A smart city data visualization method and system based on multi-source data fusion

Abstract

The application discloses a kind of based on multi-source data fusion's wisdom city data visualization method and system, belong to wisdom city technical field, by weight isolated forest and pan-Kriging algorithm preprocessing data;With the improved D-S evidence theory resolves conflict across source;Adopt DQN reinforcement learning in perspective and power allocation under dynamic adjustment fusion strategy;Combining with the viewpoint depth driven differential privacy and attribute-based encryption realizes accurate desensitization;Based on view cone rejection, dynamic LOD and cloud-terminal off-screen rendering technology completes multidimensional linkage display;Through PPO algorithm collection user feedback realizes closed-loop iterative optimization.The present application solves the problem of heterogeneous data conflict and high dependence of three-dimensional rendering on hardware, while ensuring the fidelity of situation analysis and individual privacy, significantly improves the rendering fluency of city-level massive data and the system evolution ability.

Description

Technical Field

[0001] This invention belongs to the field of smart city technology, and specifically relates to a smart city data visualization method and system based on multi-source data fusion. Background Technology

[0002] With the deep integration of new urbanization and the digital economy, smart city construction has become a core direction for the modernization of urban governance. During the operation of smart cities, massive amounts of heterogeneous data from multiple dimensions and structures, including IoT sensing devices, government business systems, video surveillance, internet public opinion, and geographic information systems, are generated. How to efficiently integrate this multi-source data and intuitively present the city's operational status through visualization technology is a core requirement of smart city governance.

[0003] Smart city data is highly heterogeneous, with inconsistent spatiotemporal benchmarks, data formats, and semantic standards across different data sources. There are also numerous cross-source data conflicts. Existing fusion methods are mostly static offline fusions, which cannot adapt to the dynamic changes of real-time data streams. This results in high fusion latency, low data reliability, and insufficient accuracy in multi-source data fusion, directly leading to distortion of visualization content.

[0004] In existing solutions, data fusion and visualization rendering are two independent steps. The fusion process does not consider the actual needs of the visualization scene and often performs high-precision fusion of all data, resulting in a large waste of computing resources. In large-scale 3D visualization scenes and on low-performance terminals, rendering stutters and excessive latency are very likely to occur, resulting in serious waste of computing resources and a poor user experience.

[0005] Smart city data contains a large amount of sensitive government data and citizens' personal information. Existing solutions mostly encrypt data only at the data storage stage. In the entire chain of data integration and visualization interaction, there is a lack of fine-grained privacy protection and access control, resulting in a high risk of sensitive data leakage and failing to meet the relevant requirements of the Data Security Law.

[0006] Existing visualization solutions mostly display indicators in a single dimension, which cannot achieve interconnected drill-down of spatiotemporal, business, and event dimensions. Users cannot complete a full-dimensional analysis of the city's operational status from macro to micro. Due to the lack of multi-dimensional linkage, the practical value of visualization is limited and cannot support refined decision-making in urban governance.

[0007] In summary, there is an urgent need to provide an optimized smart city data visualization solution to address the core technical challenges of dynamic fusion and visualization scenario adaptation, end-to-end privacy protection, and multi-dimensional interactive linkage, so as to meet the real-time, highly secure, and highly interactive visualization requirements of smart city refined governance.

[0008] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0009] The purpose of this invention is to provide a smart city data visualization method and system based on multi-source data fusion, which can overcome the defects of existing technologies such as low fusion accuracy, poor dynamic adaptability, disconnect between fusion and rendering, and weak interactive capabilities. It can achieve high-precision real-time fusion and highly adaptable visualization of multi-source data in smart cities, and improve the credibility, fluency, security and practicality of smart city visualization.

[0010] To achieve the above objectives, this invention provides a smart city data visualization method based on multi-source data fusion, comprising the following steps: Step S1: Collect multi-dimensional heterogeneous raw data of smart city, perform data cleaning, format standardization and outlier removal on the raw data to obtain a preprocessed dataset; Step S2: Align the preprocessed dataset with a unified spatiotemporal reference to generate a spatiotemporally standardized dataset. Based on the improved DS evidence theory, perform weight allocation and conflict resolution on the cross-source conflict data in the spatiotemporally standardized dataset to obtain a trustworthy fusion dataset. Step S3: Obtain the scene features, user interaction commands and terminal computing power parameters of the current visualization scene. Based on the preset scene-policy mapping model, dynamically adjust the granularity, dimension and computing priority of data fusion to generate a target fusion dataset adapted to the scene. Step S4: Combine the rendering layer depth of the target fusion dataset in the current visualization scene, dynamically allocate differential privacy budgets to different layers, and inject Laplacian noise; further combine the attribute base encryption algorithm to generate a visualization permission dataset that matches the user's access permissions; Step S5: Based on the aforementioned visualization permission dataset, construct a three-dimensional visualization scene with spatiotemporal-business-event linkage. Render the screen using a distributed cloud-terminal collaborative rendering engine based on view frustum and LOD (Level of Detail). Simultaneously, respond to user drill-down, linkage, and filtering interaction commands, and update the visualization rendering content in real time.

[0011] Optionally, in step S1, the collected multi-dimensional heterogeneous raw data of the smart city includes at least: real-time monitoring data collected by IoT sensing devices, structured business data from government business systems, unstructured video stream data from video surveillance equipment, internet public opinion data, and geospatial information data; the preprocessing specifically includes: Step S11: To address the format differences between different data sources, construct a unified semantic mapping dictionary to map the field names, data types, and units of heterogeneous raw data to a unified semantic standard. Step S12: Use a combination of the 3σ criterion and the isolated forest algorithm to remove outliers and noisy data from the original data; Step S13: For missing data, use the Universal Kriging interpolation algorithm that incorporates spatiotemporal distance decay factors to complete the data and obtain the preprocessed dataset.

