Smart city space 3D model control method and device and storage medium

Abstract

The invention discloses a smart city space 3D model control method and device, and a storage medium. The method comprises the following steps: S1, obtaining and preprocessing real-time data of a smart city space; s2, constructing a dynamic graph convolutional network, mapping images and point cloud data, initializing a 3D model and establishing a topological structure; s3, calculating a sensor data change rate in combination with time characteristics, and updating a topological structure and a state; s4, optimizing rotation, zooming, displacement and deformation parameters by using the gating map attention network, and adjusting a motion track and a shape; s5, compressing the features by adopting a variable topological aggregation mechanism, and optimizing a topological structure, boundary curvature and grid density; and S6, outputting to a mobile terminal to realize visualization and multi-terminal interaction. And S7, storing and updating the 3D model, and optimizing feature learning and an adaptive adjustment strategy. According to the invention, the dynamic adaptability and interaction precision of the 3D model are improved, and efficient modeling, optimal control and multi-terminal cooperative operation of the smart city space are realized.

Description

Technical Field

[0001] The present invention relates to the technical fields of intelligent space and digital twin, and particularly to a control method, device, and storage medium for a 3D model of a smart city space. Background Art

[0002] In the construction and management of a smart city space, 3D models are widely used as the core data representation form in multiple fields such as building information modeling (BIM), intelligent manufacturing, digital twin, and smart city. The existing control technologies for 3D models mainly rely on rule setting, manual adjustment, or predefined parameter optimization, and cannot achieve adaptive adjustment to complex dynamic environments, which limits their intelligent applications in the smart city space. Especially when it comes to real-time interaction, precise control, and environmental adaptive optimization of large-scale 3D data, traditional methods have many limitations and are difficult to meet the requirements of the smart city space for high-precision, low-latency, and real-time adjustable 3D model control.

[0003] The current mainstream 3D model control methods rely on traditional methods based on geometric constraints and mathematical transformations, such as implementing basic operations such as rotation, scaling, and displacement through mathematical models such as Euler angles and quaternions. These methods are usually applicable to static scenarios or manual interactions, but when the 3D model needs to be adaptively adjusted according to sensor data or external environmental changes, traditional mathematical transformation methods cannot effectively handle complex dynamic relationships. For example, in an intelligent factory or autonomous driving simulation, the 3D model needs to be updated in real time with the change of sensor data, while traditional methods cannot establish a mapping relationship from environmental data to 3D model control parameters, resulting in lag or distortion of model adjustment. In addition, traditional 3D model control methods usually rely on predefined rules or fixed thresholds and lack self-learning and optimization capabilities, making the adaptability of the model poor and unable to meet the real-time and highly dynamic change requirements of the smart city space.

[0004] In recent years, the development of deep learning and computer vision technologies has provided new solutions for 3D model control. 3D model optimization methods based on deep learning technologies such as convolutional neural networks (CNNs) and graph neural networks (GNNs) can achieve intelligent model adjustment by extracting features from inputs such as images and point cloud data. However, existing deep learning methods still face many challenges when dealing with 3D models in the smart city space. First, existing 3D deep learning methods are usually trained based on static datasets and are difficult to adapt to the real-time changes in the smart city space, resulting in lagging model updates. Second, when dealing with 3D topological structures, deep learning models usually need to regularize the models, such as using uniform sampling or fixed grid partitioning, but this method is prone to losing key geometric details, leading to a decrease in the control accuracy of 3D models. In addition, most existing 3D deep learning methods are based on global feature learning, ignoring the local topological information of 3D models, making it easy to have problems such as local structure distortion or detail loss during model adjustment.

[0005] In the smart city space, the real-time control of 3D models requires the integration of multi-source data such as Internet of Things sensors, point cloud scanning, and cameras to achieve dynamic adjustment of the models. However, existing 3D model control methods have weak capabilities in fusing multi-source data and are difficult to achieve real-time optimization based on sensor feedback. For example, there are significant differences in the format, dimension, and temporal characteristics between the point cloud data obtained by lidar and the image data collected by cameras. When existing 3D model control methods fuse these data, they usually adopt simple interpolation or rule matching methods, which are difficult to accurately represent the dynamic changes in the smart city space. In addition, Internet of Things sensor data has a high degree of temporal correlation, and existing 3D model control methods lack the ability to model temporal information, resulting in an inability to accurately predict future states when dealing with the dynamic changes in the smart city space, making the adjustment of 3D models lagging or unstable.

[0006] Therefore, how to provide a control method, device, and storage medium for 3D models in the smart city space is an urgent problem for those skilled in the art to solve. Summary of the Invention

[0007] An object of the present invention is to propose a control method for 3D models in the smart city space. The present invention combines a dynamic graph convolutional network, a gated graph attention network, and a variable topology aggregation mechanism, and details the dynamic adjustment, intelligent optimization, and interactive control methods of 3D models in the smart city space, with the advantages of high real-time performance, high precision, and strong adaptability.

