Multi-source data fusion and anomaly detection method based on three-dimensional GIS and intelligent optimization

Abstract

The invention provides a multi-source data fusion and anomaly detection method based on a three-dimensional GIS (Geographic Information System) and intelligent optimization. The method comprises the following steps: collecting various data sources, carrying out preprocessing by utilizing a multi-modal data fusion and self-adaptive data enhancement technology, realizing data synchronization and integration by adopting a graph-based data structure technology and through a multi-level data fusion algorithm, constructing a high-precision dynamic three-dimensional model by using a three-dimensional geographic information system and an augmented reality technology, realizing visualization, and realizing real-time real-time monitoring. And multi-view and multi-scale real-time interaction is supported. And in combination with self-supervised learning and a graph attention network, key modes and trends are analyzed and identified in real time. A generative adversarial network and an attention mechanism are applied to carry out anomaly detection, and abnormal conditions are rapidly identified and early warned. And providing real-time response suggestions based on a data analysis result, optimizing a decision process by adopting a reinforcement learning algorithm, and dynamically coping with potential risks. According to the method, multi-source data are efficiently fused, and the method has accurate real-time monitoring and response capabilities.

Description

Technical Field

[0001] The present invention relates to the fields of data fusion and anomaly detection, and particularly to a multi-source data fusion and anomaly detection method based on 3D GIS and intelligent optimization. Background Art

[0002] With the rapid development of technology and social progress, the acquisition and utilization of data have become increasingly important. Especially in the fields of urban management, environmental monitoring, disaster warning, and intelligent transportation, the fusion and analysis of multi-source data have become the key to improving the accuracy and timeliness of decision-making. However, there are still many deficiencies in the existing technologies for multi-source data fusion and anomaly detection.

[0003] In the existing technologies, most of the methods for multi-source data fusion are limited to simple data superposition and statistical analysis, and it is difficult to effectively integrate data from different sources. For example, data sources such as satellite images, meteorological data, traffic flow information, and social media dynamics have their own characteristics, and direct superposition or simple analysis often cannot fully exploit the useful information therein. In addition, the existing data fusion methods lack consideration of data heterogeneity and are difficult to cope with the differences in time and space of different data sources, resulting in poor fusion effects.

[0004] In terms of data preprocessing, the existing technologies mainly rely on traditional normalization and filtering methods and fail to fully utilize modern machine learning and adaptive technologies. These methods show obvious limitations when dealing with complex and variable actual data and cannot effectively improve data quality and consistency. The applications of 3D geographic information systems and augmented reality technologies in the existing technologies are also relatively limited. Although some studies have attempted to apply 3D GIS and AR technologies to data visualization, most of them stay at the static display level and lack real-time reflection and interaction support for dynamic changes. This results in the inability to update and display the latest data changes in actual applications, affecting the accuracy and timeliness of decision-making.

[0005] In terms of anomaly detection, the existing technologies mainly rely on traditional methods based on rules or statistics and are difficult to cope with complex and variable abnormal situations. Especially when dealing with multi-source data, the detection accuracy and efficiency of traditional methods often cannot meet the actual requirements. In addition, the existing anomaly detection methods lack intelligence and adaptability and cannot dynamically adjust the detection strategy according to the actual situation, resulting in frequent false negatives and false positives.

[0006] Therefore, how to provide a multi-source data fusion and anomaly detection method based on 3D GIS and intelligent optimization is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0007] An object of the present invention is to propose a multi-source data fusion and anomaly detection method based on 3D GIS and intelligent optimization. The present invention makes full use of technologies such as multi-modal data fusion, adaptive data augmentation, 3D geographic information system, augmented reality, generative adversarial network, graph attention network, and reinforcement learning, and details the implementation methods for efficient fusion, real-time analysis, and anomaly detection of multi-source data, having the advantages of good data integration effect, strong real-time performance, high detection accuracy, and strong decision-making optimization ability.

[0008] A multi-source data fusion and anomaly detection method based on 3D GIS and intelligent optimization according to an embodiment of the present invention includes the following steps: S1. Collect multiple data sources including satellite images, meteorological sensor data, traffic flow information, and social media dynamics, and preprocess the data through multi-modal data fusion and adaptive data augmentation technologies; S2. In a three-dimensional space, adopt a graph-based data structure technology to synchronize and integrate each data source through a multi-level data fusion algorithm; S3. Through 3D geographic information system technology and augmented reality technology, construct and visualize a high-precision dynamic 3D model of the city and natural landscapes, and support real-time interactive operations from multiple perspectives and scales; S4. Apply a hybrid model combining self-supervised learning and graph attention network to analyze and identify key patterns and trends in multi-source data in real time; S5. Implement an anomaly detection technology based on generative adversarial network and attention mechanism to quickly identify and warn of anomalies caused by human activities or natural events from the data; S6. Provide real-time response suggestions based on the data analysis results, adopt a reinforcement learning algorithm to optimize the decision-making process, and support dynamic responses to identified anomalies and potential risks.

