Digital entity model integration and interaction method and system considering data interconnection and fusion

Abstract

This application relates to the field of data processing technology, specifically to a method and system for integrating and interacting digital entity models considering data interconnection and fusion. The method includes: collecting road features and constructing semantic key vectors by combining sensor types and historical features; obtaining query vectors, obtaining comprehensive weights based on the similarity between the query vectors and semantic key vectors and data quality factors, and determining fused observations by combining them with value vectors; constructing fused observation matrices and comprehensive weight matrices based on the fused observations and comprehensive weights respectively, and obtaining observation fidelity terms based on the difference between the latent state matrix and the fused observation matrix combined with the comprehensive weight matrix; constructing a Laplace matrix, using it as weights and combining it with the latent state matrix to obtain topological smoothing terms; determining an objective function based on these two terms to obtain the topology-corrected latent state matrix and the entity's trust level; and updating the model based on the topology-corrected latent state matrix and the trust level. This application improves the credibility of digital entity models.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a method and system for integrating and interacting digital entity models that considers data interconnection and fusion. Background Technology

[0002] In transportation infrastructure, a digital entity model (DEM) refers to a dynamic mapping of physical entities, such as bridges, roads, and vehicles, in digital space. It is a digital twin formed by integrating the entity's geometry, physical attributes, behavioral characteristics, and real-time status data. This model not only contains static structural information but also reflects the dynamic changes of the entity through real-time data, thereby supporting full lifecycle monitoring, prediction, and optimization decisions for physical entities. Integrated interaction refers to the semantic alignment and fusion of data from different sources and modalities, thereby collaboratively correcting multiple entity models and updating the digital models of each entity.

[0003] Traditional integrated interaction methods for digital entity models typically first aggregate heterogeneous data from cameras, point clouds, and sensors into a unified database through format conversion, protocol alignment, or middleware. Then, they use simple averaging, voting, or single-model inference to generate and update entity states. However, this method fails to consider the interconnectivity between data sources. Each source is treated as equally weighted and trustworthy, without semantically measuring its relevance to the current entity or distinguishing whether the source is currently reliable. Furthermore, adjacent entities on the topology are updated independently, without utilizing spatial continuity to constrain outliers. Therefore, isolated noise is amplified layer by layer. This application addresses the technical problem of unreliable updates for the same entity due to observational conflicts caused by a lack of constraints in multi-source heterogeneous data by incorporating data interconnection and fusion techniques. Summary of the Invention

[0004] To address the technical problem of low reliability in updated models, this application provides a method and system for integrating and interacting digital entity models that considers data interconnection and fusion. The specific technical solution adopted is as follows:

[0005] In a first aspect, this application proposes a digital entity model integration and interaction method that considers data interconnection and fusion, the method comprising the following steps:

[0006] Different road features of each entity are collected by different sensors, and the road features are combined with the sensor type and the historical features of the entity to form the semantic key vector of each road feature;

[0007] The historical features of the entity are encoded and projected to obtain the query vector. For each road feature of the entity, the product of the similarity between the query vector and the semantic key vector and the data quality factor is used as the comprehensive weight. Then, the fused observation value of the entity is obtained by combining the value vectors in all road features.

[0008] A fusion observation matrix and a comprehensive weight matrix are constructed based on the fused observations and comprehensive weights of all entities. The difference between the hidden state matrix to be solved and the fusion observation matrix is ​​used to calculate the squared error through the comprehensive weights to obtain the observation fidelity term. For the connection between entities in the road network composed of all entities, the Laplace matrix is ​​obtained. The Laplace matrix is ​​used as the weight to calculate the squared error of the hidden state matrix to be solved, and then combined with the regularization coefficient to obtain the topology smoothing term. The objective function is obtained based on the topology smoothing term and the observation fidelity term. The hidden state matrix of topology correction and the confidence degree of each road feature are obtained by minimizing the objective function.

[0009] The entity model is updated based on the difference between the hidden state matrix after topology correction and the parameters of the entity model before the update, combined with the confidence level of road features.

[0010] In the above scheme, this application uses dot product similarity and real-time quality factor as weights to obtain a fused observation vector from multi-source heterogeneous observation data. Then, it uses the topological relationship of multiple adjacent entity nodes to construct a weighted graph Laplacian regularization and solve it linearly. Noise suppression and missing data imputation under spatial continuity constraints are achieved through a single sparse matrix solution. Finally, the model parameters of each road segment are weighted and averaged by combining the trust degree obtained in the solution process to complete the parameter update of the cloud model. It achieves efficient denoising of multi-source heterogeneous data, missing data repair under spatial continuity constraints, and cross-node trust parameter aggregation, which significantly improves the trustworthiness and robustness of digital entity models, while avoiding the high computational and communication burden caused by complex attention mechanisms, iterative filtering, and federated learning.

