the collection and standardization of multi-source heterogeneous data, hierarchical storage with edge collaboration, deep feature extraction, dynamic association graph construction, incremental learning-driven model iteration, collaborative retrieval output, and final closed-loop verification. This structure embodies the core technological logic of edge awareness, central aggregation, and continuous evolution.

[0059] Reference Figure 2 This diagram highlights the initial processing and storage allocation mechanisms for data at the edge. The system supports multi-source data such as vector, image, and trajectory data, achieving standardization through coordinate transformation and spatiotemporal alignment. At the storage level, the system implements a strategy of storing hot data at the edge and archiving cold data in the data center based on access frequency, and combines Geohash and R-trees to build a two-layer distributed index, ensuring high efficiency for cross-node retrieval.

[0060] Reference Figure 3This figure illustrates how the system extracts high-dimensional features from different types of spatial data. It uses CNN to process the texture of raster images, GCN to process the topological semantics of vector data, and TCN to process temporal variations, ultimately fusing and stitching them together to generate feature vectors of uniform dimension. These vectors are stored in a feature library that combines edges and centers, providing a standardized comparison basis for subsequent similarity calculations.

[0061] Reference Figure 4 This diagram illustrates how spatial data units establish connections. The system calculates connections based on four dimensions: spatial distance, topological relationship, semantic similarity, and temporal relevance. By introducing a time decay coefficient and topological deviation correction, a precise comprehensive association similarity is calculated. Units that meet the similarity criteria are used as nodes to construct a weighted association graph, thereby realizing dynamic connections between geographic entities.

[0062] Reference Figure 5 This figure illustrates the system's self-evolution capability. The system filters high-value incremental samples based on information entropy and correlation contribution, and achieves incremental updates of the algorithm through pre-training at edge nodes and aggregation with the central node model. Simultaneously, the correlation results undergo a three-level verification system (topological, semantic, and spatiotemporal logic), and the verification results are fed back to the system for reverse optimization of the index structure, collaborative mechanisms, and feature weights, forming a continuously evolving closed loop.

[0063] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A spatial data dynamic association method based on edge collaboration and incremental learning, characterized in that, Includes the following steps: The distributed acquisition and standardized preprocessing steps for multi-source heterogeneous spatial data are used to complete the distributed acquisition of multi-source heterogeneous spatial data such as geospatial vectors, remote sensing images, point trajectories, and attribute labels through edge node clusters. The standardized preprocessing steps include data format normalization, coordinate system unification, noise filtering, and missing value completion, generating standardized spatial data units with unique spatiotemporal labels. The steps of edge-collaborative spatial data hierarchical storage and distributed index construction are used to complete the cold and hot hierarchical distributed storage of spatial data based on the geographical coverage and computing resources of edge nodes, construct a two-layer distributed spatial index structure based on the spatiotemporal dimension, and establish a data synchronization and collaborative access mechanism between edge nodes. The steps of multi-dimensional spatial feature deep extraction and feature vector library construction are used to complete the multi-dimensional deep extraction of spatial geometric features, semantic features, temporal features, and topological relationship features of standardized spatial data based on a hybrid deep learning network, generate high-dimensional spatial feature vectors of unified dimension, and construct a distributed spatial feature vector library in collaboration between edge nodes and central nodes. The steps for dynamic association matching and association graph construction of spatial data under spatiotemporal constraints are used to complete multi-dimensional dynamic association matching of spatial data units based on spatial distance thresholds, topological relationship constraints, semantic similarity matching, and temporal correlation analysis, construct a weighted dynamic association graph of spatial data, and store it in a distributed spatial database. The dynamic update of the association model and the iterative optimization of the feature library based on incremental learning are used to collect new spatial data and association matching results. The incremental update of the spatial feature extraction network and the association matching model is completed through the edge-center collaborative incremental learning framework, and the iterative optimization of the distributed feature vector library and spatial index structure is completed simultaneously. The edge collaborative spatial data association retrieval and result fusion output steps are used to receive spatial data retrieval requests, complete cross-edge node collaborative association retrieval based on dynamic association graphs, and complete the fusion, sorting, deduplication and structured output of multi-node retrieval results. The spatial data association result end-to-end verification and consistency closed-loop optimization steps are used to complete the accuracy verification, topology consistency verification, and spatiotemporal logic verification of the association results. Based on the verification results, the association model, index structure, and collaborative access mechanism are optimized in a closed loop.

2. The spatial data dynamic association method based on edge collaboration and incremental learning according to claim 1, characterized in that, It also includes steps for comprehensive quantitative calculation of multi-dimensional spatial data association similarity and adaptive adjustment of matching thresholds. These steps are used to construct a comprehensive spatial data association similarity calculation model. Combining multi-dimensional parameters such as spatial geometric distance, topological relationship fit, semantic similarity, and temporal relevance, the model calculates the comprehensive association similarity between spatial data units. Based on similarity distribution characteristics, it performs adaptive dynamic adjustment of the association matching threshold. The expression for the comprehensive spatial data association similarity calculation is as follows: ； This represents the comprehensive correlation similarity between two spatial data units, with a value ranging from 0 to 1. , , , These are the weight coefficients for spatial geometric distance, topological compatibility, semantic similarity, and temporal relevance, respectively, and the sum of all coefficients is 1. This represents the normalized numerical value of spatial geometric distance similarity. This represents the normalized numerical value of the topological fit. This is a normalized numerical value for semantic feature similarity. This represents the normalized values ​​for time-series correlation. The time decay coefficient, , These are the acquisition timestamps for the two spatial data units, The maximum effective time span for spatial data association. This represents the spatial topological deviation value between two spatial data units. This represents the maximum permissible threshold for spatial topological deviation.