[0012] Optionally, step S2 specifically includes: Step S21: Construct a unified spatiotemporal reference system, using the WGS84 coordinate system as the spatial reference and the UTC+8 time zone as the time reference, and convert the spatial coordinates and timestamps of all data in the preprocessed dataset to the unified spatiotemporal reference. Step S22: For cross-source data at the same spatiotemporal location, calculate the time decay weight, spatial matching weight, and historical reliability weight of each data source to obtain the comprehensive weight of each data source. Step S23: Obtain the basic probability allocation (BPA) function for a certain fusion event from each data source. When calculating the conflict coefficient between pairs of evidence sources, introduce the Jousselme evidence distance to quantify the similarity between highly conflicting evidence. Step S24: When the conflict coefficient is greater than the preset conflict tolerance coefficient, the comprehensive weight is used as the absolute credibility conversion factor to modify the BPA function. The probability quality of the conflict part is dynamically redistributed to the specified proposition according to the comprehensive weight ratio of each data source, instead of all being allocated to the unknown set, to generate a credible fusion dataset.

[0013] Optionally, in step S22, the formula for calculating the time decay weight is:

[0014] in, For time decay weight, The preset attenuation coefficient, This is the time difference between the data collection time and the current time. The formula for calculating the spatial matching degree weight is:

[0015] in, For spatial matching degree weight, The preset spatial attenuation coefficient, The Euclidean distance between the data acquisition location and the target spatiotemporal location; The historical reliability weight is dynamically updated based on the historical data accuracy of the data source, with an initial value of 1.

[0016] Optionally, step S3 specifically includes: Step S31: Extract scene features of the current visualization scene, including scene type, rendering range, and data dimension requirements; obtain user interaction commands, including zoom, pan, drill-down, and filter operations; collect computing power parameters of the current terminal, including CPU / GPU utilization and network bandwidth. Step S32: Input scene features, user interaction commands, and computing power parameters into the pre-trained scene-policy mapping model, and output the optimal fusion strategy. The optimal fusion strategy includes at least the temporal granularity, spatial granularity, business dimension, and computational priority of data fusion. The state space S of the pre-trained scene-policy mapping model... t Represented as a state vector: In the formula: L view R represents the height hierarchy parameter of the camera in the current visualization scene. fov U represents the geographical area covered by the camera's field of view. cpu U gpu These are the real-time CPU load rate and real-time GPU load rate of the visualization terminal, respectively; B net For the terminal's real-time network downlink bandwidth; I op Encoding vectors for multi-dimensional user interaction commands; The action space A of the model t Represented as an action vector: A t =[Δg s ,Δg t ,P cal ], where: Δg s Δg is the adjustment amount for the aggregation or refinement order of the spatial grid fusion granularity; t P is the step size adjustment value for time fusion slices; cal This refers to the cloud computing power offloading ratio and rendering pipeline priority parameters corresponding to different business data within the current field of view; Step S33: According to the optimal fusion strategy, perform secondary fusion processing on the trusted fusion dataset to generate a target fusion dataset that is adapted to the scenario. Step S34: Monitor the frame rate and latency of the visualization rendering in real time. When the frame rate is lower than the preset threshold or the latency is higher than the preset threshold, dynamically adjust the fusion strategy to reduce the fusion granularity of non-core data and ensure smooth rendering.

[0017] Optionally, the scene-policy mapping model is a deep Q-network model based on reinforcement learning, and its training process includes: Using scene rendering smoothness, data fidelity, and computational efficiency as reward functions, and historical visualized scene data, corresponding fusion strategies, and rendering effects as the training set, the deep Q-network model is iteratively trained until convergence. The reward function R of the deep Q-network model is calculated using the following formula: Among them, F real For real-time rendering frame rate of the visualization scene; F target F min These are the preset target rendering frame rate and the minimum tolerable rendering frame rate, respectively. Dorg D fuse These are the information entropy of the original high-precision business data and the information entropy of the fused data, respectively; C use C max These represent the real-time computing power usage and total computing power throughput of the visualization terminal, respectively; θ is the preset terminal computing power warning threshold; α, β, and γ are preset weight hyperparameters, respectively.

[0018] Optionally, step S4 specifically includes: Step S41: Identify sensitive data fields in the target fusion dataset, add differential privacy noise to the sensitive data fields based on the Laplace mechanism, and complete the data desensitization process; Step S42: Based on the attribute-based encryption algorithm, generate corresponding attribute private keys for different users according to their roles, departments, and access permissions; Step S43: Perform layered encryption on the desensitized target fusion dataset to generate a visualization permission dataset that matches different access permissions. The visualization data of the corresponding level can only be decrypted when the user's attribute private key meets the access policy.

[0019] Optionally, step S5 specifically includes: Step S51: Based on the visual permission dataset, construct a four-level visualization hierarchy system including a macro overview layer, a regional monitoring layer, a single-entity details layer, and an event tracing layer; Step S52: Establish a linkage mapping relationship between the spatiotemporal dimension, business dimension, and event dimension. When a user performs filtering or drill-down operations on the data in any dimension, the visualization content of the other two dimensions is updated synchronously. Step S53: Using a distributed rendering engine, the rendering tasks are distributed to cloud rendering nodes and local terminal nodes. The rendering resolution and effect level are adapted according to the terminal performance to complete the visualization rendering adapted to multiple terminals. Step S54: Respond to user interaction commands in real time, and update data fusion strategies and rendering content synchronously to achieve real-time linkage between interaction and visualization.

