Data processing method and application based on multi-source data fusion

Abstract

The invention relates to the technical field of data processing, and discloses a data processing method and application based on multi-source data fusion. The method comprises the following steps: acquiring multi-source time sequence data (including sensor, geographic space, user behavior data and the like) and space vector base map data of a target area; de-noising and normalizing the multi-source time series data, extracting an associated feature map through a heterogeneous model, generating a regional data object set based on base map adaptive segmentation, and extracting multi-dimensional features; optimizing a data object set through spatial topology verification, feature matching screening and weighted fusion; and generating statistics, space-time coupling analysis and a dynamic prediction result by utilizing optimized data, or generating a visual map supporting interaction by mapping a base map. The method improves the multi-source data fusion precision and spatial analysis capability, is suitable for multi-field cross-modal decision support, and can be used for scenes of smart cities, environment monitoring and the like.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a data processing method and application based on multi-source data fusion. Background Art

[0002] In the information age, data in various fields is experiencing explosive growth, and multi-source data fusion has become a key technical direction for mining data value. Traditional data processing methods generally face the following technical bottlenecks when dealing with multi-source heterogeneous data such as sensor data, geospatial data, and user behavior data:

[0003] The heterogeneity of multi-source data makes fusion difficult. Different data sources vary significantly in format, sampling frequency, spatial reference system, and other aspects. For example, sensor data is typically high-frequency time series data, geospatial data is presented in vector or raster form, and user behavior data is discrete and dynamic. Traditional methods lack effective cross-modal feature matching mechanisms, making it difficult to discover potential connections between different data sources. This results in incomplete representation of fused data features and an inability to accurately reflect the true state of the target area.

[0004] Insufficient spatial constraint processing capabilities. Existing technologies for spatial analysis of multi-source data mostly employ fixed grid division or simple polygon overlay methods, failing to adaptively segment data based on its spatial distribution characteristics. This results in deviations between the generated regionalized data object sets and actual geographic boundaries (such as administrative regions and physical geographic units), reducing the reliability of subsequent analysis results in the spatial dimension. This is particularly true in scenarios requiring precise spatial positioning (such as urban planning and environmental monitoring), where imprecise spatial constraints can lead to erroneous decisions.

[0005] The data optimization mechanism is not perfect. Traditional methods lack spatial topology verification and cross-modal consistency assessment of data objects during the data fusion process. For example, boundary conflicts between data objects and spatial vector base maps are not effectively eliminated, which may lead to "spatial dislocation" in the fusion results; data screening based on feature matching may introduce a large amount of low-quality data, affecting the accuracy of the analysis results. In addition, the data quality assessment value is not fully utilized, and it is impossible to achieve reasonable weighting of high-confidence data through weighted fusion, which reduces the reliability of data fusion.

[0006] Visualization and analysis capabilities have significant limitations. Traditional data visualization methods typically only provide a static display of data, lacking dynamic overlay and interactive support for spatiotemporal dimensions. This makes it difficult to meet users' needs for in-depth analysis, such as switching data focus and zooming in on local features. Regarding dynamic trend forecasting, existing methods are mostly based on single data sources or simple statistical models, failing to construct effective spatiotemporal correlation analysis matrices and accurately capturing the spatiotemporal coupling characteristics between different data objects. This results in insufficient prediction accuracy for regional data evolution models, making it difficult to provide forward-looking decision-making support information.

[0007] Noise suppression technology needs to be improved. Multi-source time series data is susceptible to environmental interference during collection and transmission, generating noise data such as outliers and missing values. Traditional noise suppression methods (such as simple interpolation and fixed threshold filtering) are difficult to adapt to the complex noise patterns of multi-source data. They may result in the loss of effective information or residual noise, affecting the accuracy of subsequent feature extraction and fusion analysis.

[0008] The rapid development of smart cities, intelligent transportation, precision agriculture, and other fields has placed higher demands on the accuracy, real-time nature, and spatial analysis capabilities of multi-source data fusion. Existing technologies are limited in their ability to meet the complex data processing requirements of these emerging scenarios. There is an urgent need for new data processing methods that can effectively address multi-source heterogeneous data fusion, spatially constrained segmentation, data optimization, and visualization analysis to enhance cross-modal data-driven decision support capabilities. Summary of the Invention

[0009] The purpose of the present invention is to provide a data processing method and application based on multi-source data fusion to solve the problems raised in the above background technology.

[0010] To achieve the above object, the present invention provides the following technical solution: a data processing method based on multi-source data fusion, the method comprising:

[0011] Obtain multi-source time series data and spatial vector base map data corresponding to the target area;

[0012] Extracting correlation feature graphs of different data sources in the multi-source time series data through a preset heterogeneous data feature fusion model; and performing spatial constraint segmentation on the multi-source time series data based on the spatial vector base map data to generate a regionalized data object set;

[0013] The regionalized data object set is optimized according to the spatial vector base map data and the associated feature map, and the optimized data object set is used to generate a data fusion analysis result corresponding to the target area.

