GIS data dynamic updating optimization algorithm based on three-dimensional digital twinning

Abstract

The invention discloses a GIS (Geographic Information System) data dynamic updating optimization algorithm based on three-dimensional digital twinning, which comprises the following steps of: S1, acquiring GIS data, preprocessing and generating a three-dimensional observation data set with a unified structure; s2, dividing a target area by adopting an octree structure, and constructing a three-dimensional digital twin model and a spatial index system; s3, constructing a spatial topological graph model based on the spatial adjacency relation and the semantic tag; s4, according to a topological graph node structure, constructing an improved hierarchical Bayesian network model, calculating an update posterior probability, and generating an update priority sequence; s5, performing difference comparison based on the updating priority sequence, and executing incremental updating; and S6, the prior distribution of the Bayesian network nodes is dynamically corrected, the topological graph updating probability is adjusted, and loop optimization is carried out until convergence. According to the invention, real-time, efficient and accurate updating of the GIS data is realized.

Description

Technical Field

[0001] The present invention relates to the field of geographic information technology, and particularly to a GIS data dynamic update and optimization algorithm based on three-dimensional digital twin. Background Art

[0002] With the rapid development of digital twin technology and three-dimensional geographic information system (GIS), GIS data based on three-dimensional digital twin has been widely applied in fields such as smart city, emergency management, and urban spatial planning. By integrating remote sensing images, lidar, and edge sensor data, refined management and real-time monitoring of the urban environment are achieved. However, in the actual application process, there are significant differences in the geographical features, spatial structures, and dynamic changes of urban areas, posing higher requirements for the real-time performance, accuracy, and dynamic update efficiency of GIS data, which brings huge challenges to traditional GIS data update methods.

[0003] Currently, most GIS data update methods still rely on manual periodic collection and unified batch update. These traditional methods usually require separate data collection, processing, comparison, and import for each region. Although this method can improve the accuracy of data periodically, obvious technical defects are also exposed: First, the manual batch update process requires a large amount of human resources and time costs, with low efficiency and difficult to meet the needs of rapidly developing urban management. Second, most existing update methods lack a unified fusion mechanism for multi-source heterogeneous data. The format differences and spatial scale inconsistencies of remote sensing images, lidar, and sensor data will lead to data redundancy and fusion difficulties, and cannot effectively improve the refinement and accuracy of updates. In addition, traditional methods lack refined partitioning means for representing complex three-dimensional spatial structures and establishing spatial indexes. Usually, only simple two-dimensional or single-layer spatial index structures are adopted, resulting in low spatial retrieval efficiency and lack of optimization in update path selection, seriously affecting the efficiency and real-time performance of GIS data updates.

[0004] In recent years, incremental update and automated difference detection technologies have gradually been introduced into the field of 3D GIS in order to improve the efficiency of GIS data update. By constructing a spatial topology graph or using a probability model to detect data changes, these methods have alleviated the problem of low efficiency of traditional update methods to a certain extent. However, the existing incremental update methods still have the following deficiencies: on the one hand, in the process of constructing nodes and edges of the current spatial topology graph structure, the fine-grained expression ability of spatial units is limited, ignoring the comprehensive influence of structural differences, attribute features, and semantic similarity between spatial units, making it difficult to achieve efficient update decisions; on the other hand, most of the existing update strategies based on probability inference models such as Bayesian networks adopt static or simple parameter configurations, without fully considering the changes in the states of spatial units and the influence of dynamic interactions with adjacent regions during the long-term update process, resulting in the lack of dynamic adaptability of update decisions, poor stability in long-term operation, and the inability to effectively avoid the problems of redundant updates and update delays in important regions.

[0005] In summary, the existing technologies still have significant deficiencies in the dynamic update efficiency of 3D GIS data, the construction of refined spatial indexes, and the dynamic optimization of update strategies, making it difficult to meet the urgent needs of smart cities for real-time and accurate spatial data.

[0006] Therefore, how to provide a GIS data dynamic update optimization algorithm based on 3D digital twins is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0007] An object of the present invention is to propose a GIS data dynamic update optimization algorithm based on 3D digital twins. The present invention makes full use of octree space division, spatial topology graph modeling, and hierarchical Bayesian network inference technologies, and details a GIS data real-time update algorithm that can dynamically adapt to changes in the urban 3D spatial structure, with the advantages of high data update real-time performance, high spatial retrieval efficiency, strong update path optimization ability, and good dynamic adaptability.

[0008] A GIS data dynamic update optimization algorithm based on 3D digital twins according to an embodiment of the present invention includes the following steps:

[0009] S1. Collect GIS data, perform preprocessing, calculate the change frequency based on multi-temporal observation results, determine the data coverage range as the target area, and generate a three-dimensional observation data set with a unified structure;

[0010] S2. In the target area, use the octree structure to divide the space into multi-level spatial units, use each spatial unit to construct a 3D digital twin model, and establish a three-dimensional spatial index system based on the level and coordinate position of the spatial unit;

[0011] S3. Map each spatial unit to a node in the graph model, establish edge connections based on the spatial adjacency relationship and semantic labels, and construct a spatial topology graph model;

[0012] S4. According to the structure and connection relationship of the nodes in the spatial topology graph model, construct an improved hierarchical Bayesian network model, input the change frequency, structural complexity, and attribute characteristics of each node in the three-dimensional observation dataset, calculate and update the posterior probability, and generate an update priority sequence;

[0013] S5. Select nodes from the spatial topology graph model in turn according to the update priority sequence, perform differential comparison in combination with the three-dimensional observation dataset, identify the spatial regions where the structure or attributes have changed, and perform incremental updates;

[0014] S6. After each round of update, correct the prior distribution of the updated nodes in the improved hierarchical Bayesian network model, dynamically adjust the update probability of adjacent nodes in the spatial topology graph model, and repeat until the termination condition is met.