[0009] Optionally, the step S1 specifically includes: S11. Obtain raw data from four data sources: satellite images, meteorological sensors, traffic flow information, and social media dynamics; S12. Perform time series analysis on each data source, use an adaptive method to determine the analysis window and time step, and normalize each data source: ; wherein, is the normalized data, is the raw data, is the mean within the time window is the standard deviation, is the adjustment parameter, is the attenuation coefficient, is the past time point, is the past time point, is the attenuation factor, are respectively , , and ; S13. Use the multi-modal data fusion technology to extract features from each data source: ; Among them, is the non-linear activation function, is the layer normalization function, is the weighting coefficient, and are the weight matrix and the bias term, is the original data of the th data source; Align the time and space features of different data sources to form an aligned feature matrix ; S14. Apply the adaptive data augmentation technology according to the data characteristics to enhance the feature matrix and generate an enhanced data set : ; Among them, is the enhancement parameter, is the weight coefficient, is the data augmentation function, is the adversarial noise weight, is the adversarial noise term.

[0010] Optionally, the step S2 specifically includes: S21. Adopt the graph-based data structure technology in the three-dimensional space to integrate multiple data sources into a unified data graph, and each data source corresponds to a different sub-graph in the graph; S22. Synchronize different data sources in time and space through the multi-level data synchronization algorithm: ; Among them, is the state of the time time node , is the set of neighbor nodes of the node , is the synchronization weight between the node and the node , is the synchronization delay time, is the th attention weight coefficient, is the node Attention mechanism between the sum node ; S23. Through a multi-level data fusion algorithm, fuse the data from different data sources into a unified feature space, denoted as : ; Among them, is a non-linear activation function, is the feature matrix of the th layer, and are the weight matrix and bias term of the th layer respectively, is the query, key and value matrix, is the feature dimension, is the total number of layers of the fusion layer.

[0011] Optionally, the step S3 specifically includes: S31. Through three-dimensional geographic information system technology, map the fused multi-source data to a three-dimensional space to form an initial three-dimensional model; S32. Combine augmented reality technology to dynamically update the initial three-dimensional model through real-time data stream , denoted as : ; Among them, is the initial three-dimensional model, is the update coefficient, is a gated recurrent unit, is time at the moment of real-time data, is the update time interval, is the number of update times; S33. Use an intelligent optimization algorithm to optimize the dynamic three-dimensional model: ; Among them, is the optimized three-dimensional model, is the loss function, is the real data label, is the regularization coefficient, is the regularization term; S34. Visually display the optimized three-dimensional model to provide real-time interactive operations with multiple perspectives and scales.

[0012] Optionally, the step S4 specifically includes: S41. Combine self-supervised learning and graph attention network techniques to construct a deep learning model for real-time analysis and identification of key patterns and trends in multi-source data; S42. Process the real-time data stream from multi-source data through self-supervised learning methods , generate a new feature representation, denoted as , and the formula is as follows: ; where is the feature representation generated by self-supervised learning, is the attention mechanism, is the historical feature representation, and are adjustment parameters, is the start time, is the past time point, is the current time; S43. Use the deep learning model to analyze the feature representation of the real-time data stream and identify key patterns and trends: ; where is the key pattern recognition function, is the graph attention network, is the node feature of the th layer of the deep learning model, is the attention graph generated by self-supervised learning, and are weighting coefficients, is the gated recurrent unit, is the representation of the dynamic three-dimensional model at time ; S44. Determine the trend in the data by further analyzing the key patterns: ; where is the trend analysis function, is the trend fusion function, is the historical trend data, is the weighting coefficient, is the prediction model, is the number of prediction models.