[0014] Preferably, the step of extracting correlation feature graphs of different data sources in the multi-source time series data by using a preset heterogeneous data feature fusion model includes:

[0015] Normalizing the multi-source time series data to generate a standardized data sequence; wherein the multi-source data includes sensor data, geospatial data, and user behavior data;

[0016] Performing cross-modal feature matching on the standardized data sequence through the heterogeneous data feature fusion model to generate an initial feature map corresponding to each data source;

[0017] Performing a spatiotemporal alignment operation on the initial feature map to generate a multi-source joint feature map;

[0018] A topological structure analysis is performed on the multi-source joint feature map to obtain a correlation feature map corresponding to the target area.

[0019] Preferably, the multi-source time series data includes static attribute data and dynamically changing data; and the step of performing spatial constraint segmentation on the multi-source time series data based on the spatial vector base map data includes:

[0020] The static attribute data and the dynamically changing data are superimposed into a composite data layer, and the composite data layer is segmented using an adaptive grid division method with the spatial vector base map data as a boundary to obtain the regionalized data object set.

[0021] Preferably, the step of generating a regionalized data object set includes:

[0022] For each data object, extract its corresponding static attribute features, dynamic change features and spatial distribution features;

[0023] The static attribute features include data type identification and data quality assessment value, and the dynamic change features include time series fluctuation index and trend correlation.

[0024] Preferably, the step of optimizing the regionalized data object set according to the spatial vector base map data and the associated feature map comprises:

[0025] Performing a spatial topology check on the regionalized data object set to eliminate data objects that conflict with the boundaries of the spatial vector base map data;

[0026] and, based on the cross-modal consistency index in the associated feature graph, screening data objects in the regionalized data object set whose feature matching degree is higher than a preset threshold;

[0027] And, weighted fusion is performed on the screened data objects according to the data quality evaluation value to generate an optimized data object set.

[0028] Preferably, the step of generating the data fusion analysis result corresponding to the target area using the optimized data object set includes:

[0029] Performing aggregation calculation on the optimized data object set to generate data distribution statistics corresponding to the target area;

[0030] And / or, mapping the optimized data object set to a spatial vector base map to generate a multi-source data fusion visualization map corresponding to the target area.

[0031] Preferably, after the step of obtaining multi-source time series data and spatial vector base map data corresponding to the target area, the method further includes:

[0032] Noise suppression processing is performed on the multi-source time series data to obtain denoised time series data; wherein the noise suppression processing includes outlier removal, data smoothing and missing value interpolation.

[0033] Preferably, performing aggregate calculation on the optimized data object set includes:

[0034] constructing a spatiotemporal correlation analysis matrix based on the data distribution statistics to extract spatiotemporal coupling features between different data objects in the target area;

[0035] A regional data evolution model is generated according to the spatiotemporal coupling characteristics, and a dynamic trend prediction result corresponding to the target area is output.

[0036] Preferably, generating a multi-source data fusion visualization map corresponding to the target area includes:

[0037] Performing multi-dimensional layered rendering on the optimized data object set to generate an interactive map with superimposed time and space dimensions;

[0038] The interactive graph supports data focus switching and local feature magnification display based on user operations.

[0039] Preferably, the present invention also includes an application of a data processing method based on multi-source data fusion, applying any of the above-mentioned data processing methods to a multi-domain data fusion analysis scenario to generate cross-modal data-driven decision support information.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] At the multi-source data feature fusion level, by normalizing multi-source time series data such as sensor data, geospatial data, and user behavior data, the differences in format and dimension between different data sources are eliminated, laying the foundation for subsequent cross-modal feature matching. The application of the heterogeneous data feature fusion model realizes cross-modal feature matching of standardized data sequences, generates initial feature maps corresponding to each data source, and constructs correlation feature maps that can reflect the inherent correlations of multi-source data through spatiotemporal alignment and topological structure analysis. This process effectively explores the potential connections between different data sources, breaking through the limitations of traditional methods in extracting cross-modal data correlation features, and enabling the fused data to more comprehensively and accurately characterize the comprehensive characteristics of the target area.

[0042] In terms of spatial constraint segmentation, static attribute data and dynamically changing data are superimposed into a composite data layer. Using the spatial vector basemap data as the boundary, an adaptive gridding method is used for segmentation to generate a set of regionalized data objects. This method fully considers the spatial distribution characteristics and actual geographic boundaries of the data, avoiding the disconnection between data and spatial boundaries caused by traditional fixed gridding. This allows the segmented regionalized data objects to better align with actual geographic units, improving the accuracy and rationality of spatial analysis. Furthermore, static attribute features, dynamically changing features, and spatial distribution features are extracted for each data object, providing multi-dimensional feature support for subsequent data optimization and analysis.