[0015] Optionally, the GIS data includes remote sensing images, lidar, and edge sensor data. The preprocessing includes format standardization, time synchronization, and spatial registration, extracts spatial structures, attribute characteristics, and semantic information, and the semantic labels are generated from the extracted semantic information.

[0016] Optionally, the specific steps of S2 are as follows:

[0017] S21. Divide the target area into cubic units with a fixed side length in the three-dimensional coordinate space to generate an initial spatial division structure;

[0018] S22. Recursively divide the initial spatial units using an octree structure. Each parent unit is divided into two sub-units in the horizontal axis, vertical axis, and elevation axis directions to obtain eight sub-spatial units. Set the maximum division level as L, and stop dividing when the current level reaches L;

[0019] S23. Calculate the Morton code for the spatial units at the l-th layer:

[0020]

[0021] where, I l is the index code of the spatial unit at the l-th layer, a k is the discrete coordinate of the spatial unit at the k-th layer in the horizontal axis direction, b k is the discrete coordinate of the spatial unit at the k-th layer in the vertical axis direction, c k is the discrete coordinate of the spatial unit at the k-th layer in the elevation axis direction, << is the bitwise left shift operator, k is the layer number of the current summation term, l is the layer number where the current spatial unit is located, and L is the maximum division level of the octree structure;

[0022] S24. Construct the index codes of all levels into an index tree structure, establish the mapping relationship from three-dimensional space coordinates to the Morton coding path, form a hierarchical three-dimensional space index system that can be recursively queried, and store it in the space index table;

[0023] S25. Establish a one-to-one mapping between each index item in the space index table and the corresponding space unit in the three-dimensional digital twin model.

[0024] Optionally, the specific steps of S3 are as follows:

[0025] S31. Map the space units generated by the octree structure division to the nodes in the space topology graph model in sequence;

[0026] S32. Based on the spatial adjacency relationship, establish an undirected edge connection relationship between the nodes, and add edge connections to the space unit nodes with shared faces or shared edges;

[0027] S33. Introduce attribute similarity and structural adjacency degree as the graph edge weight factors, and assign a weight value w to each edge in the graph structure ij :

[0028] w ij = α·d ij + β·r ij + γ·s ij ;

[0029] where w ij is the edge weight between node i and node j, d ij is the spatial distance between node i and node j, r ij is the historical change correlation degree between node i and node j, s ij is the semantic similarity degree between node i and node j, and α, β, γ are weighting coefficients;

[0030] S34. Structurally store the constructed space topology graph model in the form of an adjacency matrix for modeling and calculation.

[0031] Optionally, the specific steps of S4 are as follows:

[0032] S41. Construct an improved hierarchical Bayesian network model, set three types of nodes: the state layer, the region layer, and the unit layer. The state layer corresponds to the global control variables, the region layer corresponds to the subgraph structure in the space topology graph model, and the unit layer corresponds to the space unit nodes. Establish directed connections based on the space index relationship and the semantic label relationship;

[0033] S42. Extract the change frequency, structural complexity, semantic coding, and adjacency perturbation degree of each space unit, and form a feature vector after normalization processing;

[0034] S43. Calculate the updated posterior probability of each spatial unit:

[0035]

[0036] where P i is the updated posterior probability of spatial unit i, X i is the normalized feature vector of spatial unit i, R i is the state variable of spatial unit i, P(X i |R i = 1) is the conditional probability density of the feature vector under the condition that the state is updated, P(R i = 1) is the prior probability that the state is updated, and P(X i ) is the marginal probability density of the feature vector;

[0037] S44. Calculate the marginal probability density using Bayes' theorem and the feature distribution model:

[0038] P(X i ) = P(X i |R i = 1)·P(R i = 1) + P(X i |R i = 0)·P(R i = 0);

[0039] where P(X i |R i = 0) is the conditional probability density of the feature vector under the condition that the state is not updated, and P(R i = 0) is the prior probability that the state is not updated;

[0040] S45. Sort the updated posterior probabilities corresponding to all spatial units in descending order of value to generate an update priority sequence.

[0041] Optionally, S44 specifically includes:

[0042] S441. Classify the normalized feature vectors of each spatial unit node according to the state variable, and construct two sample sets with the states of updated and not updated;

[0043] S442. Calculate the feature mean vector and covariance matrix for each type of sample set;

[0044] S443. To improve the numerical stability of the model, introduce a regularization term to adjust the covariance matrix and construct a modified covariance matrix

[0045]

[0046] Among them, is the corrected covariance matrix, Σ (r) is the characteristic covariance matrix when the state is r, ∈ is a positive real constant, and I is the identity matrix;

[0047] S444. Based on the corrected covariance matrix and the mean vector, construct a conditional probability density function:

[0048]

[0049] Among them, P(X i |R i =r) is the characteristic joint probability density of node i under the condition that the state is r, X i is the characteristic vector of node i, μ (r) is the characteristic mean vector when the state is r, is the corrected covariance matrix, Σ (r) is the characteristic covariance matrix when the state is r, is the vector transpose, (·) -1 is the matrix inversion, |·| is the matrix determinant, exp is the natural exponential function, π is the constant of pi, and d is the dimension of the characteristic vector;

[0050] S445. Apply the conditional probability density function to the posterior probability calculation process of S43 in claim 5 to generate an updated priority sequence.