[0011] In one embodiment, the road features include crack features, indentation features, material features, and traffic features; the road features are composed of a value vector and a data quality factor.

[0012] In one embodiment, the method for constructing the semantic key vector for each road feature by combining the road features with the sensor type and the historical features of the entity is as follows:

[0013] The road feature value vector and data quality factor are mapped to 16-dimensional data using a high-dimensional mapping algorithm; the sensor type is mapped to 32-dimensional data using one-hot encoding; historical features are embedded using embedding and concatenated into 64-dimensional data.

[0014] All road features, the corresponding sensor types, and historical features are combined into a 128-dimensional semantic key vector.

[0015] In one embodiment, the method of using the product of the similarity between the query vector and the semantic key vector and the data quality factor as a comprehensive weight, and then combining it with the value vectors from all road features to obtain the fused observation value of the entity is as follows:

[0016] , This represents the query vector for the nth entity. The semantic key vector representing the j-th road feature of the n-th entity. The data quality factor represents the feature of the nth entity and the jth road. This represents the value vector of the feature of the nth entity and the jth road. Represents extremely small integers. This represents the fused observation value of the nth entity.

[0017] In one embodiment, the fused observation matrix is ​​a matrix composed of the fused observations of all entities as the diagonal elements of the matrix; the comprehensive weight matrix is ​​a matrix composed of the comprehensive weights of all entities as the diagonal elements of the matrix.

[0018] In one embodiment, the method for obtaining the Laplace matrix is ​​as follows:

[0019] For a road network consisting of all entities, if two entities are connected, the elements of the adjacency matrix are 1; if two entities are not connected, the elements of the adjacency matrix are 0. The number of direct connections between each entity and the other entities is used as the elements of the degree matrix. The rows and columns of both the adjacency matrix and the degree matrix are entities, and the row and column indices of the same entity are the same. The degree matrix is ​​a diagonal matrix.

[0020] In one embodiment, the expression for the objective function is:

[0021] , Represents the trace of a matrix. Let represent the hidden state matrix to be solved. This represents the fused observation matrix of all entities. Represents the Laplace matrix, Representation matrix transpose, This represents the transpose of matrix H. Represents the regularity coefficient. Let represent the objective function. The optimal objective function is obtained by minimizing the objective function. Indicates the observation fidelity term; This represents the topological smoothing term.

[0022] In one embodiment, the method for obtaining the hidden state matrix of topology correction and the trust level of each entity by minimizing the objective function is as follows:

[0023] By taking the first-order optimality condition, we differentiate H and transform the objective function using the differential property of the trace. For the transformed objective function, we set its derivative to 0 to obtain the equation. Then, we use the preconditional conjugate gradient algorithm to solve for the hidden state matrix of topology correction. The parameters of the entity in the last step of the preconditional conjugate gradient algorithm are used as the entity's confidence level.

[0024] In one embodiment, the method for updating the entity model based on the difference between the hidden state matrix obtained from topology correction and the entity model parameters before the update, combined with the entity's trust level, is as follows:

[0025] The hidden state matrix of topology correction is mapped to an incremental vector using a hard matrix. The incremental vector is then subjected to gradient clipping and normalization with the parameters before the entity model update to obtain the local contribution parameters of each entity. The local common line parameters of each entity are then weighted and averaged with the trust level of each entity to obtain the updated model parameters, thus completing the entity model update.

[0026] Secondly, embodiments of this application also provide a digital entity model integration and interaction system that considers data interconnection and fusion, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any of the above-described digital entity model integration and interaction methods that consider data interconnection and fusion.