[0043] The data optimization phase achieves multi-level optimization of the regionalized data object set through spatial topology verification, feature matching screening based on cross-modal consistency indicators, and weighted fusion based on data quality assessment values. Spatial topology verification eliminates data objects that conflict with the spatial vector base map boundary, ensuring the spatial consistency of the data. The introduction of cross-modal consistency indicators effectively screens data objects with high feature matching, improving data reliability. The weighted fusion of data quality assessment values ​​differentiates data objects based on data quality, increasing the weight of high-credibility data, further improving the overall quality of the optimized data object set, and providing a higher-quality data foundation for subsequent analysis.

[0044] In terms of generating data fusion analysis results, data distribution statistics are generated through aggregate calculations, and a spatiotemporal correlation analysis matrix is ​​constructed to extract spatiotemporal coupling features, thereby generating a regional data evolution model and achieving accurate predictions of dynamic trends in the target area. This analysis method based on the spatiotemporal correlation of multi-source data breaks through the limitations of traditional single data sources or simple statistical models, and can more accurately capture the interactions and evolution laws between data objects, providing forward-looking predictive information for decision-making. At the same time, the optimized data object set is mapped to a spatial vector base map, and an interactive multi-source data fusion visualization map is generated through multi-dimensional layered rendering. This supports users to switch data focus and magnify local features, meeting users' needs for in-depth interactive analysis of data and improving the practicality of data visualization and decision support efficiency.

[0045] The noise suppression processing stage adopts multiple technologies such as outlier removal, data smoothing and missing value interpolation to perform comprehensive noise suppression on multi-source time series data, effectively removing the noise generated during data acquisition and transmission, retaining real and valid data information, improving the quality of original data, and providing more reliable data input for subsequent feature extraction and fusion analysis.

[0046] When applied to multi-domain data fusion analysis scenarios, this method can generate cross-modal data-driven decision support information, which can be widely used in smart city management, environmental monitoring, traffic flow analysis, commercial site selection, and other fields. By integrating multi-source data and conducting in-depth fusion analysis, it provides more comprehensive and accurate data analysis results for various fields, assisting decision makers in making scientific and reasonable decisions, and promoting the application and development of data-driven decision-making models in more fields, with significant social and economic benefits and technological innovation value. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a working principle diagram of the data processing method based on multi-source data fusion according to the present invention;

[0048] Figure 2 Flowchart for extracting correlation feature graphs for heterogeneous data feature fusion models;

[0049] Figure 3 This is a flowchart of spatial constraint segmentation of multi-source time series data based on spatial vector base map;

[0050] Figure 4 Flowchart optimized for regionalized data object sets;

[0051] Figure 5 Flowchart for generating a visual atlas of multi-source data fusion. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0053] See also Figure 1-Figure 5 The present invention relates to a data processing method based on multi-source data fusion, and the specific implementation steps are as follows:

[0054] Obtain multi-source time series data and spatial vector basemap data corresponding to the target area. Multi-source time series data includes data sources of different types and formats, such as sensor data, geospatial data, and user behavior data. These data have time series characteristics and can reflect the state and changes of the target area at different time points. Spatial vector basemap data is used to represent the geospatial structure and boundary information of the target area, providing a spatial reference for subsequent data processing.

[0055] Through the preset heterogeneous data feature fusion model, the correlation feature maps of different data sources in multi-source time series data are extracted. The specific process is as follows: first, the multi-source time series data is normalized, and data of different scales and units are converted into a unified standardized data sequence to facilitate subsequent feature extraction and analysis; then, the heterogeneous data feature fusion model is used to perform cross-modal feature matching on the standardized data sequence, and corresponding initial feature maps are generated for each data source. These initial feature maps respectively reflect the feature information of each data source; then, the initial feature maps are aligned in time and space to ensure that the feature maps of different data sources are consistent in time and space dimensions, thereby generating a multi-source joint feature map; finally, the multi-source joint feature map is topologically analyzed to explore the correlation relationship between different data sources and obtain the correlation feature map corresponding to the target area.

[0056] Based on the spatial vector basemap data, multi-source time series data is spatially constrained and segmented to generate a regionalized data object set. Since multi-source time series data includes static attribute data and dynamically changing data, the static attribute data and the dynamically changing data are first superimposed into a composite data layer. Then, using the spatial vector basemap data as the boundary, the composite data layer is segmented using an adaptive grid partitioning method. The target area is divided into multiple sub-regions, each corresponding to a data object, thus obtaining a regionalized data object set.