[0051] Optionally, the S5 specifically includes:

[0052] S51. Construct an initial candidate node set, set the maximum number of jump rounds T max , starting from the spatial unit node with the largest updated probability value, initialize the current node and the jump round counter;

[0053] S52. Construct an adjacent candidate node set at the current node, and calculate the sampling acceptance probability for any candidate node j:

[0054]

[0055] Among them, α i→j is the acceptance probability of jumping from node i to node j, P i , P j are the updated probabilities of nodes i and j, β is the non-linear adjustment coefficient, γ is the energy adjustment coefficient, is the joint energy term in the jump direction, is the corresponding information entropy function value, and exp is the natural exponential function;

[0056] S53. After accepting the jump to the current node, construct a perturbation subgraph centered on the current node; calculate the perturbation intensity ψ for each node k in the perturbation subgraph k :

[0057]

[0058] where ψ k is the perturbation intensity of node k, Ω k is the three-dimensional spatial region covered by node k, is the gradient vector field of the geometric structure, is the gradient vector field of the attribute features, is the gradient vector field of the semantic information, λ1, λ2, and λ3 are weighting coefficients, ∥·∥ represents the vector norm, and ∫∫∫ represents the three-dimensional integral operation;

[0059] S54. Sort the nodes in the perturbation subgraph according to the perturbation intensity from high to low, and select the top n nodes to perform the incremental data replacement operation;

[0060] S55. Update the jump round counter. If the current round number is less than the maximum jump round number T max , then return to step S52 to continue the jump and update. If the jump round number limit is reached, end the path selection and incremental update process.

[0061] Optionally, the specific content of S6 includes:

[0062] S61. Record all updated spatial unit nodes after each round of incremental update, and locate the corresponding unit layer nodes in the improved hierarchical Bayesian network model;

[0063] S62. For each updated node, update its prior distribution in the Bayesian network model using the exponential weighted correction method according to the feature vector and state variable:

[0064] P ′ (R i = 1) = (1 - α)·P(R i = 1) + α·δ i ;

[0065] where P ′ (R i = 1) represents the corrected prior distribution of spatial unit node i, P(R i = 1) is the original prior probability, δ i is the observed state label of node i, and α is the prior update coefficient;

[0066] S63. Extract the adjacent spatial unit nodes of each updated node in the spatial topology graph model, construct an adjacency set, and recalculate the updated posterior probability for the adjacent nodes

[0067]

[0068] wherein, P(R j = 1|X j ) ′ represents the updated posterior probability correction value of the adjacent node j, P(R j = 1|X j ) is the original posterior probability, U is the set of updated nodes, ω i,j is the propagation weight from node i to node j, and η1, η2 are weighting parameters satisfying η1 + η2 = 1;

[0069] S64. Perform normalization processing on all the corrected updated posterior probabilities to maintain the consistency and additivity in the Bayesian network joint probability inference process;

[0070] S65. Set the jump termination criterion, including that the posterior probability change rate is lower than the convergence threshold, the average correction amplitude is lower than the perturbation threshold, and the number of jump rounds reaches the preset upper limit. If any of the conditions is met, terminate the update process. If not, return to step S5 to continue performing path sampling and incremental update.

[0071] The beneficial effects of the present invention are as follows:

[0072] The present invention proposes a GIS data dynamic update optimization algorithm based on three-dimensional digital twin. By fusing multi-source heterogeneous data such as remote sensing images, lidar, and edge sensors, it realizes the efficient perception and refined modeling of three-dimensional space scenes, overcomes the technical bottlenecks in traditional GIS data update such as difficult data fusion, insufficient fineness, long update cycle, and low spatial index efficiency, and effectively improves the real-time performance, accuracy of data update, and the efficiency of spatial analysis.

[0073] The present invention uses an octree structure to perform multi-level spatial unit division on the target area, constructs a refined three-dimensional spatial index system, and realizes efficient spatial position query and fast retrieval through Morton coding, significantly improving the efficiency of update path selection and spatial positioning. At the same time, using the spatial topology graph modeling technology, combined with attribute similarity and structural adjacency degree, a refined topological relationship between nodes is established, which can accurately reflect the spatial structure difference and semantic similarity, thus significantly improving the accuracy of GIS data update and the optimization effect of node selection.

[0074] Based on an improved hierarchical Bayesian network model, the present invention fully considers the change frequency, structural complexity, attribute features, and semantic information of spatial units, dynamically calculates the update priority and posterior probability of spatial units, and forms an adaptive node update path. During the incremental update process, the present invention can dynamically correct the parameters of the network model and the update probability of adjacent nodes, achieving efficient and accurate updates under long-term stable operation, significantly reducing the problems of redundant updates and update delays in key areas, and ensuring the continuous high-precision performance of GIS data in complex urban environments.