[0020] Optionally, after step S5, the following may also be included: Step S6: Iterative optimization of visualization effects and data fusion quality: Collect user interaction behavior data, visualization rendering effect data, and data fusion accuracy data. Based on user feedback and running data, iteratively optimize the scene-policy mapping model, conflict resolution algorithm, and rendering engine parameters.

[0021] A second aspect of the present invention also provides a smart city data visualization system based on multi-source data fusion, comprising the following modules: The data preprocessing module is configured to collect multi-dimensional heterogeneous raw data from smart cities, perform data cleaning, format standardization, and outlier removal on the raw data, and generate a preprocessed dataset. The spatiotemporal benchmark alignment module is connected to the data preprocessing module. It is used to perform unified spatiotemporal benchmark alignment on the preprocessed dataset, generate a spatiotemporally standardized dataset, and perform weight allocation and conflict resolution on cross-source conflict data based on the improved DS evidence theory, and output a credible fusion dataset. The fusion strategy adjustment module is connected to the spatiotemporal reference alignment module. By acquiring the scene features, user interaction commands and terminal computing power parameters of the current visualization scene, it dynamically adjusts the granularity, dimension and computing priority of data fusion based on the preset scene-strategy mapping model to generate a target fusion dataset adapted to the scene. The privacy management module is connected to the fusion strategy adjustment module, performs differential privacy desensitization processing on the target fusion dataset, and generates a visual permission dataset that matches user access permissions based on the attribute base encryption algorithm. The rendering interaction module is connected to the privacy control module. It constructs a three-dimensional interactive visualization scene based on the visualization permission dataset, which integrates time, space, business, and events. It completes multi-terminal adapted screen rendering through a distributed rendering engine and updates the rendered content in real time in response to the user's drill-down, linkage, and filtering interaction commands.

[0022] Compared with existing technologies, the smart city data visualization method and system based on multi-source data fusion according to the present invention has the following advantages or beneficial effects: This invention normalizes cross-source heterogeneous data through a unified spatiotemporal benchmark alignment technique, eliminating the spatiotemporal misalignment problem of multi-source data. Building upon this, it overcomes the inherent defect of distorted fusion results when processing highly conflicting cross-source data by improving the conflict resolution mechanism of the DS evidence theory. Simultaneously, through a dynamic weight allocation mechanism based on time decay, spatial matching, and historical credibility, it adaptively assigns differentiated weights to data sources with different collection times, spatial coverage, and historical data precision, effectively suppressing the interference of low-quality, highly conflicting data on the fusion results. Based on the above progressive technical solutions, high-precision dynamic fusion of multi-source heterogeneous data for smart cities is achieved, solving the core technical problems of low credibility of fusion results and inability to adapt to real-time data streams caused by spatiotemporal mismatch of cross-source data and severe evidence conflicts in existing technologies.

[0023] This invention proposes a visualization scene-driven adaptive adjustment mechanism for fusion strategies, constructing a closed-loop linkage logic between fusion computing and visualization rendering. Specifically, through a reinforcement learning-based Deep Q-Network (DQN) model, the system takes the visualization camera height level, view frustum coverage, and terminal CPU / GPU real-time load as state inputs, and spatial grid fusion granularity, temporal fusion slice step size, and cloud computing power offloading ratio as action outputs to construct a reward function. This allows the model to autonomously learn and output the optimal fusion strategy for different scenarios. Based on this technical solution, an on-demand fusion mode is achieved, resolving the contradiction between high precision and high smoothness caused by fixed fusion strategies. This improves the utilization of computing resources and the real-time rendering frame rate, significantly expanding the system's terminal adaptability boundaries.

[0024] In the fusion computing stage, this invention uses differential privacy technology to add Laplace noise to sensitive location data and business statistics data, conforming to privacy budget constraints. This achieves irreversible desensitization of sensitive data without affecting the statistical validity of the fused data, preventing the leakage of sensitive information during the fusion process. In the rendering and distribution stage, attribute-based encryption (ABE) technology is used to perform layered encryption on the visualization data layers, eliminating the need for separate encryption and distribution for users with different permissions. Based on these technical solutions, the risk of sensitive data leakage in the smart city data visualization process is blocked from the entire chain. This solves the technical problems of fragmented privacy protection and insufficient granularity of permission control in existing technologies, achieving compliant sharing of visualization data while ensuring data security.

[0025] This invention constructs a four-level visualization hierarchy system comprising a macro-level overview layer, a regional monitoring layer, a unit detail layer, and an event tracing layer. It establishes a linkage mapping relationship between spatiotemporal, business, and event dimensions, enabling synchronous linkage and automatic adaptation of corresponding business data, event data, and spatiotemporal scope when switching visualization levels. This breaks down the technical barriers of the separation between spatiotemporal, business, and event dimensions in existing smart city visualization, achieving full-dimensional visualization from a macro-level overview of city operations to micro-level event tracing. Users can perform multi-dimensional analysis of the city's operational status through the visualization interface, enhancing the decision-making support capabilities and practical value of the smart city visualization system. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating a smart city data visualization method based on multi-source data fusion according to an embodiment of the present invention. Figure 2 This is a flowchart of step S2 in a smart city data visualization method based on multi-source data fusion according to an embodiment of the present invention. Figure 3 This is a flowchart of step S3 in a smart city data visualization method based on multi-source data fusion according to an embodiment of the present invention. Figure 4 This is a flowchart of step S4 in a smart city data visualization method based on multi-source data fusion according to an embodiment of the present invention. Figure 5 This is a flowchart of step S5 in a smart city data visualization method based on multi-source data fusion according to an embodiment of the present invention. Figure 6 This is a flowchart of step S53 in a smart city data visualization method based on multi-source data fusion according to an embodiment of the present invention. Figure 7 This is a structural block diagram of a smart city data visualization system based on multi-source data fusion according to an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein.