[0008] According to the control method for 3D models in the smart city space of the embodiments of the present invention, the following steps are included: S1. Obtain real-time data of the smart city space and preprocess the real-time data; S2. Build a dynamic graph convolutional network, map the preprocessed image data and point cloud data to a unified feature space, calculate the spatial structure information, initialize the 3D model, and establish the topological structure of the 3D model; S3. Combine the temporal characteristics of the dynamic graph convolutional network and the preprocessed sensor data, calculate the change rate of the sensor data, update the topological structure and state change trend of the 3D model, and keep the 3D model in dynamic consistency with the smart city space; S4. Based on the preprocessed instruction parameters, calculate the difference between the control target and the 3D model state, use the improved gated graph attention network to calculate the importance weights of the control target in the topological structure, adjust the rotation, scaling, displacement and deformation parameters of the 3D model, optimize the motion trajectory of the 3D model, and adjust the change range of the 3D model shape based on the topological structure; S5. Use the improved variable topology aggregation mechanism to perform feature compression on the 3D model topological structure, extract local geometric features, and based on the deformation distribution of the motion trajectory, adjust the edge connection method between adjacent vertices, and correct the boundary curvature and mesh density of the 3D model; S6. Output the 3D model to the mobile terminal to achieve visual display, and synchronize it to the remote computing platform to support multi-user collaborative interaction, achieve multi-person remote sharing control, and record interaction data; S7. Store the 3D model state, input the interaction data into the dynamic graph convolutional network, update the feature learning ability of the 3D model, optimize the adaptive adjustment strategy, and improve the response accuracy to the real-time environment and the interaction experience.

[0009] Optionally, the real-time data includes Internet of Things sensor data, image data collected by a camera, point cloud data obtained by a lidar, and control instructions input by a user. The preprocessing includes performing noise filtering and normalization processing on the sensor data, performing denoising, histogram equalization and feature extraction on the image data, performing downsampling, surface reconstruction and normal vector calculation on the point cloud data, and parsing the control instructions input by the user to extract rotation, scaling, displacement and deformation parameters; Optionally, the S2 specifically includes: S21. Based on the preprocessed image data and point cloud data, construct a 3D model topological structure G(V,E), where V is the set of vertices in the 3D model, E is the set of edges between vertices, and initialize the feature vector of each vertex and the feature vector of each edge ; S22. Use the dynamic graph convolutional network to perform topological structure learning on G, calculate the connection relationship between vertices using the dynamic adjacency matrix, and define as the vertex and The correlation weight between them, and adjust the adjacency matrix through a normalization operation: ; Among them, is the normalized adjacency matrix, is the adjacency matrix at time t, is 's degree matrix, and ; S23. Update the eigenvector of each vertex. Adopt the dynamic graph convolution calculation method, and perform feature aggregation through the message passing mechanism to calculate the vertex state: ; Among them, is the eigenvector of vertex v at time t + 1, is the training parameter, is the non-linear activation function, N(v) is the neighborhood set of vertex v, is the normalized correlation weight between vertex v and u, is the eigenvector of vertex u at time t; S24. Update the topological structure G(V, E) of the 3D model by updating the eigenvector of the vertex, and calculate the spatial structure information: ; Among them, is the local curvature value of vertex v of the 3D model at time t + 1, is the weight between vertex v and u, is the curvature value at vertex u, is the curvature value at vertex v.

[0010] Optionally, the specific steps of S3 include: S31. Define the sensor data set as where represents the data of the i-th sensor at time t, n is the number of sensors, and calculate the state vector of the current time step: ; Among them, is the state vector at time t, is the mapping weight matrix, is the bias term; S32. Calculate the change rate between the state vector of the current time step and the state vector of the previous time step, and define the change rate calculation formula: ; Among them, is the state vector at time t - 1, is the rate of change; Will As the input of the dynamic graph convolutional network, adjust the adjacency matrix in the topology structure and update the adjacency relationship: ; in, is the normalized adjacency matrix, is the adjacency matrix at time t-1, Based on The calculated topology adjustment matrix, is the degree matrix; S33. Based on the updated topological structure G(V,E), the dynamic graph convolutional network is used to calculate the state change trend and define the vertex feature update formula: ; in, is the eigenvector of vertex v at time t+1, is the training parameter, is a nonlinear activation function, N(v) is the neighborhood set of vertex v, is the normalized association weight between vertices v and u, is the eigenvector of vertex u at time t, is the bias term; S34, calculate the spatial position change of the updated vertex feature in the 3D model, and define the position update formula of the vertex v in the three-dimensional space: ; in, is the coordinate position of vertex v at time t+1, is the coordinate position of vertex v at time t, is the step size adjustment coefficient, represents the position change of vertex v, is the local curvature value of the 3D model vertex v at time t+1; S35. Adjust the 3D model status to ensure that the 3D model maintains dynamic consistency with the smart city space.