[0075] Therefore, the present invention has significant beneficial effects such as strong data fusion ability, efficient spatial indexing, strong self-adaptability of the update strategy, and good long-term stability. It can be widely applied in fields such as smart cities, emergency management, and urban planning, effectively supporting the refined management and intelligent decision-making needs of urban three-dimensional digital twins. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. In the accompanying drawings:

[0077] Figure 1 is a flowchart of a GIS data dynamic update optimization algorithm based on three-dimensional digital twins proposed by the present invention;

[0078] Figure 2 is a flowchart of the recursive partitioning of the octree structure of spatial units and the generation of Morton coding in the present invention;

[0079] Figure 3 is a schematic structural diagram of an improved hierarchical Bayesian network model proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0080] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0081] Refer to Figures 1-3 , a GIS data dynamic update optimization algorithm based on three-dimensional digital twins, includes the following steps:

[0082] S1. Collect GIS data, perform preprocessing, calculate the change frequency based on multi-temporal observation results, determine the data coverage area as the target area, and generate a three-dimensional observation data set with a unified structure;

[0083] S2. Within the target area, use the octree structure to divide the space into multi-level spatial units, construct a 3D digital twin model using each spatial unit, and establish a 3D spatial index system based on the level and coordinate position of the spatial unit;

[0084] S3. Map each spatial unit to a node in the graph model, establish edge connections based on spatial adjacency relationships and semantic labels, and construct a spatial topology graph model;

[0085] S4. According to the structure and connection relationships of the nodes in the spatial topology graph model, construct an improved hierarchical Bayesian network model, input the change frequency, structural complexity, and attribute characteristics of each node in the 3D observation dataset, calculate the updated posterior probability, and generate an update priority sequence;

[0086] S5. Select nodes from the spatial topology graph model in sequence according to the update priority sequence, perform differential comparison in combination with the 3D observation dataset, identify the spatial areas where the structure or attributes have changed, and perform incremental updates;

[0087] S6. After each round of update, correct the prior distribution of the updated nodes in the improved hierarchical Bayesian network model, dynamically adjust the update probabilities of adjacent nodes in the spatial topology graph model, and repeat until the termination condition is met.

[0088] Through the construction of an octree space division and a fine 3D spatial index system, the present invention significantly improves the retrieval and positioning efficiency of spatial data. At the same time, by combining the spatial topology graph model and the improved hierarchical Bayesian network, it realizes the dynamic calculation of the updated posterior probability and the update priority ranking, effectively optimizing the update path selection strategy. In addition, through incremental updates and dynamic correction of network parameters, the GIS data update process becomes more accurate and real-time, significantly reducing redundant calculations and enhancing the stability and reliability of long-term operation.

[0089] In this embodiment, the GIS data includes remote sensing images, lidar, and edge sensor data. The preprocessing includes format standardization, time synchronization, and spatial registration, extracting spatial structure, attribute characteristics, and semantic information. The semantic labels are generated from the extracted semantic information.

[0090] By fusing multi-source data such as remote sensing images, lidar, and edge sensors, the present invention improves the richness and comprehensiveness of the GIS data sources. At the same time, through preprocessing measures such as format standardization, time synchronization, and spatial registration, it ensures the consistency of the data in terms of structure and spatial scale. In addition, by extracting spatial structure, attribute characteristics, and semantic information to generate accurate semantic labels, it provides reliable support for the subsequent construction of the spatial topology graph, thereby improving the accuracy and semantic correlation of data updates.

[0091] In this embodiment, S2 specifically includes:

[0092] S21. Divide the target area into cube units with a fixed side length in the three-dimensional coordinate space to generate an initial space division structure;

[0093] S22. Recursively divide the initial space units using an octree structure. Each parent unit is divided into two sub-units in the horizontal axis, vertical axis, and elevation axis directions to obtain eight sub-space units. Set the maximum division level to L, and stop dividing when the current level reaches L;

[0094] S23. Calculate the Morton code for the space units at the l-th level:

[0095]

[0096] where I l is the index code of the space units at the l-th level, a k is the discrete coordinate of the space units at the k-th level in the horizontal axis direction, b k is the discrete coordinate of the space units at the k-th level in the vertical axis direction, c k is the discrete coordinate of the space units at the k-th level in the elevation direction, << is the bitwise left shift operator, k is the level number of the current summation term, l is the level number where the current space unit is located, and L is the maximum division level of the octree structure;

[0097] S24. Construct the index codes of all levels into an index tree structure, establish a mapping relationship from the three-dimensional space coordinates to the Morton code path, form a hierarchical three-dimensional space index system that can be recursively queried, and store it in the space index table;

[0098] S25. Establish a one-to-one mapping between each index item in the space index table and the corresponding space unit in the three-dimensional digital twin model.

[0099] In the present invention, by using an octree structure to perform fine-grained space division in the target area, a cube space unit structure with clear levels is generated. By introducing the Morton code for space unit identification, efficient mapping and rapid retrieval between three-dimensional space coordinates and index codes are achieved, significantly improving the query efficiency of space data. In addition, by establishing an index tree structure and an accurate one-to-one mapping relationship with the space units of the three-dimensional digital twin model, the accuracy, scalability of space information management, and the convenience of data maintenance are effectively improved.