[0029] It should be understood that in the various embodiments of the present invention, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0030] It should be understood that in this invention, "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0031] It should be understood that in this invention, "multiple" refers to two or more. "And / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, "and / or B" can represent: A existing alone, A and B existing simultaneously, and B existing alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "Contains A, B, and C", "Contains A, B, and C" means that all three A, B, and C are contained; "Contains A, B, or C" means that one of A, B, and C is contained; "Contains A, B, and / or C" means that any one, two, or three of A, B, and C are contained.

[0032] It should be understood that in this invention, "B corresponding to A", "B corresponding to A", "A and B correspond", or "B and A correspond" means that B is associated with A, and B can be determined based on A. Determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information. Matching A and B is defined as a similarity between A and B that is greater than or equal to a preset threshold.

[0033] Depending on the context, "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection."

[0034] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0035] like Figure 1 As shown, the smart city data visualization method based on multi-source data fusion according to a preferred embodiment of the present invention includes the following steps: Step S1: Collect multi-dimensional heterogeneous raw data of smart city, perform data cleaning, format standardization and outlier removal on the raw data to obtain a preprocessed dataset; In this embodiment of the invention, the multi-dimensional heterogeneous raw data of smart cities collected in step S1 includes at least: real-time monitoring data collected by IoT sensing devices, structured business data from government business systems, unstructured video stream data from video surveillance devices, internet public opinion data, and geospatial information data; the preprocessing specifically includes: Step S11: To address the format differences between different data sources, construct a unified semantic mapping dictionary to map the field names, data types, and units of heterogeneous raw data to a unified semantic standard. Step S12: Use a combination of the 3σ criterion and the isolated forest algorithm to remove outliers and noisy data from the original data; Step S13: For missing data, use the Universal Kriging interpolation algorithm that incorporates spatiotemporal distance decay factors to complete the data and obtain the preprocessed dataset.

[0036] This invention introduces an isolated forest with added dimensional correlation weights, which can discover "contextual anomalies" in high-dimensional space that appear normal in a single dimension but violate physical laws in combined dimensions. Furthermore, the pan-kriging algorithm incorporates a time decay window, breaking the limitations of traditional interpolation that relies solely on spatial distance. This ensures the accuracy of repairing dynamic urban data in both spatiotemporal dimensions, improves the reliability of the underlying data, and avoids visualization distortion caused by "garbage in, garbage out."

[0037] Step S2: Align the preprocessed dataset with a unified spatiotemporal reference to generate a spatiotemporally standardized dataset. Based on the improved DS evidence theory, perform weight allocation and conflict resolution on the cross-source conflict data in the spatiotemporally standardized dataset to obtain a trustworthy fusion dataset. In embodiments of the present invention, such as Figure 2 As shown, step S2 specifically includes: Step S21: Construct a unified spatiotemporal reference system, using the WGS84 coordinate system as the spatial reference and the UTC+8 time zone as the time reference, and convert the spatial coordinates and timestamps of all data in the preprocessed dataset to the unified spatiotemporal reference. Step S22: For cross-source data at the same spatiotemporal location, calculate the time decay weight, spatial matching weight, and historical reliability weight of each data source to obtain the comprehensive weight of each data source; in step S22, the formula for calculating the time decay weight is:

[0038] in, For time decay weight, The preset attenuation coefficient, This is the time difference between the data collection time and the current time. The formula for calculating the spatial matching degree weight is:

[0039] in, For spatial matching degree weight, The preset spatial attenuation coefficient, The Euclidean distance between the data acquisition location and the target spatiotemporal location; The historical reliability weight is dynamically updated based on the historical data accuracy of the data source, with an initial value of 1.

[0040] Step S23: Obtain the basic probability allocation (BPA) function for a certain fusion event from each data source. When calculating the conflict coefficient between pairs of evidence sources, introduce the Jousselme evidence distance to quantify the similarity between highly conflicting evidence. Step S24: When the conflict coefficient is greater than the preset conflict tolerance coefficient, the comprehensive weight is used as the absolute credibility conversion factor to modify the BPA function. The probability quality of the conflict part is dynamically redistributed to the specified proposition according to the comprehensive weight ratio of each data source, instead of all being allocated to the unknown set, to generate a credible fusion dataset.

[0041] It should be noted that this invention introduces "Jousselme evidence distance" and "comprehensive weighting based on spatiotemporal and historical accuracy." When the conclusions given by sensors are completely contradictory, the system no longer blindly assigns probabilities to the "uncertain set." Instead, it forces a skewed redistribution of probabilities based on which sensor is closer, more recent, and has a higher historical credit score. This solves the pain point of system paralysis caused by conflicting cross-source data, ensuring the uniqueness and high reliability of the fusion judgment.