[0011] Optionally, the S4 specifically includes: S41, based on the 3D model topology structure G(V,E) and preprocessed instruction parameters , define the control target set: ; Where b is the number of control targets, represents the i-th control target at time t, and the control target includes rotation, scaling, displacement and deformation parameters; S42. Calculate the importance weights of the control objectives in the topological structure using the improved gated graph attention network, and define the influence weights of the control objectives on the vertices: ; Wherein, is the control objective, is the vertex, is the weight of the control objective on the vertex . is the feature vector of the vertex at time step t, is the neighborhood set of the vertex , is the natural exponential function with base e, is the interaction function of the control objective and the vertex feature, defined as: ; Wherein, and are trainable parameter matrices, is the bias term; S43. Based on the influence weights of the control objectives on the vertices, adjust the rotation, scaling, displacement and deformation parameters of the 3D model, and define the control update matrix: ; Wherein, is the new position of the vertex at time step t + 1, is the original position at time step t, and m is the number of control objectives; S44. Calculate the adjusted 3D model motion trajectory, and define the trajectory smoothing optimization objective function: ; Wherein, is the trajectory smoothing loss, and respectively represent the new positions of adjacent vertices v and u at time step t + 1, V is the set of vertices in the 3D model, N(v) is the neighborhood set of vertex v, represents the square of the Euclidean norm; S45. Adjust the morphological change range of the 3D model through the motion trajectory, and calculate the final morphological adjustment function: ; Wherein, is the final morphological adjustment position of vertex v at time step t + 1, is the morphological change control coefficient.

[0012] Optionally, the S5 specifically includes: S51. Based on the 3D model topological structure G(V, E) and the adjacency matrix , form a topological feature matrix by combining the feature vectors of each vertex at time t: ; where, is the vertex feature matrix, a is the number of vertices at time t, represents the vertex 's feature vector at time step t; Improve the variable topology aggregation mechanism and adaptively update the topological feature matrix: ; where, is the topological feature matrix at time step t, , , , is the trainable weight matrix, is 's transpose matrix, is the topological level adjustment parameter, is the adjacency matrix at time t, is the topological adjacency matrix of the k-th layer, and K is the number of topological levels; S52. Based on the topological aggregation matrix, obtain the motion trajectory matrix and calculate the motion deformation influence factor: ; ; where, is the motion trajectory matrix at time step t, is the motion trajectory feature of vertex at time step t, a is the number of vertices at time t, is the motion deformation influence factor at time step t, is the neighborhood set of vertex , is the motion deformation control parameter, is the natural exponential function with base e, represents the square of the Euclidean norm; S53. Combine the motion trajectory matrix and the motion deformation influence factor, calculate the boundary curvature adjustment matrix and the grid density distribution at time step t, and optimize the grid density: ; ; ; where, is the boundary curvature adjustment matrix at time step t, is the degree matrix, is the Laplacian matrix, is the curvature smoothing coefficient, is the grid density at time step t, is the grid density control parameter, and V is the set of vertices in the 3D model, is the grid density loss, is the target grid density; S54. Optimize the topological structure based on the boundary curvature and grid density, and update the 3D model state: ; wherein, is the 3D model state calculated at time step t + 1, I is the identity matrix, and are the balance coefficients, is the degree matrix at time step t, is the boundary curvature adjustment matrix calculated at time step t, is the optimization step size, is the grid density gradient descent adjustment term.

[0013] The control device for the 3D model of the smart city space according to the embodiment of the present invention includes: A network construction module, configured to construct a dynamic graph convolutional network, map the preprocessed image data and point cloud data to a unified feature space, calculate the spatial structure information, initialize the 3D model, and establish the topological structure of the 3D model; A state update module, configured to combine the time characteristics of the dynamic graph convolutional network and the preprocessed sensor data, calculate the change rate of the sensor data, update the topological structure and the state change trend of the 3D model, and keep the 3D model in dynamic consistency with the smart city space; An adjustment calculation module, configured to calculate the difference between the control target and the 3D model state based on the preprocessed instruction parameters, calculate the importance weight of the control target in the topological structure by using the improved gated graph attention network, adjust the rotation, scaling, displacement, and deformation parameters of the 3D model, optimize the motion trajectory of the 3D model, and adjust the change range of the 3D model shape based on the topological structure; A topological optimization module, configured to perform feature compression on the 3D model topological structure by using the improved variable topology aggregation mechanism, extract local geometric features, and adjust the edge connection mode between adjacent vertices based on the deformation distribution of the motion trajectory, and correct the boundary curvature and grid density of the 3D model; A synchronization and sharing module, configured to output the 3D model to the mobile terminal to implement visual display, and synchronize it to the remote computing platform, support multi-user collaborative interaction, implement multi-person remote sharing control, and record the interaction data; A storage update module for storing the 3D model state, inputting interaction data into a dynamic graph convolutional network, updating the feature learning ability of the 3D model, optimizing the adaptive adjustment strategy, and improving the response accuracy to the real-time environment and the interaction experience.

[0014] According to an embodiment of the present invention, a computer-readable storage medium stores a computer program, which, when executed by a processor, enables the processor to execute a control method for a 3D model of a smart city space.

[0015] The beneficial effects of the present invention are as follows: First, the present invention establishes the temporal dependence relationship of the 3D model through a dynamic graph convolutional network, enabling the model to adaptively adjust based on historical sensor data and current environmental data, improving the real-time performance of model control, enabling the 3D model to accurately respond to changes in the external environment, and ensuring the continuity and stability of the adjustment process.