[0100] In this embodiment, S3 specifically includes:

[0101] S31. Sequentially map the space units generated by the octree structure division to the nodes in the space topology graph model;

[0102] S32. Establish an undirected edge connection relationship between nodes based on the spatial adjacency relationship, and add edge connections to the spatial unit nodes with shared faces or shared edges;

[0103] S33. Introduce attribute similarity and structural adjacency degree as graph edge weight factors, and assign a weight value w to each edge in the graph structure ij :

[0104] w ij = α·d ij + β·r ij + γ·s ij ;

[0105] wherein, w ij is the edge weight between node i and node j, d ij is the spatial distance between node i and node j, r ij is the historical change correlation degree between node i and node j, s ij is the semantic similarity degree between node i and node j, and α, β, γ are weighting coefficients;

[0106] S34. Structurally store the constructed spatial topology graph model in the form of an adjacency matrix for modeling and calculation.

[0107] In the present invention, by mapping spatial units to the nodes of the topology graph model and accurately establishing the edge connections between nodes according to the spatial adjacency relationship, the topological structure and spatial correlation relationship between spatial units are clearly reflected. At the same time, by introducing attribute similarity and structural adjacency degree to construct edge weights, the expression ability of the topology graph model for spatial structure differences and semantic information is effectively enhanced. In addition, through the structural storage of the adjacency matrix, the calculation efficiency and data processing ability of the topology graph model are significantly improved, providing an accurate and efficient basic support for subsequent dynamic update decisions.

[0108] In this embodiment, the S4 specifically includes:

[0109] S41. Construct an improved hierarchical Bayesian network model, set three types of nodes: a state layer, a region layer, and a unit layer. The state layer corresponds to global control variables, the region layer corresponds to the sub-graph structure in the spatial topology graph model, and the unit layer corresponds to spatial unit nodes, and establish directed connections according to the spatial index relationship and semantic label relationship;

[0110] S42. Extract the change frequency, structural complexity, semantic encoding, and adjacency perturbation degree of each spatial unit, and form a feature vector after normalization processing;

[0111] S43. Calculate the updated posterior probability of each spatial unit:

[0112]

[0113] Among them, P i is the updated posterior probability of the spatial unit i, X i is the normalized eigenvector of the spatial unit i, R i is the state variable of the spatial unit i, P(X i |R i =1) is the conditional probability density of the eigenvector under the condition that the state is updated, P(R i =1) is the prior probability that the state is updated, P(X i ) is the marginal probability density of the eigenvector;

[0114] S44. Calculate the marginal probability density using Bayes' theorem and the feature distribution model:

[0115] P(X i ) = P(X i |R i =1)·P(R i =1) + P(X i |R i =0)·P(R i =0);

[0116] Among them, P(X i |R i =0) is the conditional probability density of the eigenvector under the condition that the state is not updated, P(R i =0) is the prior probability that the state is not updated;

[0117] S45. Sort the updated posterior probabilities corresponding to all spatial units in descending order of value to generate an update priority sequence.

[0118] In the present invention, by constructing an improved hierarchical Bayesian network model of the state layer, region layer, and unit layer, the hierarchical relationship between the global control variable, regional topological structure, and spatial unit is clearly expressed. Through feature extraction and normalization processing of the change frequency, structural complexity, semantic coding, and adjacency perturbation degree of the spatial unit, the accuracy and stability of the calculation of the updated posterior probability are effectively improved. In addition, the Bayesian theorem is used to accurately calculate the update priority sequence of the spatial unit, making the update process more efficient and accurate, and effectively reducing redundant calculations and update delays.

[0119] In this embodiment, the S44 specifically includes:

[0120] S441. Classify the normalized eigenvectors of each spatial unit node according to the state variable to construct two sample sets with updated and non-updated states;

[0121] S442. Calculate the feature mean vector and covariance matrix for each type of sample set;

[0122] S443. To improve the numerical stability of the model, a regularization term is introduced to adjust the covariance matrix and construct a corrected covariance matrix.

[0123]

[0124] Among them, is the corrected covariance matrix, Σ (r) is the characteristic covariance matrix when the state is r, ∈ is a positive real constant, and I is the identity matrix;

[0125] S444. Based on the corrected covariance matrix and the mean vector, construct a conditional probability density function:

[0126]

[0127] Among them, P(X i ∣R i =r) is the characteristic joint probability density of node i under the condition that the state is r, X i is the characteristic vector of node i, μ (r) is the characteristic mean vector when the state is r, is the corrected covariance matrix, Σ (r) is the characteristic covariance matrix when the state is r, is the vector transpose, (·) -1 is the matrix inversion, |·| is the matrix determinant, exp is the natural exponential function, π is the constant of pi, and d is the dimension of the characteristic vector;

[0128] S445. Apply the conditional probability density function to the posterior probability calculation process of S43 in claim 5 to generate an updated priority sequence.

[0129] The present invention classifies the characteristic vectors of the spatial unit nodes and constructs a sample set, accurately calculates the characteristic means and covariance matrices in each state, and at the same time introduces a regularization term to adjust the covariance matrix, effectively improving the numerical stability and generalization ability of the model. In addition, by constructing a conditional probability density function based on the corrected covariance matrix, the accuracy and reliability of the posterior probability calculation are further improved, thereby optimizing the update priority sorting of the spatial unit and making the data update strategy more efficient, stable and accurate.