[0057] The regionalized data object set is optimized based on the spatial vector basemap data and the associated feature map, and the optimized data object set is used to generate data fusion analysis results corresponding to the target area. The specific optimization process includes: performing spatial topology verification on the regionalized data object set, checking the consistency of the data objects with the spatial vector basemap data boundaries, and eliminating data objects that conflict with the boundaries; based on the cross-modal consistency index in the associated feature map, filtering data objects in the regionalized data object set with feature matching degrees above a preset threshold to ensure high consistency between the features of the data objects; and performing weighted fusion on the filtered data objects based on the data quality assessment value to generate an optimized data object set. Finally, the optimized data object set is aggregated and calculated to generate data distribution statistics corresponding to the target area, or it is mapped to the spatial vector basemap to generate a multi-source data fusion visualization map corresponding to the target area.

[0058] Example 1:

[0059] After acquiring multi-source time series data and spatial vector basemap data corresponding to the target area, noise reduction processing is required to improve data quality. Multi-source time series data may contain noise due to sensor failure, transmission interference, acquisition errors, and other factors. Noise reduction processing includes outlier removal, data smoothing, and missing value interpolation. Outlier removal identifies and removes data points that significantly deviate from the normal range by setting a reasonable range. For example, in a temperature data series collected by a sensor, if the data at a certain time point is significantly higher or lower than the data at multiple consecutive time points and exceeds the normal fluctuation range, it can be identified as an outlier and removed. Data smoothing uses appropriate algorithms to reduce random data fluctuations. For example, for a frequency of access series in user behavior data, a moving average method can be used to average the number of points before and after each data point to ensure that the series reflects trends more smoothly. Missing value interpolation fills in missing data by using adjacent or related data. For example, if vegetation cover data for a certain area in geospatial data is missing, it can be filled in by interpolating vegetation cover data from adjacent areas.

[0060] After noise suppression is completed and denoised time series data is obtained, the process begins with extracting associated feature graphs using a pre-set heterogeneous data feature fusion model. First, the multi-source time series data is normalized. Because different data sources have significantly different dimensions and scales, such as temperature units in degrees Celsius for sensor data, access counts for user behavior data, and distance units in meters for geospatial data, it is necessary to convert them into standardized data sequences with a uniform scale. Specifically, this involves linearly transforming the raw data from each data source so that it falls within a specific interval (e.g., [0, 1]) to facilitate subsequent cross-modal feature matching.

[0061] The standardized data sequence is then subjected to cross-modal feature matching using a heterogeneous data feature fusion model. This model, capable of identifying potential correlations between data of different modalities, extracts features from different data sources, such as sensor data, geospatial data, and user behavior data, to generate initial feature maps. For sensor data, features such as the fluctuation pattern of the time series, the frequency of peaks, and the duration of valleys are extracted. For geospatial data, features such as the density of spatial distribution, shape index, and distance relationships to geographical elements such as roads and rivers are extracted. For user behavior data, features such as the temporal patterns of behavior, spatial clustering, and the frequency of transitions between different behavior types are extracted. Each data source corresponds to an initial feature map, and the elements in the map reflect the key feature information of that data source.

[0062] The initial feature maps are then subjected to a spatiotemporal alignment operation. Because data collection times and spatial locations may differ between different data sources, the temporal axes of the initial feature maps must be unified to a common temporal coordinate system for the target region, ensuring consistent feature correspondence at the same time points. In the spatial dimension, based on the coordinate system of the spatial vector basemap data, feature maps from data sources other than geographic spatial data are transformed to align the spatial locations of all feature maps with the spatial vector basemap, generating a multi-source joint feature map that integrates feature information from multiple sources at the same spatiotemporal location.

[0063] Finally, a topological analysis of the multi-source joint feature graph is performed. By studying the connectivity, clustering patterns, and path orientation of the feature points within the feature graph, the associations between features from different data sources are explored. For example, the spatial proximity between sensor data feature points and geospatial data feature points can be analyzed to determine whether sensor data changes are associated with a specific geographic region. The temporal sequence of user behavior data feature points and sensor data feature points can be analyzed to determine whether user behavior is influenced by sensor monitoring indicators. This results in a correlation feature graph corresponding to the target region, clearly demonstrating the inherent connections between multi-source data and providing a key basis for subsequent data processing.

[0064] Example 2:

[0065] Multi-source time series data consists of static attribute data and dynamically changing data. When performing spatial constraint segmentation based on spatial vector basemap data, the two types of data must first be superimposed into a composite data layer. Static attribute data is relatively stable basic information within the target area, such as the location, type, and floor area of ​​buildings, land use type, property rights, and the distribution of public facilities. This data does not change frequently over a given period of time. Dynamically changing data is data that is continuously updated over time, such as air quality index and traffic flow data collected in real time by sensors, as well as user movement trajectories and consumption behavior records. It can reflect the real-time status or short-term trends of the target area. By superimposing static attribute data and dynamically changing data at a spatial location, each spatial location point possesses both static attribute and dynamic change information, forming a composite data layer with rich dimensions, providing a more comprehensive data foundation for subsequent spatial segmentation.