[0027] The beneficial effects of this application are as follows:

[0028] This application uses dot product similarity and real-time quality factor as weights to obtain a fused observation vector from multi-source heterogeneous observation data. Then, it constructs a weighted graph Laplacian regularization using the topological relationships of multiple adjacent entity nodes and solves it linearly. Noise suppression and missing data imputation under spatial continuity constraints are achieved through a single sparse matrix solution. Finally, the model parameters of each road segment are weighted and averaged by combining the trust level obtained during the solution process to complete the parameter update of the cloud model. It achieves efficient denoising of multi-source heterogeneous data, missing data repair under spatial continuity constraints, and cross-node trust parameter aggregation, which significantly improves the trustworthiness and robustness of digital entity models, while avoiding the high computational and communication burden caused by complex attention mechanisms, iterative filtering, and federated learning. Attached Figure Description

[0029] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1This is a flowchart illustrating a digital entity model integration and interaction method considering data interconnection and fusion, provided as an embodiment of this application. Detailed Implementation

[0031] To further illustrate the technical means and effects adopted by this application to achieve the intended inventive objective, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the digital entity model integration and interaction method and system considering data interconnection and fusion proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0032] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0033] Implementation examples of digital entity model integration and interaction methods and systems considering data interconnection and fusion:

[0034] The following description, in conjunction with the accompanying drawings, details the specific scheme of the digital entity model integration and interaction method and system that considers data interconnection and fusion provided in this application.

[0035] Please see Figure 1 The diagram illustrates a flowchart of a digital entity model integration and interaction method considering data interconnection and fusion, according to an embodiment of this application. The method includes the following steps:

[0036] Step S001: Collect road features and combine them with sensor type and historical features to form a semantic key vector.

[0037] Each segment of the traffic infrastructure is treated as an entity. The road surface map of the entity is collected through the RTSP stream of the roadside camera. Frames are extracted locally at 30fps. The crack detection YOLOv8-seg model is used to perform real-time inference on the road surface map and output a road surface crack probability heat map. The proportion of crack pixels with a confidence level greater than 0.5 is taken as the value vector. At the same time, the reciprocal of the frame loss rate within 1 minute is used as the data quality factor. Crack features are obtained based on this.

[0038] Then, the road surface depressions were obtained using a 64-line lidar. The PCL library was used to count the number of depressions larger than 0.5 cm within a 10m solid area. The RMS value of the rut depth was calculated as the value vector, and the point cloud density was used as the data quality factor. Based on this, the depression features were obtained.

[0039] Furthermore, the SCADA system pushes six sets of asphalt temperature, humidity, and strain gauge readings for the entity every 5 minutes. After Kalman filtering to remove noise, the maximum value of the temperature and humidity gradient is taken as the value vector. At the same time, the reciprocal of the sensor packet loss rate is used as the data quality factor, and the material characteristics are obtained based on this.

[0040] Finally, the real-time vehicle speeds of entities are retrieved every 2 minutes via the map API as value vectors, and the number of sampled vehicles returned by the API is used as a data quality factor to obtain traffic features.

[0041] The acquired crack features, indentation features, material features, and traffic features are collectively referred to as road features. Each road feature is a heterogeneous data source. The value vector represents the numerical representation of the specific measurement or inference results of each data source for the target entity. After normalization and dimensional unification, it directly participates in the calculation. Its elements correspond to the physical or statistical indicators of the entity's state, ultimately generating a single comprehensive observation vector of the corresponding entity's current state. The data quality factor represents a scalar coefficient for real-time evaluation of the data source's credibility. It can be mapped to the 0-1 range using one or more of the following: frame drop rate, noise level, latency, calibration status, or certificate validity. It is used to evaluate the credibility of the corresponding observations in the weighted average, suppressing the negative impact of low-quality data on the fusion results and improving the system's robustness.

[0042] For the road features obtained above, synchronization is achieved through interpolation. After normalization of the data in the road features, a high-dimensional mapping algorithm is used to map the value vector and data quality factor of each road feature into 16-dimensional data. In this embodiment, the high-dimensional mapping algorithm uses high-dimensional polynomial mapping. The sensor type of the road features is collected and mapped to 32-dimensional data using one-hot encoding. Historical observation data of the road segment is compiled into a static record table. Historical features of entities are extracted from the static record table. These historical features include five fields: road grade, surface layer type, design year, year of last major or medium repair, and average historical crack density. These fields are then discretely binned, embedded, and concatenated into 64-dimensional data.

[0043] A 128-dimensional semantic key vector is constructed from the multidimensional data features of all road features, the corresponding sensor types, and historical features. The semantic key vector is a low-dimensional embedding vector in a unified coding space used to characterize the correlation between the semantic content of the observed data from the data source and the context of the digital entity. It is obtained by mapping the prior information of sensor type, observed physical quantity, and digital entity to a shared vector space and is used to quantify the semantic matching degree.