[0130] In this embodiment, the S5 specifically includes:

[0131] S51. Construct an initial candidate node set, set the maximum number of jump rounds T max , starting from the spatial unit node with the largest update probability value, initialize the current node and the jump round counter;

[0132] S52. Construct an adjacency candidate node set at the current node, and calculate the sampling acceptance probability for any candidate node j:

[0133]

[0134] where α i→j is the acceptance probability of jumping from node i to node j, P i , P j are the update probabilities of nodes i and j, β is a non-linear adjustment coefficient, γ is an energy adjustment coefficient, is the joint energy term in the jump direction, is the corresponding information entropy function value, and exp is the natural exponential function;

[0135] S53. After accepting the jump to the current node, construct a perturbation subgraph centered on the current node; calculate the perturbation intensity ψ for each node k in the perturbation subgraph k :

[0136]

[0137] where ψ k is the perturbation intensity of node k, Ω k is the three-dimensional space region covered by node k, is the gradient vector field of the geometric structure, is the gradient vector field of the attribute feature, is the gradient vector field of the semantic information, λ1, λ2, and λ3 are weighting coefficients, ∥·∥ represents the vector norm, and ∫∫∫ represents the three-dimensional integral operation;

[0138] S54. Sort the nodes in the perturbation subgraph according to the perturbation intensity from high to low, and select the top n nodes to perform the incremental data replacement operation;

[0139] S55. Update the jump round counter. If the current round number is less than the maximum jump round number T max , then return to step S52 to continue performing jumps and updates. If the jump round number limit is reached, end the path selection and incremental update process.

[0140] The present invention effectively controls the iteration range and computational cost of the update process by constructing an initial candidate node set based on update probabilities and introducing a limit on the number of jump rounds. The introduction of the sampling acceptance probability combines the update probabilities and energy information between nodes, realizing an adaptive path jump mechanism, improving the rationality of the update path selection and the global exploration ability. At the same time, by constructing a perturbation subgraph and introducing multi-dimensional gradient field to calculate the perturbation intensity, the key change regions are accurately identified, realizing efficient incremental data replacement. The overall process not only ensures the update accuracy but also greatly improves the efficiency and intelligence of spatial data update.

[0141] In this embodiment, step S6 specifically includes:

[0142] S61. Record all updated spatial unit nodes after each round of incremental update, and locate the corresponding unit layer nodes in the improved hierarchical Bayesian network model;

[0143] S62. For each updated node, update its prior distribution in the Bayesian network model by using the exponential weighted correction method according to the feature vector and state variable:

[0144] P ′ (R i =1)=(1 - α)·P(R i =1)+α·δ i ;

[0145] Where, P ′ (R i =1) represents the corrected prior distribution of the spatial unit node i, P(R i =1) is the original prior probability, δ i is the observed state label of node i, and α is the prior update coefficient;

[0146] S63. Extract the adjacent spatial unit nodes of each updated node in the spatial topology graph model, construct an adjacency set, and recalculate the updated posterior probability for the adjacent nodes:

[0147]

[0148] Where, P(R j =1|X j ) ′ represents the corrected value of the updated posterior probability of the adjacent node j, P(R j =1|X j ) is the original posterior probability, U is the set of updated nodes, ω i,j is the propagation weight from node i to node j, and η1, η2 are weighting parameters satisfying η1 + η2 = 1;

[0149] S64. Normalize all the updated posterior probabilities after correction to maintain the consistency and additivity in the Bayesian network joint probability inference process;

[0150] S65. Set the jump termination criterion, including that the change rate of the posterior probability is lower than the convergence threshold, the average correction amplitude is lower than the perturbation threshold, and the number of jump rounds reaches the preset upper limit. If any condition is met, terminate the update process. If not, return to step S5 to continue the path sampling and incremental update.

[0151] Through the recording and positioning of the updated spatial units in each round, the present invention realizes the dynamic correction of the prior distribution in the Bayesian network model, enhances the adaptability of the model to real-time changes, adopts an exponentially weighted update strategy, effectively integrates historical knowledge and the latest observations, improves the robustness of the prior estimation. At the same time, through the propagation correction of the posterior probabilities of adjacent nodes and the normalization process, it ensures the consistency and additivity in the continuous update of the Bayesian network inference process. Combined with the setting of the jump termination criterion, the update process has convergence and stability, significantly improving the accuracy and computational efficiency of the dynamic update strategy.

[0152] Embodiment 1:

[0153] To verify the feasibility and superiority of the present invention in real scenarios, a central area of a coastal city is selected as the test area in this embodiment. This area is a key area for urban planning, with a total area of about 16 square kilometers, including dense high-rise buildings, underground passages, and commercial district transportation hubs. During the high-incidence period of natural disasters (such as typhoons and heavy rains), the original GIS system has problems such as response lag, outdated data, and low update efficiency, and cannot meet the real-time requirements of urban dynamic management and emergency decision-making.

[0154] The GIS data dynamic update and optimization algorithm based on three-dimensional digital twin proposed by the present invention is deployed in this area, and a comparative experiment is carried out with the traditional manual + batch remote sensing update system. The experimental data uses three types of fusion information sources: high-resolution remote sensing images (resolution 0.2m), vehicle-mounted lidar data (sampling at 10Hz), and edge node multi-modal sensors (120 nodes are deployed). A three-dimensional observation data set with a unified structure is constructed through format standardization, spatial registration, and time synchronization.