[0066] An adaptive gridding method is used to segment the composite data layer, using spatial vector basemap data as boundaries. The spatial vector basemap data defines the geographic extent and spatial structure of the target area, such as administrative boundaries, block outlines, and physical geographic boundaries, serving as spatial constraints for data segmentation. The adaptive gridding method automatically adjusts the size and shape of the grid based on characteristics such as the distribution density and spatial heterogeneity of the data in the composite data layer. In areas with dense data distribution and complex feature variations (such as commercial centers and transportation hubs), smaller grids are used to more precisely capture data differences. In areas with sparse data distribution and relatively uniform features (such as suburban and rural areas), larger grids are used to reduce data processing. This method segments the composite data layer into multiple spatially continuous subregions, each corresponding to a data object, thus generating a regionalized set of data objects. Each data object contains all static attribute data and dynamically changing data within the subregion, forming a relatively independent spatial data unit.

[0067] In the process of generating a regionalized data object set, it is necessary to extract static attribute features, dynamic change features, and spatial distribution features for each data object. The data type identifier in the static attribute features is used to clarify the source and category of the data. For example, temperature data collected by sensors is identified as "environmental monitoring-temperature", road layers in geospatial data are identified as "infrastructure-roads", and shopping behavior records in user behavior data are identified as "user activities-consumption", so that different types of data can be quickly identified and managed in subsequent processing. The data quality assessment value is a comprehensive evaluation of data reliability. By evaluating indicators such as data completeness (such as whether fields are missing), accuracy (such as the degree of consistency with field survey data), and consistency (such as the consistency of descriptions of the same entity by different data sources), the quality of each data object is quantitatively scored to provide a basis for data optimization.

[0068] The time series volatility index, a dynamic change feature, measures the severity of data changes over time. For example, traffic flow data for a certain area may show significant peaks during rush hour in the morning and evening, while traffic flow is lower during off-peak hours. By calculating the amplitude and frequency of fluctuations in this data series, an index reflecting its volatility can be obtained. Trend correlation is used to analyze the consistency of trends in a data series with other related series. For example, analyzing the time series of air quality data for a certain area and production emission data from surrounding industrial enterprises can determine whether there is a positive or negative correlation between the two trends and calculate the degree of correlation through statistical methods.

[0069] Spatial distribution characteristics primarily include the spatial location and distribution range of data objects. Spatial location is represented by geographic coordinates (such as longitude and latitude), defining the specific location of the data object within the target area. Distribution range is described by geometric figures such as minimum enclosing rectangles and polygonal boundaries, reflecting the size of the spatial area covered by the data object. Furthermore, by combining functional zoning information (such as commercial, residential, and industrial areas) within the spatial vector basemap, each data object is assigned a corresponding spatial function label, enabling data classification and analysis based on regional functions. For example, this allows for analyzing the correlation between user consumption behavior data and traffic flow data within a commercial area.

[0070] Through these steps, we achieve spatially constrained segmentation and data object feature extraction for multi-source time series data, forming a structured set of regionalized data objects, laying the foundation for subsequent data optimization and fusion analysis. Each data object not only contains spatial location information but also integrates static attributes and dynamically changing features. This allows for a comprehensive and detailed reflection of the data characteristics of different sub-regions within the target region, providing a rich information carrier for the deep fusion of multi-source data.

[0071] Example 3:

[0072] When optimizing the regionalized data object set based on the spatial vector basemap data and the associated feature map, a spatial topology check must be performed first. The spatial vector basemap data clearly defines the geographic boundaries and spatial structure of the target area, such as administrative division boundaries, natural geographic feature boundaries, etc. During the check process, each data object's spatial extent is checked one by one to see if it is consistent with the basemap boundary. For example, it is determined whether the geometric extent of the data object is completely within the target area boundary, or whether it conflicts with the spatial direction of linear features such as roads and rivers in the basemap. Data objects that exceed the boundary or conflict with the spatial position of key geographic features in the basemap are determined to be spatial topology conflict data and are eliminated to ensure that the spatial position of the remaining data objects conforms to the actual geographic scope of the target area, avoiding deviations in subsequent analysis results due to spatial position errors.