[0044] At this point, the semantic key vector of each road feature in the entity has been obtained.

[0045] Step S002: Obtain the query vector, obtain the comprehensive weight based on the similarity between the query vector and the semantic vector and the data quality factor, and determine the fused observation value by combining the value vector.

[0046] There are both semantic matching errors and confidence differences regarding sensor malfunctions among the observation data of multiple road features. If only cosine similarity is used, semantic deviations between road features or sensor malfunctions can lead to deviations in the state of the actual road segment. If only data quality is considered, semantically unrelated but high-quality neighboring noise may be used as the basis for road segment identification. Therefore, the entity fusion observation vector must consider both semantic matching degree and real-time reliability.

[0047] For each entity, the entity's road history features are encoded using one-hot encoding and then subjected to a linear projection layer to obtain a 128-dimensional vector, which is denoted as the query vector.

[0048] For each road feature of an entity, the similarity between the query vector and the semantic key vector, as well as the data quality factor, are used as weights to combine the value vectors from all road features to obtain the entity's fused observation value.

[0049] Preferably, in this embodiment, the expression for the fused observations is:

[0050] , This represents the query vector for the nth entity. The semantic key vector representing the j-th road feature of the n-th entity. The data quality factor represents the feature of the nth entity and the jth road. This represents the value vector of the feature of the nth entity and the jth road. This represents a very small integer, and its purpose is to prevent the denominator from being 0. This represents the fused observation value of the nth entity.

[0051] In this system, the query vector represents the semantic space direction of greatest interest to the entity. The dot product of the semantic key vector and the query vector represents the similarity between the vectors, indicating the degree of semantic matching in the vector space; a higher value indicates greater relevance. A larger data quality factor indicates better data quality and higher reliability. Multiplying this factor by the vector similarity yields a comprehensive weight, giving higher weight to observations with high semantic matching quality. A weighted average of the value vectors is then applied using this comprehensive weight; a larger value indicates a more significant anomaly, while a smaller value indicates a more normal entity state or less reliable results due to missing observations. The comprehensive weight is... .

[0052] At this point, the fused observations for each entity have been obtained.

[0053] Step S003: Construct a fused observation matrix and a comprehensive weight matrix based on the fused observations and comprehensive weights respectively; obtain the observation fidelity term by combining the comprehensive weight matrix with the difference between the hidden state matrix and the fused observation matrix; construct a Laplace matrix and combine it with the hidden state matrix as weights to obtain the topology smoothing term; determine the objective function based on the two terms and then obtain the hidden state matrix and confidence level of topology correction.

[0054] Even after fusion, the weighted vector may still contain sensor noise, communication packet loss, and semantic mapping errors. Therefore, spatial autocorrelation between adjacent entities can be utilized, i.e., the structure and anomalous evolution of adjacent entities often have strong coupling relationships. This application introduces graph Laplace regularization, which integrates the effectiveness of observations and the consistency of the neighborhood by solving a weighted least squares problem with topological constraints. A unified objective function is used to represent spatial continuity with the Laplace matrix corresponding to the infrastructure adjacency graph. Noise suppression and missing data imputation are achieved in the analytical solution, so that the corrected hidden state not only ensures high reliability of local observations but also conforms to the smooth prior of continuous road segments in space.

[0055] In order to update the entity model, it is necessary to obtain the contribution of each entity. This application constructs an objective function to calculate the hidden state matrix of all entities. The hidden state matrix is ​​the most likely true state of each number under topological constraints after Laplace smoothing. Each row corresponds to an entity and each column corresponds to a state.

[0056] The fused observation value of each entity is used as an element of a diagonal matrix. A fused observation matrix is ​​constructed based on the fused observation values ​​of all entities. The overall credibility of an entity is quantified by the squared error of the difference between the hidden state matrix to be solved and the fused observation matrix. The larger the value, the more the final solution tends to retain the original observation of that point; otherwise, it is more likely to trust the smoothness of the neighborhood constraint. When using the squared error, the weights used are the combined weights of each entity. The term used to calculate the weighted squared error is denoted as the observation fidelity term and is used as the first term of the objective function.