[0016] Second, by using a gated graph attention network, the present invention optimizes the dynamic distribution of control objectives in the topological structure of the 3D model, enabling precise adjustment of rotation, scaling, displacement, and deformation parameters, avoiding problems such as local structure distortion or control response lag in traditional methods, improving the interaction accuracy, and making the operation of the 3D model smoother and more natural.

[0017] Finally, the present invention adopts a variable topology aggregation mechanism to perform fine-grained optimization on the local structure of the 3D model, enabling the mesh density to remain uniform and the boundary curvature to be smooth during the model deformation process, enhancing the geometric stability and morphological consistency of the 3D model, thereby improving its applicability in complex smart city spaces, achieving high-precision, real-time, and intelligent dynamic adjustment of the 3D model, and improving the digital management and interaction experience of the smart city space. Description of the Drawings

[0018] The 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 drawings: Figure 1 It is a flowchart of the control method for the 3D model of the smart city space proposed by the present invention. Detailed Embodiment

[0019] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only showing the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0020] Refer to Figure 1 , the control method for the 3D model of the smart city space includes the following steps: S1. Obtain the real-time data of the smart city space and preprocess the real-time data; S2. Construct a dynamic graph convolutional network, map the preprocessed image data and point cloud data to a unified feature space, calculate the spatial structure information, initialize the 3D model, and establish the topological structure of the 3D model; S3. Combine the time characteristics of the dynamic graph convolutional network and the preprocessed sensor data, calculate the change rate of the sensor data, update the topological structure and state change trend of the 3D model, and make the 3D model maintain dynamic consistency with the smart city space; S4. Based on the preprocessed instruction parameters, calculate the difference between the control target and the state of the 3D model, use the improved gated graph attention network to calculate the importance weights of the control target in the topological structure, adjust the rotation, scaling, displacement and deformation parameters of the 3D model, optimize the motion trajectory of the 3D model, and adjust the change range of the 3D model's shape based on the topological structure; S5. Use the improved variable topology aggregation mechanism to perform feature compression on the 3D model topological structure, extract local geometric features, and based on the deformation distribution of the motion trajectory, adjust the edge connection method between adjacent vertices, and correct the boundary curvature and mesh density of the 3D model; S6. Output the 3D model to the mobile terminal to achieve visual display, and synchronize it to the remote computing platform to support multi-user collaborative interaction, achieve multi-person remote shared control, and record the interaction data; S7. Store the 3D model state, input the interaction data into the dynamic graph convolutional network, update the feature learning ability of the 3D model, optimize the adaptive adjustment strategy, and improve the response accuracy and interaction experience for the real-time environment.

[0021] In this embodiment, the real-time data includes Internet of Things sensor data, image data collected by a camera, point cloud data obtained by a lidar, and control instructions input by a user. The preprocessing includes performing noise filtering and normalization processing on the sensor data, performing denoising, histogram equalization and feature extraction on the image data, performing downsampling, surface reconstruction and normal vector calculation on the point cloud data, and parsing the control instructions input by the user to extract rotation, scaling, displacement and deformation parameters; In this embodiment, the S2 specifically includes: S21. Based on the preprocessed image data and point cloud data, construct a 3D model topological structure G(V, E), where V is the set of vertices in the 3D model, E is the set of edges between vertices, and initialize the feature vector of each vertex and the feature vector of each edge ; S22. Use the dynamic graph convolutional network to perform topological structure learning on G, calculate the connection relationship between vertices using the dynamic adjacency matrix, and define as the vertex and the associated weight between them, and adjust the adjacency matrix through a normalization operation: ; wherein, is the normalized adjacency matrix, is the adjacency matrix at time t, is 's degree matrix, and ; S23. Update the eigenvector of each vertex. Adopt the dynamic graph convolution calculation method, and perform feature aggregation through the message passing mechanism to calculate the vertex state: ; wherein, is the eigenvector of vertex v at time t + 1, is the training parameter, is the non-linear activation function, N(v) is the neighborhood set of vertex v, is the normalized associated weight between vertex v and u, is the eigenvector of vertex u at time t; S24. Update the topological structure G(V, E) of the 3D model by updating the eigenvector of the vertex, and calculate the spatial structure information: ; wherein, is the local curvature value of the 3D model vertex v at time t + 1, is the weight between vertex v and u, is the curvature value at vertex u, is the curvature value at vertex v.

[0022] In this embodiment, the S3 specifically includes: S31. Define the sensor data set as , wherein represents the data of the i-th sensor at time t, n is the number of sensors, and calculate the state vector of the current time step: ; wherein, is the state vector at time t, is the mapping weight matrix, is the bias term; S32. Calculate the change rate between the state vector of the current time step and the state vector of the previous time step, and define the change rate calculation formula: ; wherein, is the state vector at time t-1, is the rate of change; Take as the input of the dynamic graph convolutional network, adjust the adjacency matrix in the topological structure, and update the adjacency relationship: ; Among them, is the normalized adjacency matrix, is the adjacency matrix at time t-1, is the topological structure adjustment matrix calculated according to , is the degree matrix; S33. Based on the updated topological structure G(V,E), use the dynamic graph convolutional network to calculate the state change trend, and define the vertex feature update formula: ; Among them, is the feature vector of vertex v at time t+1, is the training parameter, is the non-linear activation function, N(v) is the neighborhood set of vertex v, is the normalized correlation weight between vertex v and u, is the feature vector of vertex u at time t, is the bias term; S34. Calculate the spatial position change of the updated vertex features in the 3D model, and define the position update formula of vertex v in the three-dimensional space: ; Among them, is the coordinate position of vertex v at time t+1, is the coordinate position of vertex v at time t, is the step adjustment coefficient, represents the position change amount of vertex v, is the local curvature value of vertex v of the 3D model at time t+1; S35. Adjust the state of the 3D model to keep the 3D model in dynamic consistency with the smart city space.