[0155] The system constructs a three-dimensional space index based on the octree structure, and combines the spatial topology graph and the Bayesian network model for dynamic update inference. In each round of update, it automatically identifies the changed area and performs difference comparison and data replacement according to the perturbation intensity and the updated posterior probability. During the test period, 9 new structures, 2 old buildings demolished, and 4 underground facility adjustments occurred in this area. The system executed a total of 3 rounds of complete update processes, with a total time of 10.4 hours, which is significantly more efficient than the traditional scheme (about 18 hours per round).

[0156] Data analysis shows that the system of the present invention is comprehensively superior to the comparative method in terms of building boundary recognition error, topological structure recognition rate, semantic matching degree, and high-risk area detection coverage rate, and maintains a stable performance in terms of computing resource usage. The specific comparison results are shown in Table 1:

[0157] Table 1 Performance comparison between the GIS data dynamic update algorithm of the present invention and traditional methods

[0158]

[0159]

[0160] As described above, the above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A dynamic update and optimization algorithm for GIS data based on three-dimensional digital twins, characterized in that, It includes the following steps: S1. Collect GIS data, perform preprocessing, calculate the change frequency based on multi-temporal observation results, determine the data coverage area as the target area, and generate a three-dimensional observation dataset with a unified structure; S2. In the target area, use the octree structure to divide the space into multi-level spatial units, construct a three-dimensional digital twin model using each spatial unit, and establish a three-dimensional spatial index system based on the level and coordinate position of the spatial unit; S3. Map each spatial unit to a node in the graph model, establish edge connections based on spatial adjacency relationships and semantic labels, and construct a spatial topology graph model; S4. According to the structure and connection relationships of the nodes in the spatial topology graph model, construct an improved hierarchical Bayesian network model, input the change frequency, structural complexity, and attribute characteristics of each node in the three-dimensional observation dataset, calculate the updated posterior probability, and generate an update priority sequence; S5. Select nodes from the spatial topology graph model in turn according to the update priority sequence, perform differential comparison in combination with the three-dimensional observation dataset, identify the spatial areas where the structure or attributes have changed, and perform incremental updates; S6. After each round of update, correct the prior distribution of the updated nodes in the improved hierarchical Bayesian network model, dynamically adjust the update probabilities of adjacent nodes in the spatial topology graph model, and repeat the execution until the termination condition is met.

2. The dynamic update and optimization algorithm for GIS data based on 3D digital twin according to claim 1, wherein, The GIS data includes remote sensing images, lidar, and edge sensor data. The preprocessing includes format standardization, time synchronization, and spatial registration, extracting spatial structures, attribute characteristics, and semantic information. The semantic labels are generated from the extracted semantic information.

3. A dynamic update and optimization algorithm for GIS data based on three-dimensional digital twin according to claim 1, characterized in that, The specific content of S2 includes: S21. Divide the target area into cube units with a fixed side length in the three-dimensional coordinate space to generate an initial spatial division structure; S22. Recursively divide the initial spatial unit using the octree structure. Each parent unit is divided into two sub-units in the horizontal axis, vertical axis, and elevation axis directions to obtain eight sub-spatial units. Set the maximum division level as L, and stop dividing when the current level reaches L; S23. Calculate the Morton code for the spatial units at the l-th level; Among them, I l is the index code of the l-th layer spatial unit, a k is the discrete coordinate of the k-th layer spatial unit in the horizontal axis direction, b k is the discrete coordinate of the k-th layer spatial unit in the vertical axis direction, c k is the discrete coordinate of the k-th layer spatial unit in the elevation direction, 《 is the bitwise left shift operator, k is the level number of the current summation term, l is the level number where the current spatial unit is located, and L is the maximum division level of the octree structure; S24. Construct the index codes of all levels into an index tree structure, establish a mapping relationship from the three-dimensional space coordinates to the Morton code path, form a hierarchical three-dimensional spatial index system that can be recursively queried, and store it in the spatial index table; S25. Establish a one-to-one mapping between each index item in the spatial index table and the corresponding spatial unit in the three-dimensional digital twin model.

4. An optimized algorithm for dynamic update of GIS data based on three-dimensional digital twin according to claim 1, characterized in that, The specific content of S3 includes: S31. Map the spatial units generated by the octree structure division to the nodes in the spatial topology graph model in turn; S32. Establish an undirected edge connection relationship between the nodes based on the spatial adjacency relationship, and add edge connections to the spatial unit nodes with shared faces or shared edges; S33. Introduce attribute similarity and structural adjacency degree as graph edge weight factors, and assign a weight value w to each edge in the graph structure ij : w ij = α·d ij + β·r ij + γ·s ij ; Among them, w ij is the edge weight between node i and node j, d ij is the spatial distance between node i and node j, r ij is the relevance of historical changes between node i and node j, s ij is the semantic similarity between node i and node j, and α, β, γ are weighting coefficients; S34. Structurally store the constructed spatial topology graph model in the form of an adjacency matrix for modeling and calculation.