[0057] In a road network composed of all entities, entities are marked as 1 if they are connected and 0 if they are not connected. An adjacency matrix is ​​constructed for all entities, where rows and columns represent different entities, and the same entity has the same number of rows and columns. Then, the number of direct connections between each entity and all other entities is used as diagonal elements. A diagonal matrix is ​​constructed from these diagonal elements, denoted as the degree matrix for all entities. The rows and columns of the degree matrix represent different entities, and the same entity has the same number of rows and columns. The difference between the degree matrix and the adjacency matrix is ​​used as the Laplace matrix of the road network. The Laplace matrix is ​​used as weights. For the hidden state matrix to be solved, the product of the squared error and the regularization coefficient is used as the topological smoothing term, which is the second term in the objective function.

[0058] The sum of the first and second terms is used as the objective function.

[0059] In this embodiment, the expression for the objective function is:

[0060] , Represents the trace of a matrix. Let represent the hidden state matrix to be solved. This represents the fused observation matrix of all entities. Represents the Laplace matrix, Representation matrix transpose, This represents the transpose of matrix H. Represents the regularity coefficient. Let represent the objective function. The minimum of the objective function is considered the optimal objective function. The smaller the value of the regularization coefficient, the smaller the impact of the smoothing term. When it approaches 0, the solution degenerates into a point-by-point independent weighted average, preserving all local observation details, but noise and missing data remain. The larger the value, the greater the impact of the smoothing term. When it approaches positive infinity, it prioritizes minimizing adjacent differences, the hidden state tends to a global constant, achieving thorough smoothing but potentially masking true anomaly signals. In this embodiment, the value is set to 10.

[0061] By applying the first-order optimality condition, differentiating H with respect to H, and utilizing the differential property of the trace, the objective function is transformed into... Setting the derivative to 0, we obtain a linear system, which, when rearranged, gives... ,because It is a symmetric positive definite matrix and a large sparse symmetric matrix. It can be solved using the preconditional conjugate gradient (PCG) method, and a solution exists. The residual of each entity in the last step of the preconditional conjugate gradient algorithm is used as the confidence level of the road feature.

[0062] Thus, by solving, the hidden state matrix of the topology correction corresponding to the road network and the trust degree of each entity are obtained. The algorithm globally and analytically gives the optimal balance point of each entity under two opposite constraints. It does not require repeated gradient iteration or hyperparameter tuning. It only needs to use the conjugate gradient sparse solver to complete the denoising, missing and spatial continuity in linear complexity.

[0063] Thus, the hidden state matrix of the road network topology correction and the confidence level of each road feature have been obtained.

[0064] Step S004: Update the model based on the hidden state matrix and confidence level obtained from topology correction.

[0065] The hidden state after topological correction exists only in the computation layer, and there may be differences between the local digital twin parameter versions maintained by multiple entities. Directly exchanging the original gradients may be risky. This application uses the quantified trust degree of the entities as the aggregation weight, and completes the global parameter merging through a single weighted average, avoiding the communication and computational overhead caused by complex federated learning or blockchain consensus.

[0066] After obtaining the hidden state matrix for topology correction, it is differiated from the local old parameters of each road feature. The difference result is then mapped to an incremental vector of the same dimension as the local entity digital model using a projection matrix. Gradient clipping and normalization are then performed on this vector and the local old parameters of each road feature to obtain the local contribution parameters. The local contribution parameters of all road features are collected in the cloud, and a weighted average is calculated using trust level as the weight. This average is then used to update the parameters of the global model in the cloud, thus obtaining the updated entity model. The old parameters refer to the parameters of the entity model before the update.

[0067] At this point, the updated entity model has been obtained.

[0068] Based on the same inventive concept as the above methods, embodiments of the present invention also provide a digital entity model integration and interaction system considering data interconnection and fusion, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described digital entity model integration and interaction methods considering data interconnection and fusion.

[0069] It should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

[0070] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A digital entity model integration and interaction method considering data interconnection and fusion, characterized in that, The method includes the following steps: Different road features of each entity are collected by different sensors, and the road features are combined with the sensor type and the historical features of the entity to form the semantic key vector of each road feature; The historical features of the entity are encoded and projected to obtain the query vector. For each road feature of the entity, the product of the similarity between the query vector and the semantic key vector and the data quality factor is used as the comprehensive weight. Then, the fused observation value of the entity is obtained by combining the value vectors in all road features. A fusion observation matrix and a comprehensive weight matrix are constructed based on the fused observations and comprehensive weights of all entities. The difference between the hidden state matrix to be solved and the fusion observation matrix is ​​used to calculate the squared error through the comprehensive weights to obtain the observation fidelity term. For the connection between entities in the road network composed of all entities, the Laplace matrix is ​​obtained. The Laplace matrix is ​​used as the weight to calculate the squared error of the hidden state matrix to be solved, and then combined with the regularization coefficient to obtain the topology smoothing term. The objective function is obtained based on the topology smoothing term and the observation fidelity term. The hidden state matrix of topology correction and the confidence degree of each road feature are obtained by minimizing the objective function. The entity model is updated based on the difference between the hidden state matrix after topology correction and the parameters of the entity model before the update, combined with the confidence level of road features.