[0013] Optionally, the specific steps of step S5 include: S51. Combine the generative adversarial network and the attention mechanism to construct an anomaly detection model for quickly identifying anomaly situations from multi-source data; S52. Use the attention mechanism to extract and fuse features from multi-source data to generate an input feature matrix for anomaly detection; S53. Input the input feature matrix into the anomaly detection model to generate an anomaly score . The anomaly detection steps specifically include: S531. Generate a high-dimensional feature representation from the input feature matrix through a multi-layer non-linear transformation; S532. Perform attention-weighted fusion on the high-dimensional feature representation; S533. Input the attention-weighted fused high-dimensional feature representation into the generator and discriminator in the generative adversarial network to generate adversarial features and discriminant scores respectively: ; wherein, is the adversarial generated feature, is the discriminator score, is the generator, is the discriminator; S534. Calculate the anomaly score based on the generated adversarial features and discriminator scores: ; wherein, is the norm, is the adjustment parameter, is the weight coefficient, is the attention mechanism; S54. Anomaly warning: Set a warning threshold according to the anomaly score to generate an anomaly warning signal, denoted as . The anomaly warning steps specifically include: S541. Use the Bayesian optimization algorithm to dynamically adjust the warning threshold based on historical data and real-time data: ; wherein, is the utility function based on historical data, is the variance based on real-time data, is the adjustment parameter; S542. Generate a warning signal in real time according to the adjusted warning threshold : ; wherein, is the anomaly score at time , is the dynamically adjusted warning threshold.

[0014] Optionally, step S6 specifically includes: S61. Optimize the decision-making process using a reinforcement learning model based on a neural network: ; where is the value of taking action in state , is the immediate reward, is the discount factor, and are the parameters of the current network and the target network, is the policy function; S62. Define the state space and action space of the system, where the state includes the current anomaly score , historical trend data , real-time data features , and the action includes various response measures; S63. Design a reward function to evaluate the effect of each action in a given state: ; where , , are weight coefficients, is the cost of action ; S64. Use the reinforcement learning algorithm to generate a real-time response policy , that is, select the optimal action in each state to maximize the long-term reward: ; S65. Dynamically respond to the identified anomalies and potential risks according to the real-time response policy , and the specific steps include: S651. Execute the action generated by the response policy in the current state of the system; S652. Obtain the new state and the immediate reward according to the executed action , and update the reinforcement learning model; S653. Through continuous iteration and optimization, the reinforcement learning model continuously improves the decision-making process.

[0015] The beneficial effects of the present invention are: The present invention provides a multi-source data fusion and anomaly detection method based on 3D GIS and intelligent optimization. Through innovative technical means and optimized algorithm design, it effectively solves many deficiencies in the prior art. Through multi-modal data fusion and adaptive data augmentation techniques, the quality and consistency of data are improved, and efficient fusion of multi-source data is achieved. Based on graph data structure technology and 3D geographic information system, through multi-level data fusion algorithms, a high-precision dynamic 3D model is constructed, and combined with augmented reality technology, it supports real-time interactive operations from multiple perspectives and scales, making data display more intuitive and vivid, and improving user experience and decision-making efficiency.

[0016] The present invention also uses a hybrid model of self-supervised learning and graph attention network to analyze and identify key patterns and trends in multi-source data in real time, improving the speed and accuracy of data analysis, and being able to capture potential anomalies and risks in a timely manner. Anomaly detection technology using generative adversarial network and attention mechanism is adopted to quickly identify and warn of abnormal situations caused by human activities or natural events, significantly improving the accuracy and response speed of anomaly detection, and reducing the probability of missed reports and false alarms. Through reinforcement learning algorithms, real-time response suggestions are provided based on data analysis results, dynamically optimizing the decision-making process, enhancing the system's ability to cope with complex environments and emergencies, improving the scientificity and effectiveness of decision-making, and supporting the efficient operation of fields such as urban management, environmental monitoring, disaster warning, and intelligent transportation.

[0017] By introducing technologies such as multi-modal data fusion, adaptive data augmentation, 3D geographic information system, augmented reality, generative adversarial network, graph attention network, and reinforcement learning, the present invention successfully solves many deficiencies in multi-source data fusion and anomaly detection in the prior art, significantly improving the accuracy and efficiency of data analysis and anomaly detection, and providing strong technical support for related fields. This not only improves the overall performance of the system, but also provides new technical means and solutions for modern urban management and intelligent monitoring. BRIEF 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. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention. In the drawings: Figure 1 is the overall flowchart of a multi-source data fusion and anomaly detection method based on 3D GIS and intelligent optimization proposed by the present invention; Figure 2 is the detailed flowchart of multi-modal data fusion and adaptive data augmentation techniques in the present invention; Figure 3 is the flowchart of applying generative adversarial network and attention mechanism for anomaly detection and providing real-time response suggestions and optimizing the decision-making process through reinforcement learning algorithms. Detailed implementation mode