[0042] Step S3: Obtain the scene features, user interaction commands and terminal computing power parameters of the current visualization scene. Based on the preset scene-policy mapping model, dynamically adjust the granularity, dimension and computing priority of data fusion to generate a target fusion dataset adapted to the scene. In embodiments of the present invention, such as Figure 3 As shown, step S3 specifically includes: Step S31: Extract scene features of the current visualization scene, including scene type, rendering range, and data dimension requirements; obtain user interaction commands, including zoom, pan, drill-down, and filter operations; collect computing power parameters of the current terminal, including CPU / GPU utilization and network bandwidth. Step S32: Input scene features, user interaction commands, and computing power parameters into the pre-trained scene-policy mapping model, and output the optimal fusion strategy. The optimal fusion strategy includes at least the temporal granularity, spatial granularity, business dimension, and computational priority of data fusion. The state space S of the pre-trained scene-policy mapping model... t Represented as a state vector: In the formula: L view R represents the height hierarchy parameter of the camera in the current visualization scene. fov U represents the geographical area covered by the camera's field of view cone. cpu U gpu These are the real-time CPU load rate and real-time GPU load rate of the visualization terminal, respectively. net For the terminal's real-time network downlink bandwidth, I op Encoding vectors for multi-dimensional user interaction commands; The action space A of the model t Represented as an action vector: A t =[Δg s ,Δg t ,P cal ], where: Δg s Δg is the adjustment amount for the aggregation or refinement order of the spatial grid fusion granularity; t P is the step size adjustment value for time fusion slices; cal This refers to the cloud computing power offloading ratio and rendering pipeline priority parameters corresponding to different business data within the current field of view; Step S33: According to the optimal fusion strategy, perform secondary fusion processing on the trusted fusion dataset to generate a target fusion dataset that is adapted to the scenario. Step S34: Monitor the frame rate and latency of the visualization rendering in real time. When the frame rate is lower than the preset threshold or the latency is higher than the preset threshold, dynamically adjust the fusion strategy to reduce the fusion granularity of non-core data and ensure smooth rendering.

[0043] This invention constructs an AI intelligent scheduling system based on the DQN algorithm, directly using computer graphics parameters and terminal hardware status as input for deep reinforcement learning. When a user zooms out to view the entire city, the AI ​​instantly instructs the backend to perform coarse-grained spatial aggregation, discarding micro-details to maintain frame rate; when a user zooms in on a specific neighborhood, the AI ​​instantly refines the granularity to the building level and reduces the offloading ratio of computing power. This mechanism achieves millisecond-level dynamic handshake between data granularity and rendering computing power, supporting smooth roaming of massive city-level concurrent data with minimal hardware cost.

[0044] In this embodiment of the invention, the scene-policy mapping model is a deep Q-network model based on reinforcement learning, and its training process includes: Using scene rendering smoothness, data fidelity, and computational efficiency as reward functions, and historical visualized scene data, corresponding fusion strategies, and rendering effects as the training set, the deep Q-network model is iteratively trained until convergence. The reward function R of the deep Q-network model is calculated using the following formula: ; Among them, F real For real-time rendering frame rate of the visualization scene; F target F min These are the preset target rendering frame rate and the minimum tolerable rendering frame rate, respectively; D org D fuse These are the information entropy of the original high-precision business data and the information entropy of the fused data, respectively; C use C max These represent the real-time computing power usage and total computing power throughput of the visualization terminal, respectively; θ is the preset terminal computing power warning threshold; α, β, and γ are preset weight hyperparameters, respectively.

[0045] Step S4: Combine the rendering layer depth of the target fusion dataset in the current visualization scene, dynamically allocate differential privacy budgets to different layers, and inject Laplacian noise; further combine the attribute base encryption algorithm to generate a visualization permission dataset that matches the user's access permissions; In embodiments of the present invention, such as Figure 4 As shown, step S4 specifically includes: Step S41: Identify sensitive data fields in the target fusion dataset, add differential privacy noise to the sensitive data fields based on the Laplace mechanism, and complete the data desensitization process; Step S42: Based on the attribute-based encryption algorithm, generate corresponding attribute private keys for different users according to their roles, departments, and access permissions; Step S43: Perform layered encryption on the desensitized target fusion dataset to generate a visualization permission dataset that matches different access permissions. The visualization data of the corresponding level can only be decrypted when the user's attribute private key meets the access policy.

[0046] This step pioneers a viewpoint-depth-driven differential privacy control mechanism. Traditional anonymization methods either render data worthless through a one-size-fits-all approach or fail to encrypt data, leading to leakage risks. This application uses the distance of the visualization lens as the mathematical variable for differential privacy budget allocation: when administrators are viewing the overall situation of the city, a large privacy budget is allocated, injecting minimal noise to ensure the absolute accuracy of macro-statistical big data; when users attempt to drill down to examine sensitive cases in a particular building, the privacy budget decreases abruptly without a non-linear timeframe, injecting significant Laplace noise, ensuring that they only see blurred or statistically biased data, fundamentally preventing attackers from reverse-engineering the absolute privacy of specific citizens or buildings through the front-end interface. This approach balances the fidelity analysis of macro-level big data with strong privacy protection for micro-level individuals.

[0047] Step S5: Based on the aforementioned visualization permission dataset, construct a three-dimensional visualization scene with spatiotemporal-business-event linkage. Render the screen using a distributed cloud-terminal collaborative rendering engine based on view frustum and LOD (Level of Detail). Simultaneously, respond to user drill-down, linkage, and filtering interaction commands, and update the visualization rendering content in real time.