[0023] In this embodiment, the specific content of S4 includes: S41. Based on the 3D model topological structure G(V,E) and the preprocessed instruction parameters , define the control target set: ; Among them, b is the number of control targets, Denote the $i$-th control target at time $t$, where the control targets include rotation, scaling, displacement, and deformation parameters; S42. Calculate the importance weight of the control target in the topological structure using the improved gated graph attention network, and define the influence weight of the control target on the vertex: ; where, is the control target, is the vertex, is the weight of the control target on the vertex , is the feature vector of the vertex at time step $t$, is the neighborhood set of the vertex , is the natural exponential function with base $e$, is the interaction function between the control target and the vertex feature, defined as: ; where, and are trainable parameter matrices, is the bias term; S43. Based on the influence weight of the control target on the vertex, adjust the rotation, scaling, displacement, and deformation parameters of the 3D model, and define the control update matrix: ; where, is the new position of the vertex at time step $t + 1$, is the original position at time step $t$, and $m$ is the number of control targets; S44. Calculate the adjusted 3D model motion trajectory, and define the trajectory smoothing optimization objective function: ; where, is the trajectory smoothing loss, and respectively represent the new positions of adjacent vertices $v$ and $u$ at time step $t + 1$, $V$ is the set of vertices in the 3D model, $N(v)$ is the neighborhood set of vertex $v$, represents the square of the Euclidean norm; S45. Adjust the morphological change range of the 3D model through the motion trajectory, and calculate the final morphological adjustment function: ; where, is the final morphological adjustment position of vertex $v$ at time step $t + 1$, is the morphological change control coefficient.

[0024] In this embodiment, S5 specifically includes: S51. Based on the 3D model topological structure G(V, E) and the adjacency matrix , form a topological feature matrix by the feature vectors of each vertex at time t: ; wherein, is the vertex feature matrix, a is the number of vertices at time t, represents the feature vector of vertex at time step t; Improve the variable topology aggregation mechanism and adaptively update the topological feature matrix: ; wherein, is the topological feature matrix at time step t, , , , are trainable weight matrices, is 's transpose matrix, is the topological level adjustment parameter, is the adjacency matrix at time t, is the topological adjacency matrix of the k-th layer, and K is the number of topological levels; S52. Based on the topological aggregation matrix, obtain the motion trajectory matrix and calculate the motion deformation influence factor: ; ; wherein, is the motion trajectory matrix at time step t, is the motion trajectory feature of vertex at time step t, a is the number of vertices at time t, is the motion deformation influence factor at time step t, is the neighborhood set of vertex , is the motion deformation control parameter, is the natural exponential function with e as the base, represents the square of the Euclidean norm; S53. Combine the motion trajectory matrix and the motion deformation influence factor, calculate the boundary curvature adjustment matrix and the grid density distribution at time step t, and optimize the grid density: ; ; ; Among them, is the boundary curvature adjustment matrix at time step t, is the degree matrix, is the Laplacian matrix, is the curvature smoothing coefficient, is the mesh density at time step t, is the mesh density control parameter, V is the set of vertices in the 3D model, is the mesh density loss, is the target mesh density; S54. Optimize the topological structure based on the boundary curvature and mesh density, and update the 3D model state: ; Among them, is the 3D model state calculated at time step t + 1, I is the identity matrix, and are the balance coefficients, is the degree matrix at time step t, is the boundary curvature adjustment matrix calculated at time step t, is the optimization step size, is the mesh density gradient descent adjustment term.

[0025] The control device for the 3D model of the smart city space includes: A data processing module, which is used to obtain the real-time data of the smart city space and preprocess the real-time data; A network construction module, which is used to construct a dynamic graph convolutional network, map the preprocessed image data and point cloud data to a unified feature space, calculate the spatial structure information, initialize the 3D model, and establish the topological structure of the 3D model; A state update module, which is used to combine the time characteristics of the dynamic graph convolutional network and the preprocessed sensor data, calculate the change rate of the sensor data, update the topological structure and state change trend of the 3D model, and make the 3D model maintain dynamic consistency with the smart city space; An adjustment calculation module, which is used to calculate the difference between the control target and the 3D model state based on the preprocessed instruction parameters, calculate the importance weight of the control target in the topological structure by using the improved gated graph attention network, adjust the rotation, scaling, displacement and deformation parameters of the 3D model, optimize the motion trajectory of the 3D model, and adjust the change range of the 3D model shape based on the topological structure; A topological optimization module, which is used to perform feature compression on the 3D model topological structure by using the improved variable topology aggregation mechanism, extract local geometric features, and adjust the edge connection mode between adjacent vertices based on the deformation distribution of the motion trajectory, and correct the boundary curvature and mesh density of the 3D model; A synchronous sharing module, which is used to output the 3D model to the mobile terminal, realize visual display, and synchronize it to the remote computing platform, support multi-user collaborative interaction, realize multi-person remote sharing control, and record interaction data; A storage and update module, which is used to store the 3D model state, input the interaction data into the dynamic graph convolutional network, update the feature learning ability of the 3D model, optimize the adaptive adjustment strategy, and improve the response accuracy to the real-time environment and the interaction experience.