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

[0020] Reference Figure 1 , a multi-source data fusion and anomaly detection method based on 3D GIS and intelligent optimization, comprising the following steps: S1. Collect multiple data sources including satellite images, meteorological sensor data, traffic flow information, and social media dynamics, and preprocess the data through multi-modal data fusion and adaptive data enhancement technologies; In this implementation mode, step S1 specifically includes: S11. Obtain raw data from four data sources: satellite images, meteorological sensors, traffic flow information, and social media dynamics; S12. Perform time series analysis on each data source, use an adaptive method to determine the analysis window and time step, and normalize each data source: ; Among them, is the normalized data, is the raw data, is the mean within the time window , is the standard deviation, is the adjustment parameter, is the attenuation coefficient, is the past time point, is the attenuation factor, are respectively , , and ; S13. Use multi-modal data fusion technology to extract features from each data source: ; Among them, is the non-linear activation function, is the layer normalization function, is the weighting coefficient, and are the weight matrix and bias term, is the th raw data of the data source; Align the time and space features of different data sources to form an aligned feature matrix ; S14. Apply the adaptive data augmentation technology according to the data characteristics to enhance the feature matrix and generate an enhanced data set. : ; Among them, is the augmentation parameter, is the weight coefficient, is the data augmentation function, is the adversarial noise weight, is the adversarial noise term.

[0021] S2. Adopt the graph-based data structure technology in the three-dimensional space and achieve the synchronization and integration of each data source through the multi-level data fusion algorithm; In this embodiment, step S2 specifically includes: S21. Adopt the graph-based data structure technology in the three-dimensional space to integrate multiple data sources into a unified data graph, and each data source corresponds to a different sub-graph in the graph; S22. Through the multi-level data synchronization algorithm, synchronize different data sources in terms of time and space: ; Among them, is the time time node state, is the node neighbor node set, is the node and the node synchronization weight between, is the synchronization delay time, is the th attention weight coefficient, is the node and the node attention mechanism between; S23. Through the multi-level data fusion algorithm, fuse the data of different data sources into a unified feature space, denoted as : ; Among them, is the non-linear activation function, is the th layer feature matrix, and are respectively the th layer weight matrix and bias term, is the query, key and value matrix, is the feature dimension, is the total number of layers of the fusion layer.

[0022] Through this embodiment, time series analysis, normalization, multi-modal data fusion, and adaptive data enhancement processing are performed on the raw data obtained from four data sources: satellite images, meteorological sensors, traffic flow information, and social media dynamics, to generate a preprocessed data set.

[0023] S3. Construct and visualize a high-precision dynamic 3D model of the city and natural landscape through 3D geographic information system technology and augmented reality technology, and support real-time interactive operations from multiple perspectives and scales; In this embodiment, step S3 specifically includes: S31. Through 3D geographic information system technology, map the fused multi-source data to a 3D space to form an initial 3D model; S32. Combine augmented reality technology and dynamically update the initial 3D model through real-time data streams Denoted as : ; where is the initial 3D model, is the update coefficient, is the gated recurrent unit, is time is the real-time data at time is the update time interval, is the number of updates; S33. Use an intelligent optimization algorithm to optimize the dynamic 3D model: ; where is the optimized 3D model, is the loss function, is the real data label, is the regularization coefficient, is the regularization term; S34. Visualize the optimized 3D model and provide real-time interactive operations from multiple perspectives and scales.

[0024] Through this embodiment, each data source is constructed and synchronized in 3D space to achieve the integration and unified representation of multi-source data.

[0025] S4. Apply a hybrid model combining self-supervised learning and graph attention network to analyze and identify key patterns and trends in multi-source data in real time; In this embodiment, step S4 specifically includes: S41. Combine self-supervised learning and graph attention network technology to construct a deep learning model for real-time analysis and identification of key patterns and trends in multi-source data; S42. Process the real-time data stream from multi-source data through self-supervised learning methods , generating a new feature representation, denoted as , with the formula as follows: ; where, is the feature representation generated by self-supervised learning, is the attention mechanism, is the historical feature representation, and are adjustment parameters, is the start time, is the past time point, is the current time; S43. Use the deep learning model to analyze the feature representation of the real-time data stream and identify key patterns and trends: ; where, is the key pattern recognition function, is the graph attention network, is the node feature of the th layer of the deep learning model, is the attention graph generated by self-supervised learning, and are weighting coefficients, is the gated recurrent unit, is the representation of the dynamic three-dimensional model at time ; S44. Determine the trend in the data through further analysis of the key patterns: ; where, is the trend analysis function, is the trend fusion function, is the historical trend data, is the weighting coefficient, is the prediction model, is the number of prediction models.