5. A dynamic update and optimization algorithm for GIS data based on three-dimensional digital twins according to claim 1, characterized in that, The specific content of S4 includes: S41. Construct an improved hierarchical Bayesian network model, set three types of nodes: the state layer, the region layer, and the unit layer. The state layer corresponds to the global control variables, the region layer corresponds to the sub-graph structure in the spatial topology graph model, and the unit layer corresponds to the spatial unit nodes. Establish directed connections based on the spatial index relationship and the semantic label relationship; S42. Extract the change frequency, structural complexity, semantic encoding, and adjacency perturbation degree of each spatial unit, and form a feature vector after normalization; S43. Calculate the updated posterior probability of each spatial unit: where, P i is the updated posterior probability of spatial unit i, X i is the normalized eigenvector of spatial unit i, R i is the state variable of spatial unit i, P(X i |R i = 1) is the conditional probability density of the eigenvector under the condition that the state is updated, P(R i = 1) is the prior probability that the state is updated, P(X i ) is the marginal probability density of the eigenvector; S44. Calculate the marginal probability density using Bayes' theorem and the feature distribution model: P(X i ) = P(X i |R i = 1)·P(R i = 1) + P(X i |R i = 0)·P(R i = 0); Among them, P(X i ∣R i = 0) is the conditional probability density of the feature vector under the condition of not being updated, and P(R i = 0) is the prior probability of the state of not being updated; S45. Sort the updated posterior probabilities corresponding to all spatial units in descending order of value to generate an update priority sequence.

6. An optimization algorithm for dynamic update of GIS data based on three-dimensional digital twin according to claim 5, characterized in that The specific content of S44 includes: S441. Classify the normalized feature vectors of each spatial unit node according to the state variables, and construct two sample sets with the states of updated and not updated; S442. Calculate the feature mean vector and covariance matrix for each type of sample set; S443. To improve the numerical stability of the model, a regularization term is introduced to adjust the covariance matrix and construct a corrected covariance matrix. Among them, is the corrected covariance matrix, Σ (r) is the characteristic covariance matrix when the state is r, ∈ is a positive real constant, and I is the identity matrix; S444. Based on the corrected covariance matrix and mean vector, construct a conditional probability density function: where, P(X i ∣R i = r) is the joint probability density of the features of node i under the condition that the state is r, X i is the feature vector of node i, μ (r) is the feature mean vector when the state is r, is the corrected covariance matrix, Σ (r) is the feature covariance matrix when the state is r, is the vector transpose, (·) -1 is the matrix inversion, |·| is the matrix determinant, exp is the natural exponential function, π is the constant of pi, and d is the dimension of the feature vector; S445. Use the conditional probability density function in the posterior probability calculation process of S43 in claim 5 to generate an update priority sequence.

7. An optimized algorithm for dynamic update of GIS data based on 3D digital twin according to claim 1, characterized in that The specific content of S5 includes: S51. Construct an initial candidate node set and set the maximum number of jump rounds T max , starting from the spatial unit node with the largest updated probability value, initialize the current node and the jump round counter; S52. Construct an adjacent candidate node set at the current node, and calculate the sampling acceptance probability for any candidate node j: Among them, α i→j is the acceptance probability of jumping from node i to node j, P i , P j are the update probabilities of nodes i and j, β is the non-linear adjustment coefficient, γ is the energy adjustment coefficient, is the joint energy term in the jump direction, is the corresponding information entropy function value, and exp is the natural exponential function; S53. After accepting the jump to the current node, construct a perturbation subgraph centered on the current node; calculate the perturbation intensity ψ for each node k in the perturbation subgraph k : where, ψ k is the perturbation intensity of node k, Ω k is the three-dimensional spatial region covered by node k, is the gradient vector field of the geometric structure, is the gradient vector field of the attribute feature, is the gradient vector field of the semantic information, λ1, λ2, λ3 are weighting coefficients, ∥·∥ represents the vector norm, and ∫∫∫ represents the three-dimensional integral operation; S54. Sort the nodes in the perturbed sub-graph from high to low according to the perturbation intensity, and select the top n nodes to perform the incremental data replacement operation; S55. Update the jump round counter. If the current round number is less than the maximum jump round number T max , then return to step S52 to continue executing the jump and update. If the jump round number limit is reached, end the path selection and incremental update process.

8. An optimized algorithm for dynamic update of GIS data based on 3D digital twin according to claim 1, characterized in that The specific content of S6 includes: S61. Record all updated spatial unit nodes after each round of incremental update, and locate the corresponding unit layer nodes in the improved hierarchical Bayesian network model; S62. For each updated node, update its prior distribution in the Bayesian network model using the exponential weighted correction method according to the feature vector and the state variable; P ′ (R i = 1) = (1 - α)·P(R i = 1) + α·δ i ; where, P ′ (R i = 1) represents the corrected prior distribution of the spatial unit node i, P(R i = 1) is the original prior probability, δ i is the observation state label of node i, and α is the prior update coefficient; S63. Extract the adjacent spatial unit nodes of each updated node in the spatial topology graph model, construct an adjacent set, and recalculate the updated posterior probability for the adjacent nodes; where P(R j = 1|X j ) ′ represents the updated posterior probability correction value of the adjacent node j, P(R j = 1|X j ) is the original posterior probability, U is the set of updated nodes, ω i,j is the propagation weight from node i to node j, and η1, η2 are weighting parameters satisfying η1 + η2 = 1; S64. Perform normalization processing on all corrected updated posterior probabilities to maintain the consistency and additivity in the Bayesian network joint probability inference process; S65. Set jump termination criteria, including that the posterior probability change rate is lower than the convergence threshold, the average correction amplitude is lower than the perturbation threshold, and the number of jump rounds reaches the preset upper limit. If any condition is met, terminate the update process. If not, return to step S5 to continue the path sampling and incremental update.

Images