[0026] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor can execute the control method of the 3D model of the smart city space.

[0027] Embodiment 1: To verify the feasibility of the present invention in implementation, the present invention is applied to the intelligent manufacturing system of a certain smart industrial park, which includes multiple automated production equipment, industrial robots, Internet of Things sensors, and a 3D visualization monitoring platform. In this park, the 3D model is used to display the equipment operation status, production process flow, and spatial layout in real time, helping managers with remote monitoring, production optimization, and anomaly warning. However, the traditional 3D model control method has serious lag, and cannot adjust the model state in real time according to the operation of the equipment, resulting in information lag and affecting the precise decision-making of intelligent manufacturing. The present invention realizes the intelligent dynamic adjustment of the 3D model through the dynamic graph convolutional network, the gated graph attention network, and the variable topology aggregation mechanism, enabling the 3D model to respond to the changes in the equipment state in real time and improving the operation efficiency and intelligent level of the system.

[0028] In this smart industrial park, each device is equipped with high-precision Internet of Things sensors, including temperature sensors, pressure sensors, vibration sensors, and current monitoring devices. These sensors collect data at an interval of 0.1 second and transmit the data to the central control system through the industrial network. The control method of the traditional 3D model mainly relies on manual adjustment or rule matching. Usually, managers need to manually modify the model parameters according to the sensor data, or use preset rules to automatically execute the adjustment. However, the update frequency of this method is relatively low, usually between 30 seconds and 1 minute, resulting in poor real-time performance of the model and unable to meet the precise control requirements of intelligent manufacturing.

[0029] By constructing a dynamic graph convolutional network, this invention conducts time-series modeling on device operation data and combines a gated graph attention network to calculate the importance weights of device state changes on the topology of the 3D model, achieving precise adjustment of the model. For example, when a certain industrial robot has a joint angle deviation due to a mechanical failure, the dynamic graph convolutional network of this invention can detect the joint deviation based on the abnormal signal of the vibration sensor and complete the adaptive adjustment of the model within 0.5 seconds, enabling the 3D model to accurately reflect the actual state of the robot and avoiding misjudgment and scheduling errors caused by lagging model information. In addition, by adopting a variable topology aggregation mechanism, this invention dynamically optimizes the mesh density and boundary curvature during the model adjustment process, enabling the 3D model to maintain good geometric accuracy and structural stability even under high-frequency updates.

[0030] During the actual testing process, the application scenarios of this invention in this smart industrial park include multiple aspects such as visualization of device operation status, dynamic adjustment of the production process, and abnormal monitoring. To more intuitively demonstrate the advantages of this invention, the following data table compares the performance of this invention and traditional methods in different scenarios of the smart industrial park: Table 1 Comparison data table of 3D model control in the smart industrial park Evaluation Project Traditional Method This Invention Number of Devices 200 units 200 units 3D Model Update Cycle 45 seconds 2.3 seconds Angle Error Range 0.8°-3.5° 0.1°-0.9° Synchronization Failure Rate at Full Load of Devices 17% 1.2% Production Abnormality Detection Time 15 minutes 1.2 seconds Fault Location Time 15 minutes 3 minutes Computing Load 100% 53.5% ; By comparing the traditional method and the method of this invention, the model adjustment speed, accuracy, and computational overhead under different operating states were evaluated. As can be seen from Table 1, when 200 devices are running simultaneously, the update cycle of the traditional 3D model control method is 45 seconds, and the error range is between 0.8° and 3.5°. Moreover, when the device load exceeds 90%, the synchronous update failure rate of the model is as high as 17%. In contrast, the method of this invention shortens the model update cycle to 2.3 seconds, reduces the error range to between 0.1° and 0.9°, and even when the devices are running at full load, the model synchronous failure rate is less than 1.2%.

[0031] Meanwhile, in the production abnormal monitoring scenario, based on the dynamic graph convolutional network and the gated graph attention network, this invention can complete the adjustment of the 3D model within 1.2 seconds when detecting abnormal vibration of production equipment, and calculate possible failure reasons in combination with historical data, enabling managers to locate the abnormal source within 3 minutes, while traditional methods usually require at least 15 minutes of manual analysis to identify the fault point. In addition, the dynamic optimization mechanism of the model significantly improves the computational efficiency, reducing the computational load by approximately 46.5% compared with traditional methods, enabling the 3D model control system to operate more efficiently in the industrial field.