[0026] Construct and visualize a dynamic three-dimensional model of the city and natural landscape through three-dimensional geographic information system technology and augmented reality technology.

[0027] S5. Implement an anomaly detection technology based on generative adversarial network and attention mechanism to quickly identify and warn of anomalies caused by human activities or natural events from the data; In this embodiment, step S5 specifically includes: S51. Combining a generative adversarial network and an attention mechanism to construct an anomaly detection model for quickly identifying anomaly situations from multi-source data; S52. Using the attention mechanism to extract and fuse features from multi-source data to generate an input feature matrix for anomaly detection; S53. Inputting the input feature matrix into the anomaly detection model to generate an anomaly score , and the anomaly detection steps specifically include: S531. Generating a high-dimensional feature representation by performing multi-layer non-linear transformation on the input feature matrix ; S532. Performing attention-weighted fusion on the high-dimensional feature representation; S533. Inputting the attention-weighted fused high-dimensional feature representation into the generator and discriminator in the generative adversarial network to respectively generate adversarial features and discriminant scores: ; wherein, is the adversarial generated feature, is the discriminator score, is the generator, is the discriminator; S534. Calculating the anomaly score according to the generated adversarial features and discriminator score: ; wherein, is the norm, is the adjustment parameter, is the weight coefficient, is the attention mechanism; S54. Anomaly warning: Setting a warning threshold according to the anomaly score to generate an anomaly warning signal, denoted as , and the anomaly warning steps specifically include: S541. Using the Bayesian optimization algorithm to dynamically adjust the warning threshold according to historical data and real-time data: ; wherein, is the utility function based on historical data, is the variance based on real-time data, is the adjustment parameter; S542. Generating a warning signal in real time according to the adjusted warning threshold ​ : ; wherein, is the anomaly score at time moment, is the warning threshold after dynamic adjustment.

[0028] Combining deep learning and graph attention network technologies, analyze and identify key patterns and trends in multi-source data in real time.

[0029] S6. Provide real-time response suggestions based on the data analysis results, optimize the decision-making process using reinforcement learning algorithms, and support dynamic responses to identified anomalies and potential risks.

[0030] In this embodiment, step S6 specifically includes: S61. Use a reinforcement learning model based on a deep network to optimize the decision-making process: ; wherein, is the value of taking action in state , is the immediate reward, is the discount factor, and are the parameters of the current network and the target network, is the policy function; S62. Define the state space and action space of the system, where the state includes the current anomaly score , historical trend data , real-time data features , and the actions include various response measures; S63. Design a reward function for evaluating the effect of each action in a given state: ; wherein, , , are weight coefficients, is the cost of action ; S64. Use the reinforcement learning algorithm to generate a real-time response policy , that is, select the optimal action in each state to maximize the long-term reward: ; S65. According to the real-time response strategy , dynamically respond to the identified anomalies and potential risks. The specific steps include: S651. Execute the actions generated by the response strategy in the current state of the system; S652. Obtain a new state and an immediate reward , and update the reinforcement learning model; S653. Through continuous iteration and optimization, the reinforcement learning model continuously improves the decision-making process.

[0031] Example 1: In a modern big city A, the urban management department faces the challenges of complex multi-source data processing and anomaly detection. The data sources include satellite images, meteorological sensors, traffic flow information, and social media dynamics, etc. These data are huge in volume and from different sources, and traditional data processing methods are difficult to meet the requirements of real-time and accuracy. Especially in emergencies such as natural disasters, traffic accidents, and public security incidents, the urban management department needs to respond quickly and make decisions, while the existing technologies have obvious deficiencies in data fusion, real-time analysis, and anomaly detection.

[0032] To solve the above problems, the urban management department of city A decides to adopt the multi-source data fusion and anomaly detection method based on three dimensions and intelligent optimization of the present invention. The specific steps are as follows: First, collect a variety of data sources, including satellite images, meteorological data, traffic flow information, and social media dynamics. Through multi-modal data fusion and adaptive data augmentation techniques, preprocess these data. The data preprocessing includes time series analysis and normalization processing. Determine the analysis window and time step through an adaptive method to improve the consistency and availability of the data.