2. The digital entity model integration and interaction method considering data interconnection and fusion as described in claim 1, characterized in that, The road features include crack features, indentation features, material features, and traffic features; Road features consist of a value vector and a data quality factor.

3. The digital entity model integration and interaction method considering data interconnection and fusion as described in claim 2, characterized in that, The method for constructing the semantic key vector for each road feature by combining the road features with the sensor type and the historical features of the entity is as follows: The road feature value vector and data quality factor are mapped to 16-dimensional data using a high-dimensional mapping algorithm; the sensor type is mapped to 32-dimensional data using one-hot encoding; historical features are embedded using embedding and concatenated into 64-dimensional data. All road features, the corresponding sensor types, and historical features are combined into a 128-dimensional semantic key vector.

4. The digital entity model integration and interaction method considering data interconnection and fusion as described in claim 1, characterized in that, The method of obtaining the fused observation value of the entity by using the product of the similarity between the query vector and the semantic key vector and the data quality factor as a comprehensive weight, and then combining it with the value vectors from all road features, is as follows: , This represents the query vector for the nth entity. The semantic key vector representing the j-th road feature of the n-th entity. The data quality factor represents the feature of the nth entity and the jth road. This represents the value vector of the feature of the nth entity and the jth road. Represents extremely small integers. This represents the fused observation value of the nth entity.

5. The digital entity model integration and interaction method considering data interconnection and fusion as described in claim 1, characterized in that, The fusion observation matrix is ​​a matrix composed of the fusion observation values ​​of all entities as the diagonal elements of the matrix; the comprehensive weight matrix is ​​a matrix composed of the comprehensive weights of all entities as the diagonal elements of the matrix.

6. The digital entity model integration and interaction method considering data interconnection and fusion as described in claim 1, characterized in that, The method for obtaining the Laplace matrix is ​​as follows: For a road network consisting of all entities, if two entities are connected, the adjacency matrix has an element of 1; if two entities are not connected, the adjacency matrix has an element of 0. The number of direct connections between each entity and the other entities is used as the element of the degree matrix. The adjacency matrix and degree matrix are both composed of entities, and the row and column indices of the same entity are the same; the degree matrix is ​​a diagonal matrix.

7. The digital entity model integration and interaction method considering data interconnection and fusion as described in claim 1, characterized in that, The expression for the objective function is: , Represents the trace of a matrix. Let represent the hidden state matrix to be solved. This represents the fused observation matrix of all entities. Represents the Laplace matrix, Representation matrix transpose, This represents the transpose of matrix H. Represents the regularity coefficient. Let represent the objective function; where the minimum of the objective function is taken as the optimal objective function. Indicates the observation fidelity term; This represents the topological smoothing term.

8. The digital entity model integration and interaction method considering data interconnection and fusion as described in claim 1, characterized in that, The method for obtaining the hidden state matrix of topology correction and the trust degree of each entity by minimizing the objective function is as follows: By taking the first-order optimality condition, we differentiate H and transform the objective function using the differential property of the trace. For the transformed objective function, we set its derivative to 0 to obtain the equation. Then, we use the preconditional conjugate gradient algorithm to solve for the hidden state matrix of topology correction. The parameters of the entity in the last step of the preconditional conjugate gradient algorithm are used as the entity's confidence level.

9. The digital entity model integration and interaction method considering data interconnection and fusion as described in claim 1, characterized in that, The method for updating the entity model based on the difference between the hidden state matrix after topology correction and the entity model parameters before update, combined with the entity's trust level, is as follows: The hidden state matrix of topology correction is mapped to an incremental vector using a hard matrix. The incremental vector is then subjected to gradient clipping and normalization with the parameters before the entity model update to obtain the local contribution parameters of each entity. The local common line parameters of each entity are then weighted and averaged with the trust level of each entity to obtain the updated model parameters, thus completing the entity model update.

10. A digital entity model integration and interaction system considering data interconnection and fusion, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the digital entity model integration and interaction method considering data interconnection and fusion as described in any one of claims 1-9.

Images