[0032] 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, making equivalent substitutions or changes should be covered within the protection scope of the present invention.

Claims

1. A method for controlling a 3D model of a smart city space, characterized in that: The steps include: S1. Obtain real-time data of smart city space and pre-process the real-time data; S2. Build a dynamic graph convolutional network, map the preprocessed image data and point cloud data to a unified feature space, calculate the spatial structure information, initialize the 3D model, and establish the topological structure of the 3D model; S3, combining the temporal characteristics of the dynamic graph convolutional network and the preprocessed sensor data, calculates the rate of change of the sensor data, updates the topological structure and state change trend of the 3D model, and keeps the 3D model dynamically consistent with the smart city space; S4. Based on the preprocessed instruction parameters, the difference between the control target and the 3D model state is calculated, and the importance weight of the control target in the topological structure is calculated using the improved gated graph attention network, the rotation, scaling, displacement and deformation parameters of the 3D model are adjusted, the motion trajectory of the 3D model is optimized, and the change range of the 3D model morphology is adjusted based on the topological structure; S5. Using the improved variable topology aggregation mechanism, perform feature compression on the topological structure of the 3D model, extract local geometric features, and adjust the edge connection between adjacent vertices based on the deformation distribution of the motion trajectory to correct the boundary curvature and mesh density of the 3D model; S6. Output the 3D model to the mobile terminal for visual display and synchronize it to the remote computing platform to support multi-user collaborative interaction, realize multi-person remote shared control, and record interaction data; S7. Store the 3D model state, input the interaction data into the dynamic graph convolutional network, update the feature learning ability of the 3D model, optimize the adaptive adjustment strategy, and improve the response accuracy and interactive experience to the real-time environment.

2. The control method of the smart city space 3D model according to claim 1 is characterized in that: The real-time data includes IoT sensor data, image data collected by cameras, point cloud data obtained by lidar and control instructions input by users. The preprocessing includes performing noise filtering and normalization on sensor data, performing denoising, histogram equalization and feature extraction on image data, performing downsampling, surface reconstruction and normal vector calculation on point cloud data, parsing the control instructions input by users, and extracting rotation, scaling, displacement and deformation parameters.

3. The control method of the smart city space 3D model according to claim 1 is characterized in that: The S2 specifically includes: S21. Based on the preprocessed image data and point cloud data, construct a 3D model topology structure G(V,E), where V is the vertex set in the 3D model, E is the edge set between the vertices, and initialize the feature vector of each vertex and the eigenvector of each edge ; S22. Use the dynamic graph convolutional network to learn the topological structure of G, use the dynamic adjacency matrix to calculate the connection relationship between vertices, and define Vertex and The association weights between them are calculated and the adjacency matrix is ​​adjusted by normalization: ; in, is the normalized adjacency matrix, is the adjacency matrix at time t, for The degree matrix of ; S23. Feature vector for each vertex Update, use dynamic graph convolution calculation method, perform feature aggregation through message passing mechanism, and calculate vertex status: ; in, is the eigenvector of vertex v at time t+1, is the training parameter, is a nonlinear activation function, N(v) is the neighborhood set of vertex v, is the normalized association weight between vertices v and u, is the eigenvector of vertex u at time t; S24. Update the topological structure G(V,E) of the 3D model by updating the feature vector of the vertex and calculate the spatial structure information: ; in, is the local curvature value of the 3D model vertex v at time t+1, is the weight between vertices v and u, is the curvature value at vertex u, is the curvature value at vertex v.

4. The control method of the smart city space 3D model according to claim 1 is characterized in that: The S3 specifically includes: S31. Define the sensor data set as ,in Represents the data of the i-th sensor at time t, n is the number of sensors, and calculates the state vector of the current time step: ； in, is the state vector at time t, is the mapping weight matrix, is the bias term; S32. Calculate the rate of change between the state vector of the current time step and the state vector of the previous time step, and define the rate of change calculation formula: ； in, is the state vector at time t-1, is the rate of change; Will As the input of the dynamic graph convolutional network, adjust the adjacency matrix in the topology structure and update the adjacency relationship: ； in, is the normalized adjacency matrix, is the adjacency matrix at time t-1, Based on The calculated topology adjustment matrix, is the degree matrix; S33. Based on the updated topological structure G(V,E), the dynamic graph convolutional network is used to calculate the state change trend and define the vertex feature update formula: ； in, is the eigenvector of vertex v at time t+1, is the training parameter, is a nonlinear activation function, N(v) is the neighborhood set of vertex v, is the normalized association weight between vertices v and u, is the eigenvector of vertex u at time t, is the bias term; S34, calculate the spatial position change of the updated vertex feature in the 3D model, and define the position update formula of the vertex v in the three-dimensional space: ; in, is the coordinate position of vertex v at time t+1, is the coordinate position of vertex v at time t, is the step size adjustment coefficient, represents the position change of vertex v, is the local curvature value of the 3D model vertex v at time t+1; S35. Adjust the 3D model status to ensure that the 3D model maintains dynamic consistency with the smart city space.