[0033] In terms of data synchronization and integration, adopt the data structure technology based on graphs to achieve multi-level data fusion in three-dimensional space. Through the multi-level data fusion algorithm, integrate different data sources into a unified data graph, realizing synchronization and integration in time and space. Utilize three-dimensional geographic information system technology and augmented reality technology to construct a high-precision dynamic three-dimensional model, realizing the visualization of the city and natural landscapes, and supporting real-time interactive operations from multiple perspectives and scales. In real-time data analysis and anomaly detection, adopt a hybrid model combining self-supervised learning and graph attention network to conduct real-time analysis and recognition of multi-source data. Through the generative adversarial network and attention mechanism, quickly identify and warn of anomalies caused by human activities or natural events. Finally, based on the data analysis results, provide real-time response suggestions, and optimize the decision-making process through the reinforcement learning algorithm to dynamically respond to the identified anomalies and potential risks.

[0034] ​During the period from June to December 2023, the urban management department of City A applied the method of the present invention in the following aspects and achieved remarkable results: 1. Natural disaster warning: In August 2023, City A was hit by a severe typhoon. Through the method of the present invention, satellite images and meteorological data were synchronized in real time, and a three-dimensional typhoon path model was constructed. Based on the generative adversarial network and attention mechanism, the landing point and influence range of the typhoon were warned 3 hours in advance, winning precious time for urban evacuation and emergency preparation. Compared with the traditional method, the warning accuracy rate increased by 20%, and the response time was shortened by 2 hours.

[0035] 2. Traffic accident detection and management: During the morning rush hour in October 2023, a serious traffic accident occurred on a main road in City A. Through real-time collection and analysis of traffic flow information and social media dynamics, the method of the present invention quickly identified and located the accident site. The three-dimensional traffic flow model showed the impact of the accident on the surrounding road network, and the urban management department quickly adjusted the traffic signals and diversion plans according to the real-time response suggestions. The accident handling time was shortened from the traditional 1 hour to 30 minutes, and the traffic recovery speed increased by 50%.

[0036] 3. Public safety event response: In November 2023, during a large-scale public event in City A, social media dynamic data showed abnormal crowd gathering. Through the method of the present invention, combined with the graph attention network and the generative adversarial network, the crowd gathering pattern was analyzed in real time, and potential safety risks were warned. The urban management department immediately took measures to divert the crowd and increase security forces, effectively avoiding the occurrence of safety accidents. Compared with the traditional manual monitoring, the abnormal detection accuracy rate increased by 30%, and the response time was shortened by 40%.

[0037] 4. Environmental monitoring and governance: In December 2023, City A implemented a large-scale air quality monitoring and governance project. The method of the present invention showed the dynamic changes of the city's air quality in real time through multi-source data fusion and three-dimensional geographic information system technology. Based on the data analysis results, the urban management department optimized the pollution source control and greening measures. The accuracy of air quality monitoring increased by 15%, and the pollution treatment efficiency increased by 25%.

[0038] Table 1 Comprehensive effect data table Time Application Area Data Source Type Enhancement Effect August 2023 Natural Disaster Warning Satellite Images, Meteorological Data Warning Accuracy Increased by 20%, Warning Advanced by 3 Hours, Response Time Shortened by 2 Hours October 2023 Traffic Accident Detection Traffic Flow Information, Social Media Trends Processing Time Shortened by 30 Minutes, Traffic Recovery Speed Increased by 50% November 2023 Public Safety Response Social Media Trends Detection Accuracy Increased by 30%, Response Time Shortened by 40% December 2023 Environmental Monitoring and Governance Meteorological Data, Air Quality Monitoring Data Monitoring Accuracy Increased by 15%, Pollution Governance Efficiency Increased by 25% Table 2 Specific effect data table Time Application Area Specific Effect August 2023 Natural Disaster Warning Typhoon Warning Advanced by 3 Hours, Warning Accuracy 95%, Response Time Shortened by 2 Hours October 2023 Traffic Accident Detection Morning Rush Hour Traffic Accident Processing Time Reduced from 1 Hour to 30 Minutes, Traffic Recovery Speed Increased by 50% November 2023 Public Safety Response Abnormal Detection Accuracy in Large Public Events Increased by 30%, Response Time Shortened by 40% December 2023 Environmental Monitoring and Governance Air Quality Monitoring Accuracy Increased by 15%, Pollution Governance Efficiency Increased by 25% Through the above data table, during the period from June to December 2023, the urban management department of City A applied the multi-source data fusion and anomaly detection method based on three dimensions and intelligent optimization of the present invention, achieving remarkable results. In terms of natural disaster early warning, the early warning accuracy rate increased by 20%, the early warning time was advanced by 3 hours, and the response time was shortened by 2 hours; in traffic accident detection and management, the accident handling time was shortened by 30 minutes, and the traffic recovery speed increased by 50%; in public safety incident response, the anomaly detection accuracy rate increased by 30%, and the response time was shortened by 40%; in environmental monitoring and governance, the accuracy of air quality monitoring increased by 15%, and the pollution treatment efficiency increased by 25%. These data indicate that the method of the present invention shows significant advantages in multi-source data fusion, real-time analysis, anomaly detection, and decision optimization, significantly improving the accuracy and efficiency of data analysis and providing strong technical support for urban management.