[0073] After completing the spatial topology verification, data objects are screened based on the cross-modal consistency indicators in the association feature graph. The association feature graph generates indicators that reflect the consistency of features from different data sources by analyzing the cross-modal feature matching relationship of multi-source data, such as the degree of feature matching between sensor data and geospatial data in a specific area, the spatiotemporal correlation strength between user behavior data and sensor data, etc. A feature matching threshold is preset (e.g., set to a reasonable value based on data fusion requirements), and for each data object in the regionalized data object set, the matching score of the multi-source data features involved in the association feature graph is calculated. Data objects with scores higher than the preset threshold are screened out, data with high feature matching is retained, and data objects that are difficult to fuse due to excessive feature differences between data sources are excluded, thereby improving the feature consistency of the data object set and providing a more reliable data foundation for subsequent fusion analysis.

[0074] Filtered data objects are weighted and fused based on their data quality assessment values. Data quality assessment values ​​are comprehensive quantitative evaluations of data integrity, accuracy, and timeliness. For example, integrity can be measured by the percentage of missing data fields, accuracy by the error in comparison with authoritative data sources, and timeliness by the length of time since data collection. For each filtered data object, the higher the data quality assessment value, the more reliable the data, and the greater the weight assigned to it in weighted fusion. During weighted fusion, multi-source feature information of the data objects is comprehensively considered, and features of different data objects are combined according to weights. For example, the static attribute features and dynamic change features of multiple data objects in the same spatial area are weighted averaged or merged to generate an optimized set of data objects. This approach ensures that high-quality data objects account for a larger proportion of the fusion results, improving the quality and usability of the overall data object set and ensuring the generation of accurate data fusion analysis results.

[0075] Through the three steps of spatial topology verification, feature matching screening, and weighted fusion described above, the regionalized data object set is optimized. This process closely combines the geographic constraints of the spatial vector basemap with the cross-modal feature associations of the associated feature map, ensuring both the spatial accuracy of the data objects and the consistency of features across them. Furthermore, data quality assessments are used to differentiate the data, enabling the optimized data object set to more realistically and comprehensively reflect the multi-source data characteristics of the target region, providing high-quality data support for subsequent data aggregation calculations and visualization analysis.

[0076] Example 4:

[0077] When generating data fusion analysis results for the target area using the optimized data object set, aggregation calculations are first performed to generate data distribution statistics. Aggregation calculations apply statistical operations to various types of data within the optimized data object set, extracting metrics that reflect the overall characteristics of the data. For example, for noise decibel data collected by sensors, the average, maximum, and minimum values ​​within the target area are calculated to describe the overall distribution of noise levels. For user behavior data, the total number of visits and frequency over different time periods are calculated to reveal temporal patterns in user activity. These statistics clearly and concisely present the centralization trend, dispersion, and distribution pattern of the data within the target area, providing foundational information for preliminary analysis.

[0078] On the basis of generating data distribution statistics, a spatiotemporal correlation analysis matrix is ​​further constructed to extract spatiotemporal coupling characteristics. The spatiotemporal correlation analysis matrix is ​​used to characterize the degree of mutual correlation between different data objects in the time and space dimensions. In the specific operation, for each two data objects, it is analyzed whether their changing trends in the time series are synchronized, and whether there is a proximity or clustering relationship in spatial position. For example, the air quality data objects and traffic flow data objects in a certain area are analyzed to observe whether the fluctuations of air quality indicators are temporally consistent with the peak hours of traffic flow, and at the same time check whether the spatial distribution of the two is concentrated in the same or adjacent areas. Through this analysis, the spatiotemporal coupling patterns between data objects are identified, such as a positive correlation of synchronous changes, a negative correlation of one increasing while the other decreases, or a clustered distribution feature in space.

[0079] Based on the extracted spatiotemporal coupling features, a regional data evolution model is generated and dynamic trend prediction results are output. The regional data evolution model predicts future data trends within a target region by simulating the interactions and evolution patterns of data objects across spatiotemporal dimensions. For example, if the spatiotemporal coupling features indicate a strong positive correlation between commercial customer flow data in a certain area and the distribution density of surrounding restaurants, the model can predict the movement direction of commercial hotspots or the increase or decrease in customer flow within the region over a period of time based on the current store distribution and customer flow trends. This model integrates the spatiotemporal characteristics of historical data and can provide forward-looking reference information for decisions such as urban planning and resource allocation.

[0080] In addition to aggregation calculations, the optimized data object set is mapped to a spatial vector basemap to generate a multi-source data fusion visualization. During the mapping process, the corresponding data types are annotated at the corresponding locations on the spatial vector basemap based on the spatial distribution characteristics of the data objects, forming a visual presentation of the fused multi-source data. For example, water quality data collected by sensors is annotated at the spatial location corresponding to the river, and user travel check-in data is annotated at the location of the scenic spot, allowing different types of data to be displayed in a unified geographic spatial coordinate system.