5. The control method of the smart city space 3D model according to claim 1 is characterized in that: The S4 specifically includes: S41, based on the 3D model topology structure G(V,E) and preprocessed instruction parameters , define the control target set: ; Where b is the number of control targets, represents the i-th control target at time t, and the control target includes rotation, scaling, displacement and deformation parameters; S42. Use the improved gated graph attention network to calculate the importance weight of the control target in the topological structure and define the influence weight of the control target on the vertex: ; in, To control the target, is the vertex, To control the target Opposite Point The weight of Vertex The feature vector at time step t, Vertex The neighborhood set of is the natural exponential function with base e, To control the interaction function between the target and vertex features, it is defined as: ; in, and is the trainable parameter matrix, is the bias term; S43. Based on the influence weight of the control target on the vertex, the rotation, scaling, displacement and deformation parameters of the 3D model are adjusted, and the control update matrix is ​​defined: ; in, Vertex At the new position at time step t+1, is the original position at time step t, and m is the number of controlled targets; S44, calculate the adjusted 3D model motion trajectory, and define the trajectory smoothing optimization objective function: ; in, is the trajectory smoothing loss, and They represent the new positions of adjacent vertices v and u at time step t+1, V is the vertex set in the 3D model, N(v) is the neighborhood set of vertex v, represents the square of the Euclidean norm; S45, adjusting the morphological change range of the 3D model through the motion trajectory, and calculating the final morphological adjustment function: ; in, Adjust the position of vertex v to its final shape at time step t+1, is the morphology change control coefficient.

6. The control method of the smart city space 3D model according to claim 1 is characterized in that: The S5 specifically includes: S51, based on 3D model topology structure G(V,E) and adjacency matrix , the eigenvectors of each vertex at time t form a topological feature matrix: ; in, is the vertex feature matrix, a is the number of vertices at time t, Represents a vertex The feature vector at time step t; Improve the variable topology aggregation mechanism and adaptively update the topology feature matrix: ; in, is the topological feature matrix at time step t, , , , is the trainable weight matrix, for The transposed matrix of Adjust parameters for topological levels, is the adjacency matrix at time t, is the topological adjacency matrix of the kth layer, K is the number of topological levels; S52. Based on the topological aggregation matrix, obtain the motion trajectory matrix and calculate the motion deformation influence factor: ; ; in, is the motion trajectory matrix of time step t, Vertex The motion trajectory characteristics at time step t, a is the number of vertices at time t, is the motion deformation influence factor at time step t, Vertex The neighborhood set of is the motion deformation control parameter, is the natural exponential function with base e, represents the square of the Euclidean norm; S53. Combine the motion trajectory matrix and the motion deformation influencing factor to calculate the boundary curvature adjustment matrix and mesh density distribution at time step t, and optimize the mesh density: ; ; ; in, is the boundary curvature adjustment matrix at time step t, is the degree matrix, is the Laplace matrix, is the curvature smoothing coefficient, is the grid density at time step t, is the mesh density control parameter, V is the vertex set in the 3D model, is the mesh density loss, is the target grid density; S54. Optimize the topology structure based on boundary curvature and mesh density, and update the 3D model status: ; in, is the 3D model state calculated at time step t+1, I is the unit matrix, and is the balance coefficient, is the degree matrix at time step t, is the boundary curvature adjustment matrix calculated at time step t, To optimize the step size, is the grid density gradient descent adjustment term.

7. A control device for a smart city space 3D model, executing a control method for a smart city space 3D model according to any one of claims 1 to 6, characterized in that: include: Data processing module, used to obtain real-time data of smart city space and pre-process the real-time data; The network construction module is used to construct a dynamic graph convolutional network, map the preprocessed image data and point cloud data to a unified feature space, calculate the spatial structure information, initialize the 3D model, and establish the topological structure of the 3D model; The state update module is used to combine the time characteristics of the dynamic graph convolutional network and the preprocessed sensor data, calculate the change rate of the sensor data, update the topological structure and state change trend of the 3D model, and keep the 3D model dynamically consistent with the smart city space; An adjustment calculation module is used to calculate the difference between the control target and the 3D model state based on the preprocessed instruction parameters, calculate the importance weight of the control target in the topological structure using the improved gated graph attention network, adjust the rotation, scaling, displacement and deformation parameters of the 3D model, optimize the motion trajectory of the 3D model, and adjust the range of change of the 3D model morphology based on the topological structure; The topology optimization module is used to perform feature compression on the topological structure of the 3D model using an improved variable topology aggregation mechanism, extract local geometric features, and adjust the edge connection method between adjacent vertices based on the deformation distribution of the motion trajectory to correct the boundary curvature and mesh density of the 3D model; The synchronization and sharing module is used to output the 3D model to the mobile terminal for visual display and synchronize it to the remote computing platform, support multi-user collaborative interaction, realize multi-person remote sharing control, and record interaction data; The storage update module is used to store the 3D model state, input the interaction data into the dynamic graph convolutional network, update the feature learning ability of the 3D model, optimize the adaptive adjustment strategy, and improve the response accuracy and interactive experience to the real-time environment.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor is enabled to execute the control method of the smart city space 3D model as described in any one of claims 1 to 6.

Images