[0039] From the above examples, it can be seen that the method of the present invention shows significant advantages in data fusion, real-time analysis, anomaly detection, and decision optimization. The efficient fusion and consistency of multi-source data improve the comprehensiveness and accuracy of information; the dynamic three-dimensional model and visualization technology enhance the intuitiveness and practicality of data display; the technology combining self-supervised learning, graph attention network, and generative adversarial network improves the accuracy and response speed of data analysis and anomaly detection; the decision-making process optimized by the reinforcement learning algorithm ensures the response ability and decision-making scientificity in complex environments and emergencies.

[0040] The method of the present invention significantly improves the technical support capabilities in the fields of urban management, environmental monitoring, disaster early warning, and intelligent transportation, providing a powerful technical means for modern urban management.

[0041] The above is only the preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and all should be covered within the protection scope of the present invention.

Claims

1. A multi-source data fusion and anomaly detection method based on 3D GIS and intelligent optimization, characterized in that It includes the following steps: S1. Collect multiple data sources including satellite images, meteorological sensor data, traffic flow information, and social media dynamics, and preprocess the data through multi-modal data fusion and adaptive data augmentation techniques; S2. In three-dimensional space, adopt graph-based data structure technology to synchronize and integrate each data source through a multi-level data fusion algorithm; S3. Through three-dimensional geographic information system technology and augmented reality technology, construct and visualize a high-precision dynamic three-dimensional model of the city and natural landscape, and support real-time interactive operations from multiple perspectives and scales; S4. Apply a hybrid model combining self-supervised learning and graph attention network to analyze and identify key patterns and trends in multi-source data in real time; S5. Implement anomaly detection technology based on generative adversarial network and attention mechanism to quickly identify and warn of anomalies caused by human activities or natural events from the data; S6. Provide real-time response suggestions based on the data analysis results, adopt reinforcement learning algorithm to optimize the decision-making process, and support dynamic response to identified anomalies and potential risks.

2. A multi-source data fusion and anomaly detection method based on 3D GIS and intelligent optimization according to claim 1, characterized in that The specific steps of S1 include: S11. Obtain raw data from four data sources: satellite images, meteorological sensors, traffic flow information, and social media dynamics; S12. Conduct time series analysis on each data source, use an adaptive method to determine the analysis window and time step, and normalize each data source; ； Among them, is the normalized data, is the original data, is the mean value within the time window ; is the standard deviation, is the adjustment parameter, is the attenuation coefficient, is the past time point, is the attenuation factor, are respectively , , and ; S13. Use multi-modal data fusion technology to extract features from each data source; ； Among them, is a non-linear activation function, is a layer normalization function, is a weighting coefficient, and are the weight matrix and the bias term, is the original data of the th data source; Align the temporal and spatial features of different data sources to form an aligned feature matrix ; S14. Apply the adaptive data augmentation technique according to the data characteristics to enhance the feature matrix and generate an enhanced data set : ； wherein, is the enhancement parameter, is the weight coefficient, is the data enhancement function, is the adversarial noise weight, is the adversarial noise term.

3. A multi-source data fusion and anomaly detection method based on 3D GIS and intelligent optimization according to claim 1, characterized in that, The specific steps of S2 include: S21. In three-dimensional space, adopt graph-based data structure technology to integrate multiple data sources into a unified data graph, and each data source corresponds to a different sub-graph in the graph; S22. Through a multi-level data synchronization algorithm, synchronize different data sources in time and space; ； Among them, is the time time node status, is the set of neighbor nodes of node , is the synchronization weight between node and node , is the synchronization delay time, is the th attention weight coefficient, is the attention mechanism between node and node ; S23. Through a multi-level data fusion algorithm, fuse the data from different data sources into a unified feature space, denoted as : ； Among them, is a non-linear activation function, is the feature matrix of the th layer, and are the weight matrix and bias term of the th layer respectively, is the query, key and value matrix, is the feature dimension, is the total number of layers of the fusion layer.