3. The spatial data dynamic association method based on edge collaboration and incremental learning according to claim 1, characterized in that, It also includes dynamic scheduling and load balancing steps for computing power and storage resources of edge node clusters. This is used to collect real-time operating parameters such as computing power utilization, remaining storage, network bandwidth, and data access frequency of edge nodes, construct an edge node resource load assessment system, and complete the dynamic scheduling of spatial data storage distribution, computing task allocation, and collaborative access links based on the assessment results, so as to achieve balanced resource allocation and load balancing of edge node clusters.

4. The spatial data dynamic association method based on edge collaboration and incremental learning according to claim 1, characterized in that, Incremental learning employs an edge-center collaborative incremental sample selection and model update mechanism. Based on incremental sample information entropy, feature differences, and correlation contribution, a sample value quantification evaluation model is constructed to automate the selection of high-value incremental samples. Through a collaborative model of local incremental pre-training at edge nodes and global model aggregation and updating at center nodes, incremental updates of the spatial feature extraction network and the correlation matching model are achieved. The incremental sample value quantification evaluation expression is as follows: ； The score is the overall value rating for the incremental samples, ranging from 0 to 1. , , These are the weighting coefficients for information entropy, feature difference, and correlation contribution, respectively, and the sum of all coefficients is 1. The information entropy normalized value for incremental samples. This represents the normalized numerical value of the feature differences between the incremental sample and the existing feature library. This represents the normalized numerical value of the contribution of incremental samples to the association matching model. The influence coefficient of the model update frequency. This represents the number of historical incremental updates for the model. This is the normalized value of the association matching error rate corresponding to the incremental samples. This is the maximum allowable threshold for the association matching error rate.

5. The spatial data dynamic association method based on edge collaboration and incremental learning according to claim 1, characterized in that, In the standardization preprocessing of multi-source heterogeneous spatial data, an adaptive transformation of spatial coordinates and a spatiotemporal alignment mechanism for multi-source data are adopted to support automatic identification and unified transformation of spatial coordinate systems. Based on spatiotemporal labels, the time and spatial dimensions of multi-source spatial data are accurately aligned. At the same time, differentiated preprocessing strategies are adopted for vector spatial data, raster remote sensing image data, and point trajectory data to complete the data format normalization and quality verification, and to remove abnormal data units that do not conform to spatial topological logic.

6. The spatial data dynamic association method based on edge collaboration and incremental learning according to claim 1, characterized in that, In the process of edge-collaborative spatial data hierarchical storage and distributed index construction, a hierarchical storage strategy is adopted, which uses hot data for local storage at the edge and cold data for archive storage in the data center. Based on the access frequency, update frequency and geographical coverage of spatial data, the automatic division and dynamic migration of hot and cold data are completed. At the same time, a two-layer distributed spatial index structure based on Geohash encoding and R-tree is constructed. An R-tree index of local spatial data is built at the edge nodes and a global distributed index based on Geohash encoding is built at the central nodes. An incremental update mechanism for the index is established to synchronously complete the collaborative synchronization of the index between edge nodes.

7. The spatial data dynamic association method based on edge collaboration and incremental learning according to claim 1, characterized in that, In the process of multi-dimensional spatial feature extraction, a hybrid deep learning architecture that integrates convolutional neural networks and graph neural networks is adopted. The spatial geometric and texture features of raster spatial data are extracted through two-dimensional convolutional neural networks, the topological and semantic features of vector spatial data are extracted through graph convolutional neural networks, and the temporal variation features of spatial data are extracted through temporal convolutional networks. The features extracted from multiple branches are fused and spliced ​​to generate a high-dimensional spatial feature vector of the same dimension, and the feature vector is normalized at the same time.

8. The spatial data dynamic association method based on edge collaboration and incremental learning according to claim 1, characterized in that, In the process of dynamic association matching and association graph construction of spatial data under spatiotemporal constraints, spatial data units are used as graph nodes, and the comprehensive association similarity between data units is used as edge weights to construct a directed weighted association graph with spatiotemporal attributes. At the same time, based on the update frequency of spatial data and the temporal changes of association relationships, the association graph is dynamically and incrementally updated, invalid association relationships are removed, and valid association matching results are added. A bidirectional mapping relationship between the association graph and the distributed spatial database is established, supporting fast retrieval of spatial data and traversal of association relationships based on the association graph.

9. The spatial data dynamic association method based on edge collaboration and incremental learning according to claim 1, characterized in that, In the edge-collaborative spatial data association retrieval process, a collaborative retrieval mechanism is adopted, which involves the distributed decomposition of retrieval tasks at the edge and the fusion of results at the center. After receiving a spatial data retrieval request, the retrieval task is decomposed to edge nodes with corresponding geographical coverage based on the spatial range, semantic conditions, and temporal constraints of the retrieval request. Each edge node completes local retrieval and matching of associated data based on local association maps and spatial indexes, and sends the retrieval results back to the central node to complete fusion sorting, duplicate data removal, and association relationship completion. Finally, a complete set of spatial data retrieval results and association relationships is output.

10. The spatial data dynamic association method based on edge collaboration and incremental learning according to claim 1, characterized in that, In the process of full-link verification and consistency closed-loop optimization of spatial data association results, a three-level verification system is established, which includes spatial topology consistency verification, semantic logic rationality verification, and spatiotemporal dimension correlation verification. Full-dimensional verification is performed on the association matching results, and an accuracy score of the association results is generated. Based on the score results and the abnormal associations identified during the verification process, the feature weights of the spatial feature extraction network, the dimensional parameters of the association matching model, and the structure of the distributed spatial index are optimized in reverse. At the same time, the collaborative access strategy between edge nodes is updated, forming a closed-loop optimization of the entire process of dynamic spatial data association.

Images