[0048] In embodiments of the present invention, such as Figure 5 As shown, step S5 specifically includes: Step S51: Based on the visualization permission dataset, construct a four-level visualization hierarchy system including a macro overview layer, a regional monitoring layer, an individual detail layer, and an event tracing layer. The macro overview layer is used to display city-level situation information; the regional monitoring layer focuses on thematic data of specific administrative regions or grids; the individual detail layer delves into the refined models of individual buildings, equipment, etc.; and the event tracing layer combines a timeline to display the development and evolution of events.

[0049] Step S52: Establish a linkage mapping relationship between the spatiotemporal dimension, business dimension, and event dimension. When a user performs filtering or drill-down operations on data in any dimension, the visualization content of the other two dimensions is updated synchronously. When a user performs filtering or drill-down operations on data in any dimension through the interactive interface, the system queries the index matrix to synchronously update the visualization filtering conditions and rendering screen of the other two dimensions, thereby realizing the real-time correlation display of multi-dimensional data.

[0050] Step S53: Employing a distributed rendering engine, rendering tasks are allocated to cloud rendering nodes and local terminal nodes. The rendering resolution and effects level are adapted to the terminal's performance to complete multi-terminal adapted visual rendering. It should be noted that, as... Figure 6 As shown, the distributed multi-terminal adaptation visualization rendering process in step S53 includes: Step S53a: Based on the projection matrix of the viewpoint camera, extract the view frustum of the current image, remove the data bounding boxes outside the view frustum, and reduce the rendering overhead of invalid data outside the field of view. Step S53b: For the service data within the view frustum, calculate the percentage of the pixel projection area of ​​the corresponding 3D bounding sphere or bounding box in the screen space at this moment. Step S53c: When the pixel projection area ratio is lower than the first threshold, it automatically degenerates into point cloud or two-dimensional billboard material rendering; when it is higher than the second threshold, a high-precision luminous geometry containing complete skeleton and texture is loaded, and the LOD level is improved. In step S53d, for the area monitoring layer and individual detail layer that require high-precision rendering, the system divides their rendering load: the basic geometric mesh is rendered on the terminal GPU through WebGL / engine, while high-performance effects such as volumetric light and high dynamic range (HDR) fluid simulation are rendered off-screen in the cloud and then mapped to the screen coordinates of the corresponding terminal as a video stream, thus completing the cloud-edge-device rendering fusion.

[0051] Step S54: Respond to user interaction commands in real time, and update data fusion strategies and rendering content synchronously to achieve real-time linkage between interaction and visualization.

[0052] In this embodiment of the invention, step S5 is followed by: Step S6: Iterative Optimization of Visualization Effects and Data Fusion Quality: Collect user interaction behavior data, visualization rendering effect data, and data fusion accuracy data. Based on user feedback and operational data, iteratively optimize the scene-policy mapping model, conflict resolution algorithm, and rendering engine parameters. Interaction behavior data includes dwell time at different levels, click heatmap distribution, and roaming path habits. Visualization rendering effect data includes GPU utilization fluctuations, real-time rendering frame rate (FPS) distribution, and first-screen loading latency. Data fusion accuracy data includes conflict allocation tendency and BPA function entropy. The operational data is uploaded to the experience replay pool on the cloud server via asynchronous communication. The cloud optimization engine uses the Proximal Policy Optimization (PPO) algorithm to fine-tune the reinforcement learning model in Step S3 offline. The PPO algorithm introduces a Clipped Probability Ratio objective function to correct the action value assessment in the DQN model using the latest real user behavior data, while ensuring policy update stability. Based on user fine-tuning or feedback on conflict resolution results, the historical credibility weights in Step S2 are dynamically corrected. By calculating the weighted cross-entropy loss, the preset parameters of the conflict redistribution matrix are optimized, making the system more consistent with the expert logic of specific business scenarios in subsequent fusion. The optimized model weights are periodically distributed to the terminal, enabling automatic iteration and silent upgrading of visualization rendering parameters and fusion strategies.

[0053] This invention also provides an embodiment of a smart city data visualization system based on multi-source data fusion, specifically as follows: Figure 7 As shown, it includes the following modules: The data preprocessing module is configured to collect multi-dimensional heterogeneous raw data from smart cities, perform data cleaning, format standardization, and outlier removal on the raw data, and generate a preprocessed dataset. The spatiotemporal benchmark alignment module is connected to the data preprocessing module. It is used to perform unified spatiotemporal benchmark alignment on the preprocessed dataset, generate a spatiotemporally standardized dataset, and perform weight allocation and conflict resolution on cross-source conflict data based on the improved DS evidence theory, and output a credible fusion dataset. The fusion strategy adjustment module is connected to the spatiotemporal reference alignment module. By acquiring the scene features, user interaction instructions and terminal computing power parameters of the current visualization scene, it dynamically adjusts the granularity, dimension and computing priority of data fusion based on the preset scene-strategy mapping model to generate a target fusion dataset adapted to the scene. The privacy management module is connected to the fusion strategy adjustment module, performs differential privacy desensitization processing on the target fusion dataset, and generates a visual permission dataset that matches user access permissions based on the attribute base encryption algorithm. The rendering interaction module is connected to the privacy control module. It constructs a three-dimensional interactive visualization scene based on the visualization permission dataset, which integrates time, space, business, and events. It completes multi-terminal adapted screen rendering through a distributed rendering engine and updates the rendered content in real time in response to the user's drill-down, linkage, and filtering interaction commands.