[0081] When generating visualization maps, a multi-dimensional layered rendering technique is employed. Different characteristic dimensions of data objects (such as time, space, and data type) are mapped to separate layers, with each layer carrying a specific type of information. For example, the first layer displays the basic elements of the geospatial basemap (such as roads and buildings), the second layer overlays real-time sensor data (such as temperature and humidity), and the third layer displays the distribution density of user behavior data (such as areas of crowd concentration). By adjusting the transparency and display order of each layer, a multi-dimensional data overlay is achieved, making it easier for users to observe the spatial correlations and hierarchical relationships between different data.

[0082] The visualization map supports user interactive operations, including data focus switching and local feature magnification display. The data focus switching function allows users to highlight data objects of a specific type or specific area by clicking or selecting operations, while weakening the display effect of other data, so as to focus on the data dimension of interest. For example, users can choose to display only a certain type of sensor data and observe its distribution pattern in the target area; the local feature magnification display function zooms in on a specific area of ​​the map through mouse zooming or frame selection operations, displaying detailed attribute information (such as specific values, feature descriptions) of the data objects in the area, helping users discover subtle features and changing trends of the data in the local space. This interactive visualization method makes the data fusion analysis results more intuitive and easy to understand, facilitating users to conduct in-depth data analysis and decision support.

[0083] Through the aforementioned steps of aggregation calculation, spatiotemporal correlation analysis, model prediction, and visualization map generation, the transformation from an optimized data object set to data fusion analysis results is achieved. This process not only provides statistical descriptions and trend predictions of the data, but also enhances the readability and operability of the data through visualization. This provides comprehensive and intuitive information support for decision-making in multiple fields, fully demonstrating the advantages of multi-source data fusion in mining data value and assisting decision-making.

[0084] Example 5:

[0085] When generating a regionalized data object set, each data object must extract static attribute features, dynamic change features, and spatial distribution features. The data type identifier in the static attribute features uses a specific prefix to distinguish the data source. For example, sensor data begins with "SD-", geospatial data begins with "GD-", and user behavior data begins with "UD-". The suffix is ​​combined with the specific device number, layer code, or user ID to form a unique identifier (for example, "SD-001" represents sensor data numbered 001). Data quality assessment values ​​are calculated by comprehensively considering completeness, accuracy, timeliness, and consistency. Completeness is assessed as the ratio of the number of non-missing fields to the total number of fields, accuracy is assessed as the matching score between the data and the authoritative source, timeliness is assessed as the normalized value of (current time - data collection time), and consistency is assessed as the ratio of the number of entities consistently described by multiple sources to the total number of entities. The final assessment value is the weighted average of the scores for each dimension.

[0086] The time series volatility index within the dynamic change feature measures the degree of volatility by calculating the relative value of the standard deviation of the data series to its mean. The trend correlation uses the Pearson correlation coefficient to calculate the trend consistency between the data series and the reference series. Spatial distribution feature extraction includes the longitude and latitude coordinates of the geometric center of the data object (e.g., (lon, lat)), the coordinate range of the minimum enclosing rectangle (e.g., (xmin, ymin, xmax, ymax)), and functional zoning labels (e.g., "commercial area" and "residential area") determined based on the spatial vector basemap.

[0087] Spatial autocorrelation analysis is introduced in the aggregation calculation, and Moran's index (Moran's I) is used to detect the spatial aggregation of data. The formula is:

[0088]

[0089] in, is the number of data objects, is the spatial weight matrix The sum of the elements of Represents a data object and The spatial weights of (usually based on distance or contiguity), For data objects The attribute value of is the average value of the attribute value. The Moran index ranges from [-1, 1], where positive values ​​indicate spatial aggregation, negative values ​​indicate spatial dispersion, and 0 indicates random distribution.

[0090] When rendering in multiple dimensions, numerical data (such as temperature) is mapped to higher and lower values ​​using a color gradient (e.g., blue to red). Categorical data (such as land use type) is distinguished by different icons (e.g., □ for built-up land, △ for green space). Spatiotemporal overlays are implemented using a timeline control. Users can slide to view data distribution at different time points, and layer transparency can be adjusted to visualize the effect of multi-layered data overlays. Interactive maps support switching the focus layer (e.g., displaying only the sensor data layer) by clicking on the legend. Zooming in or out with the mouse wheel or selecting an area allows for zooming in, displaying detailed properties of data objects (e.g., temperature values, functional zoning labels).

[0091] The scale of the spatial vector basemap is selected based on the application scenario. City-level applications use a large-scale basemap of 1:5000 to show details such as buildings and roads; regional applications use a medium-scale basemap of 1:25000 to reflect macro features such as topography and administrative divisions. Basemaps must be updated regularly to ensure currency. Updates include changes in geographic features such as new roads and building demolitions.