4. A multi-source data fusion and anomaly detection method based on 3D GIS and intelligent optimization according to claim 1, characterized in that The specific steps of S3 include: S31. Map the fused multi-source data into a three-dimensional space through three-dimensional geographic information system technology to form an initial three-dimensional model; ​ S32. Combine with the augmented reality technology and dynamically update the initial three-dimensional model through real-time data streams which is denoted as : ； Among them, is the initial three-dimensional model, is the update coefficient, is the gated recurrent unit, is time the real-time data at the moment, is the update time interval, is the number of updates; S33. Use intelligent optimization algorithms to optimize the dynamic three-dimensional model; ； Among them, is the optimized 3D model, is the loss function, is the true data label, is the regularization coefficient, is the regularization term; S34. Visualize the optimized 3D model to provide real-time interactive operations from multiple perspectives and at multiple scales.

5. A multi-source data fusion and anomaly detection method based on 3D GIS and intelligent optimization according to claim 1, characterized in that The specific steps of S4 include: S41. Combine self-supervised learning and graph attention network technologies to construct a deep learning model for real-time analysis and identification of key patterns and trends in multi-source data; S42. Process the real-time data stream from multi-source data through a self-supervised learning method , generate a new feature representation, denoted as , and the formula is as follows: ； wherein, is the feature representation generated by self-supervised learning, is the attention mechanism, is the historical feature representation, and are the adjustment parameters, is the start time, is the past time point, is the current time; S43. Use the deep learning model to analyze the feature representation of the real-time data stream and identify key patterns and trends; ； Among them, is the key pattern recognition function, is the graph attention network, is the node feature of the th layer of the deep learning model, is the attention map generated by self-supervised learning, and are the weighting coefficients, is the gated recurrent unit, is the representation of the dynamic three-dimensional model at time ; S44. Determine the trend in the data through further analysis of the key patterns; ； Among them, is a trend analysis function, is a trend fusion function, is historical trend data, is a weighting coefficient, is a prediction model, is the number of prediction models.

6. A multi-source data fusion and anomaly detection method based on 3D GIS and intelligent optimization according to claim 1, characterized in that, The specific steps of S5 include: S51. Combine generative adversarial network and attention mechanism to construct an anomaly detection model for quickly identifying anomalies from multi-source data; S52. Use the attention mechanism to extract and fuse features from multi-source data to generate an input feature matrix for anomaly detection; S53. Input the input feature matrix into the anomaly detection model to generate an anomaly score . The anomaly detection steps specifically include: S531. Generate a high-dimensional feature representation by performing multi-layer non-linear transformation on the input feature matrix ; S532. Perform attention-weighted fusion on the high-dimensional feature representation; S533. High-dimensional feature representation with attention-weighted fusion Input the generator and discriminator in the generative adversarial network to generate adversarial features and discriminant scores respectively: ； Among them, is the adversarial generated feature, is the discriminator score, is the generator, is the discriminator; S534. Calculate the anomaly score based on the generated adversarial features and discriminator scores : ； Among them, is the norm, is the adjustment parameter, is the weight coefficient, is the attention mechanism; S54. Abnormal warning: According to the abnormal score Set the warning threshold value to generate an abnormal warning signal, denoted as , and the specific steps of the abnormal warning include: S541. Dynamically adjust the warning threshold according to historical data and real-time data using the Bayesian optimization algorithm : ； Among them, is the utility function based on historical data, is the variance based on real-time data, is the adjustment parameter; S542. Generate warning signals in real time according to the adjusted warning threshold :​ ； wherein, is the abnormal score at time , and is the warning threshold after dynamic adjustment.

7. A multi-source data fusion and anomaly detection method based on 3D GIS and intelligent optimization according to claim 1, characterized in that, The specific steps of S6 include: S61. Adopt a reinforcement learning model based on a deep network to optimize the decision-making process: ； wherein, is the state under which the action is taken of the value is the immediate reward is the discount factor and are the parameters of the current network and the target network is the policy function; S62. Define the state space and action space of the system, where the state includes the current anomaly score , historical trend data , real-time data features , and the actions include various response measures; S63. Design the reward function , which is used to evaluate the effect of each action in a given state: ； Among them, , , are weight coefficients, is the cost of the action . S64. Generating a real-time response policy using a reinforcement learning algorithm That is, selecting the optimal action at each state to maximize the long-term reward: ； S65. According to the real-time response strategy , dynamically respond to the identified anomalies and potential risks. The specific steps include: S651. Execute the actions generated by the response strategy in the current state of the system; S652. According to the executed action Obtain a new state and an immediate reward , and update the reinforcement learning model; S653. Through continuous iteration and optimization, the reinforcement learning model continuously improves the decision-making process.

Images