[0054] It should be noted that the information interaction and execution process between the above modules are based on the same concept as the method embodiments of this application. It is a system corresponding to the smart city data visualization method based on multi-source data fusion. All implementation methods in the above method embodiments are applicable to the embodiments of this system. For details on its specific functions and the technical effects it brings, please refer to the method embodiment section. It will not be repeated here.

[0055] In summary, this invention effectively solves the problems of cross-source data conflict and dynamic missing data by improving the DS evidence theory and spatiotemporal variation interpolation algorithm, ensuring the authenticity of the visualization presentation base and providing high-confidence data support for urban decision-making. Based on the "cloud-terminal" distributed rendering architecture and multi-precision LOD automatic degradation conversion mechanism, it can present cinematic visual effects in a TB-level urban background in real time with low local computing power overhead. The pioneering viewpoint depth-driven differential privacy mechanism achieves a dynamic authorization effect of "fidelity from afar, blurriness up close," ensuring the accuracy of macro-situational analysis while eliminating the risk of privacy leakage of micro-level individuals from the underlying mechanism. Furthermore, by constructing an asynchronous closed-loop feedback system based on the PPO algorithm, the system has the self-learning ability to continuously optimize according to user habits and business habits, achieving a long-term and robust improvement in visualization effects and data fusion accuracy.

[0056] The technical solution of this invention has strong industry applicability and can be widely applied to various digital transformation scenarios such as smart transportation, urban governance, and emergency management, with significant social benefits and economic value.

[0057] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. A smart city data visualization method based on multi-source data fusion, characterized in that, Includes the following steps: Step S1: Collect multi-dimensional heterogeneous raw data of smart city, perform data cleaning, format standardization and outlier removal on the raw data to obtain a preprocessed dataset; Step S2: Align the preprocessed dataset with a unified spatiotemporal reference to generate a spatiotemporally standardized dataset. Based on the improved DS evidence theory, perform weight allocation and conflict resolution on the cross-source conflict data in the spatiotemporally standardized dataset to obtain a trustworthy fusion dataset. Step S3: Obtain the scene features, user interaction commands and terminal computing power parameters of the current visualization scene. Based on the preset scene-policy mapping model, dynamically adjust the granularity, dimension and computing priority of data fusion to generate a target fusion dataset adapted to the scene. Step S4: Combine the rendering layer depth of the target fusion dataset in the current visualization scene, dynamically allocate differential privacy budgets to different layers, inject Laplacian noise, and further combine the attribute base encryption algorithm to generate a visualization permission dataset that matches the user's access permissions. Step S5: Based on the visualization permission dataset, construct a three-dimensional visualization scene with spatiotemporal-business-event linkage, and complete the screen rendering based on the distributed cloud-terminal collaborative rendering engine of view frustum and LOD. At the same time, respond to the user's drill-down, linkage, and filtering interaction commands, and update the visualization rendering content in real time. Step S2 specifically includes: Step S21: Construct a unified spatiotemporal reference system, using the WGS84 coordinate system as the spatial reference and the UTC+8 time zone as the time reference, and convert the spatial coordinates and timestamps of all data in the preprocessed dataset to the unified spatiotemporal reference. Step S22: For cross-source data at the same spatiotemporal location, calculate the time decay weight, spatial matching weight, and historical reliability weight of each data source to obtain the comprehensive weight of each data source. Step S23: Obtain the basic probability allocation function of each data source for a certain fusion event. When calculating the conflict coefficient of each pair of evidence sources, introduce the Jousselme evidence distance to quantify the similarity between highly conflicting evidence. Step S24: When the conflict coefficient is greater than the preset conflict tolerance coefficient, the comprehensive weight is used as the absolute credibility conversion factor to modify the BPA function. The probability quality of the conflict part is dynamically redistributed to the specified proposition according to the comprehensive weight ratio of each data source, instead of all being allocated to the unknown set, to generate a credible fusion dataset. Step S3 specifically includes: Step S31: Extract scene features of the current visualization scene, including scene type, rendering range, and data dimension requirements; obtain user interaction commands, including zoom, pan, drill-down, and filter operations; collect computing power parameters of the current terminal, including CPU / GPU utilization and network bandwidth. Step S32: Input scene features, user interaction commands, and computing power parameters into the pre-trained scene-policy mapping model, and output the optimal fusion strategy. The optimal fusion strategy includes at least the temporal granularity, spatial granularity, business dimension, and computational priority of data fusion. The state space S of the pre-trained scene-policy mapping model... t Represented as a state vector: In the formula: L view R represents the height hierarchy parameter of the camera in the current visualization scene. fov U represents the geographical area covered by the camera's field of view cone. cpu U gpu These are the real-time CPU load rate and real-time GPU load rate of the visualization terminal, respectively. net For the terminal's real-time network downlink bandwidth; I op Encoding vectors for multi-dimensional user interaction commands; The action space A of the model t Represented as an action vector: A t =[Δg s ,Δg t ,P cal ], where: Δg s Δg is the adjustment amount for the aggregation or refinement order of the spatial grid fusion granularity; t P is the step size adjustment value for the time fusion slice; cal This refers to the cloud computing power offloading ratio and rendering pipeline priority parameters corresponding to different business data within the current field of view; Step S33: According to the optimal fusion strategy, perform secondary fusion processing on the trusted fusion dataset to generate a target fusion dataset that is adapted to the scenario. Step S34: Monitor the frame rate and latency of the visualization rendering in real time. When the frame rate is lower than the preset threshold or the latency is higher than the preset threshold, dynamically adjust the fusion strategy to reduce the fusion granularity of non-core data and ensure smooth rendering.