[0092] In the noise suppression process, the IQR method is used to remove outliers and calculate the interquartile range. , will be lower than or higher The data points are considered as outliers (where is the 25th percentile, The 75th percentile is used for data smoothing. Exponential smoothing is used to give higher weight to recent data. Missing value interpolation is performed based on the data trend using linear interpolation (suitable for linear trend data) or neighboring value filling (suitable for cases with few missing points).

[0093] Through the above steps, a complete process from data feature extraction, spatial analysis to visual interaction is realized. Combined with the spatial autocorrelation analysis formula, the spatial distribution pattern of data is deeply explored, providing a systematic processing framework for multi-source data fusion.

[0094] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0095] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A data processing method based on multi-source data fusion, characterized in that: include: Obtain multi-source time series data and spatial vector base map data corresponding to the target area; Extracting correlation feature graphs of different data sources in the multi-source time series data through a preset heterogeneous data feature fusion model; and performing spatial constraint segmentation on the multi-source time series data based on the spatial vector base map data to generate a regionalized data object set; The regionalized data object set is optimized according to the spatial vector base map data and the associated feature map, and the optimized data object set is used to generate a data fusion analysis result corresponding to the target area.

2. The data processing method based on multi-source data fusion according to claim 1, characterized in that: The step of extracting correlation feature graphs of different data sources in the multi-source time series data by using a preset heterogeneous data feature fusion model includes: Normalizing the multi-source time series data to generate a standardized data sequence; wherein the multi-source data includes sensor data, geospatial data, and user behavior data; Performing cross-modal feature matching on the standardized data sequence through the heterogeneous data feature fusion model to generate an initial feature map corresponding to each data source; Performing a spatiotemporal alignment operation on the initial feature map to generate a multi-source joint feature map; A topological structure analysis is performed on the multi-source joint feature map to obtain a correlation feature map corresponding to the target area.

3. The data processing method based on multi-source data fusion according to claim 1, characterized in that: The multi-source time series data includes static attribute data and dynamically changing data; The step of performing spatial constraint segmentation on the multi-source time series data based on the spatial vector base map data includes: The static attribute data and the dynamically changing data are superimposed into a composite data layer, and the composite data layer is segmented using an adaptive grid division method with the spatial vector base map data as a boundary to obtain the regionalized data object set.

4. The data processing method based on multi-source data fusion according to claim 3 is characterized in that: The steps for generating a regionalized data object set include: For each data object, extract its corresponding static attribute features, dynamic change features and spatial distribution features; The static attribute features include data type identification and data quality assessment value, and the dynamic change features include time series fluctuation index and trend correlation.

5. The data processing method based on multi-source data fusion according to claim 4 is characterized in that: The step of optimizing the regionalized data object set according to the spatial vector base map data and the associated feature map comprises: Performing a spatial topology check on the regionalized data object set to eliminate data objects that conflict with the boundaries of the spatial vector base map data; and, based on the cross-modal consistency index in the associated feature graph, screening data objects in the regionalized data object set whose feature matching degree is higher than a preset threshold; And, weighted fusion is performed on the screened data objects according to the data quality evaluation value to generate an optimized data object set.

6. The data processing method based on multi-source data fusion according to claim 1, characterized in that: The step of generating a data fusion analysis result corresponding to the target area using the optimized data object set includes: Performing aggregation calculation on the optimized data object set to generate data distribution statistics corresponding to the target area; And / or, mapping the optimized data object set to a spatial vector base map to generate a multi-source data fusion visualization map corresponding to the target area.

7. The data processing method based on multi-source data fusion according to claim 1, characterized in that: After the step of obtaining multi-source time series data and spatial vector base map data corresponding to the target area, the method further includes: Noise suppression processing is performed on the multi-source time series data to obtain denoised time series data; wherein the noise suppression processing includes outlier removal, data smoothing and missing value interpolation.

8. The data processing method based on multi-source data fusion according to claim 6, characterized in that: The performing aggregate calculation on the optimized data object set includes: constructing a spatiotemporal correlation analysis matrix based on the data distribution statistics to extract spatiotemporal coupling features between different data objects in the target area; A regional data evolution model is generated according to the spatiotemporal coupling characteristics, and a dynamic trend prediction result corresponding to the target area is output.

9. The data processing method based on multi-source data fusion according to claim 6, characterized in that: Generating a multi-source data fusion visualization map corresponding to the target area includes: Performing multi-dimensional layered rendering on the optimized data object set to generate an interactive map with superimposed time and space dimensions; The interactive graph supports data focus switching and local feature magnification display based on user operations.

10. An application of a data processing method based on multi-source data fusion, characterized in that: The data processing method according to any one of claims 1 to 9 is applied to a multi-domain data fusion analysis scenario to generate cross-modal data-driven decision support information.

Images