2. The smart city data visualization method based on multi-source data fusion according to claim 1, characterized in that, In step S1, the collected multi-dimensional heterogeneous raw data of the smart city includes at least: real-time monitoring data collected by IoT sensing devices, structured business data from government business systems, unstructured video stream data from video surveillance devices, internet public opinion data, and geospatial information data; the preprocessing specifically includes: Step S11: To address the format differences between different data sources, construct a unified semantic mapping dictionary to map the field names, data types, and units of heterogeneous raw data to a unified semantic standard. Step S12: Use a combination of the 3σ criterion and the isolated forest algorithm to remove outliers and noisy data from the original data; Step S13: For missing data, a generalized kriging interpolation algorithm that incorporates spatiotemporal distance decay factors is used to complete the data, resulting in a preprocessed dataset.

3. The smart city data visualization method based on multi-source data fusion according to claim 1, characterized in that, In step S22, the formula for calculating the time decay weight is: in, For time decay weight, The preset attenuation coefficient, This is the time difference between the data collection time and the current time. The formula for calculating the spatial matching degree weight is: in, For spatial matching degree weight, The preset spatial attenuation coefficient, The Euclidean distance between the data acquisition location and the target spatiotemporal location; The historical reliability weight is dynamically updated based on the historical data accuracy of the data source, with an initial value of 1.

4. The smart city data visualization method based on multi-source data fusion according to claim 1, characterized in that, The scenario-policy mapping model is a deep Q-network model based on reinforcement learning, and its training process includes: Using scene rendering smoothness, data fidelity, and computational efficiency as reward functions, and historical visualized scene data, corresponding fusion strategies, and rendering effects as the training set, the deep Q-network model is iteratively trained until convergence. The reward function R of the deep Q-network model is calculated using the following formula: ； in, Real-time rendering frame rate for visualized scenes; , These are the preset target rendering frame rate and the minimum tolerable rendering frame rate, respectively. , These are the information entropy of the original high-precision business data and the information entropy of the fused data, respectively. , These represent the real-time computing power usage and total computing power throughput of the visualization terminal, respectively; θ is the preset terminal computing power warning threshold; α, β, and γ are preset weight hyperparameters, respectively.

5. The smart city data visualization method based on multi-source data fusion according to claim 1, characterized in that, Step S4 specifically includes: Step S41: Identify sensitive data fields in the target fusion dataset, add differential privacy noise to the sensitive data fields based on the Laplace mechanism, and complete the data desensitization process; Step S42: Based on the attribute-based encryption algorithm, generate corresponding attribute private keys for different users according to their roles, departments, and access permissions; Step S43: Perform layered encryption on the desensitized target fusion dataset to generate a visualization permission dataset that matches different access permissions. The visualization data of the corresponding level can only be decrypted when the user's attribute private key meets the access policy.

6. The smart city data visualization method based on multi-source data fusion according to claim 5, characterized in that, Step S5 specifically includes: Step S51: Based on the visual permission dataset, construct a four-level visualization hierarchy system including a macro overview layer, a regional monitoring layer, a single-entity details layer, and an event tracing layer; Step S52: Establish a linkage mapping relationship between the spatiotemporal dimension, business dimension, and event dimension. When a user performs filtering or drill-down operations on the data in any dimension, the visualization content of the other two dimensions is updated synchronously. Step S53: Using a distributed rendering engine, the rendering tasks are distributed to cloud rendering nodes and local terminal nodes. The rendering resolution and effect level are adapted according to the terminal performance to complete the visualization rendering adapted to multiple terminals. Step S54: Respond to user interaction commands in real time, and update data fusion strategies and rendering content synchronously to achieve real-time linkage between interaction and visualization.

7. The smart city data visualization method based on multi-source data fusion according to any one of claims 1 to 6, characterized in that, Following step S5, the following is also included: Step S6: Iterative optimization of visualization effects and data fusion quality: Collect user interaction behavior data, visualization rendering effect data, and data fusion accuracy data. Based on user feedback and running data, iteratively optimize the scene-strategy mapping model, conflict resolution algorithm, and rendering engine parameters.

8. A smart city data visualization system based on multi-source data fusion, used to execute the method as described in any one of claims 1-7, characterized in that, Includes the following modules: The data preprocessing module is configured to collect multi-dimensional heterogeneous raw data from smart cities, perform data cleaning, format standardization, and outlier removal on the raw data, and generate a preprocessed dataset. The spatiotemporal benchmark alignment module is connected to the data preprocessing module. It is used to perform unified spatiotemporal benchmark alignment on the preprocessed dataset, generate a spatiotemporally standardized dataset, and perform weight allocation and conflict resolution on cross-source conflict data based on the improved DS evidence theory, and output a credible fusion dataset. The fusion strategy adjustment module is connected to the spatiotemporal reference alignment module. By acquiring the scene features, user interaction instructions and terminal computing power parameters of the current visualization scene, it dynamically adjusts the granularity, dimension and computing priority of data fusion based on the preset scene-strategy mapping model to generate a target fusion dataset adapted to the scene. The privacy management module is connected to the fusion strategy adjustment module, performs differential privacy desensitization processing on the target fusion dataset, and generates a visual permission dataset that matches user access permissions based on the attribute base encryption algorithm. The rendering interaction module is connected to the privacy control module. It constructs a three-dimensional interactive visualization scene based on the visualization permission dataset, which integrates time, space, business, and events. It completes multi-terminal adapted screen rendering through a distributed rendering engine and updates the rendered content in real time in response to the user's drill-down, linkage, and filtering interaction commands.

Images