Multi-source spatio-temporal data fusion air quality prediction method and system

Abstract

The invention discloses a multi-source spatio-temporal data fusion air quality prediction method and system, relates to the technical field of data fusion, and comprises the steps of performing interpolation processing on a data missing region, considering data comprehensiveness, performing complementation interpolation on data on a pollution map, and providing a reliable basis for subsequent pollution analysis and air quality prediction. And the pollution characteristic condition of a pollution source is evaluated to dynamically adapt to and adjust the resolution of space and time in a targeted manner, so that the accuracy of map description is ensured, and the requirements of resources and simulation are balanced. The comprehensive contribution degree of each type of pollution diffusion cause features is defined, the pollution diffusion cause features are fused by means of the comprehensive contribution degree, the fused features are described from the three scales of time, space and physics, and the comprehensiveness of analysis is guaranteed. A pollution diffusion-air quality causal model is constructed based on fusion features, and features in multi-source directions are effectively integrated, so that the precision of air quality prediction is improved, and complex and variable air quality prediction requirements are met.

Description

Technical Field

[0001] The present invention relates to the technical field of data fusion, and in particular to an air quality prediction method and system based on multi-source spatiotemporal data fusion. Background Art

[0002] The air quality prediction solution, which integrates multi-source spatiotemporal data, aims to integrate data from various sources (such as satellite remote sensing, ground monitoring stations, and meteorological departments) and at different temporal and spatial scales to comprehensively capture the complex factors affecting air quality. This solution uses advanced data fusion techniques to preprocess, extract features, and fuse multi-source data. Combined with spatiotemporal analysis techniques, it deeply explores the spatiotemporal correlations and dynamic patterns within the data. Furthermore, predictive models such as machine learning and deep learning are used to model and analyze the fused data, achieving accurate air quality predictions. This solution leverages the complementary nature of multi-source data, improving the robustness and accuracy of the prediction model and providing strong support for air quality management and pollution prevention and control.

[0003] In the existing technology, it is impossible to effectively integrate the features of multiple sources, and the adaptability of data fusion is poor, resulting in low accuracy and reliability of air quality prediction, which cannot meet the complex and changeable air quality prediction needs.

[0004] Therefore, how to improve the accuracy and reliability of air quality prediction is a technical problem that needs to be solved. Summary of the Invention

[0005] The purpose of the present invention is to solve the problems of poor adaptability of data fusion and low accuracy and reliability of air quality prediction in the prior art, and to propose an air quality prediction method based on multi-source spatiotemporal data fusion, the method comprising: Obtain the pollution sources and their locations within the area to be analyzed, construct a pollution map, collect multi-source data according to the pollution map, determine the data missing areas in the pollution map, and perform interpolation processing on the data missing areas; Analyze the pollution characteristics of pollution sources on the pollution map to perform adaptive grid processing on the spatial and temporal resolution of different areas; Simulate the pollutant diffusion path of the pollution source, extract the pollution diffusion characteristics, define the comprehensive contribution of each type of pollution diffusion characteristics, and fuse the pollution diffusion characteristics based on the comprehensive contribution to obtain the fusion characteristics; Based on the fusion features, a causal model of pollution diffusion-air quality is constructed, and air quality prediction is achieved through the causal model of pollution diffusion-air quality.

[0006] In some embodiments of the present application, determining the data missing area in the pollution map includes: Mark the monitoring stations on the pollution map, calculate the spatial distribution density of the monitoring stations and the spatial correlation of the monitoring station data, and preliminarily screen out data-missing areas based on the spatial distribution density of the monitoring stations and the spatial correlation of the monitoring station data; Data missing areas are verified to determine the data missing areas in the final pollution map.

[0007] In some embodiments of the present application, interpolation processing is performed on the data missing area, including: The multi-source data in the area to be analyzed are divided into main variable data and auxiliary variable data. The main variable data is pollution-related data, and the auxiliary variable data is elevation-related data. The areas other than the data missing areas in the area to be analyzed are the data known areas. The similarity of geographical features between the data known areas and the data missing areas is calculated to find the data known area closest to the data missing area. Calculate the semivariogram of the known data area based on the primary variable data and the auxiliary variable data of the data known area closest to the data missing area, and calculate the semivariogram of the entire area to be analyzed based on the primary variable data and the auxiliary variable data of the entire area to be analyzed, and record the two semivariograms as the first semivariogram and the second semivariogram respectively; The third semivariogram of the data missing area is obtained based on the first and second semivariograms; The data-missing area is divided into multiple cells through the third semivariogram, and the third semivariogram of each cell is re-evaluated. A co-kriging model is constructed for each cell. The co-kriging model introduces the directional effect and scale effect of the third semivariogram under different cells, and interpolates each cell, thereby completing the interpolation of the main variable data in the data-missing area.

[0008] In some embodiments of the present application, the pollution characteristics of the pollution sources on the pollution map are analyzed to perform adaptive grid processing on the spatial and temporal resolution of different areas, including: The pollution map is divided into multiple regions according to the location of pollution sources. The average distance between monitoring stations, terrain complexity, number of pollution sources, and pollutant parameters in each region are counted. Pollutant parameters include pollutant toxicity parameters, emission parameters, meteorological parameters, and flow range. The pollution intensity in the region is calculated based on the pollutant toxicity parameters, emission parameters, meteorological parameters, flow range, and number of pollution sources. The initial value of the spatial resolution of the region is set based on the average distance between monitoring stations in the region, the complexity of the terrain, and the real-time value of the pollution intensity. The change value of the pollution intensity in the region is monitored according to the first preset monitoring period, and the initial value of the spatial resolution of the region is dynamically adjusted. The pollution source types, meteorological parameters, and pollutant categories and pollutant change rates generated by all pollution sources in each area are counted to set the initial value of the regional time resolution. The meteorological parameters and pollutant change rates in the area are monitored according to the second preset monitoring period, and the initial value of the regional time resolution is dynamically adjusted. Adaptive grid processing is performed for each region by dynamically adjusting the spatial and temporal resolution within the region.

[0009] In some embodiments of the present application, the pollutant diffusion path of the pollution source is simulated to extract the pollution diffusion characteristics, including: Establish a diffusion model for the pollution sources on the pollution map, and establish different types of diffusion models for different pollution sources to simulate the diffusion path of pollutants from the pollution sources; In the process of simulating the pollutant diffusion path of the pollution source, the pollution diffusion characteristics at the spatial scale, time scale and physical scale are extracted.

[0010] In some embodiments of the present application, the comprehensive contribution of each type of pollution diffusion factor characteristic is defined, including: Calculate the original contribution of each pollution diffusion factor feature to air quality, analyze the interaction between pollution diffusion factors, and divide pollution diffusion factors into independent features and interactive features based on the interaction; For independent features, the original contribution of the independent feature is used as the comprehensive contribution of the feature to air quality; For interactive features, the original contribution of the interactive feature under the interactive relationship is adjusted according to the interaction, and the adjusted original contribution is used as the comprehensive contribution of the feature to air quality.

[0011] In some embodiments of the present application, the pollution diffusion factor features are fused by means of comprehensive contribution to obtain fused features, including: According to the comprehensive contribution of pollution diffusion characteristics, the combined weights of the pollution diffusion characteristics at the spatial scale, time scale and physical scale are assigned; The pollution diffusion characteristics are fused according to the spatial scale, temporal scale and physical scale respectively to obtain the fusion characteristics at the three scales.

[0012] In some embodiments of the present application, a causal model of pollution diffusion-air quality is constructed based on fusion features, including: The fusion features of the three scales, namely spatial scale fusion feature, temporal scale fusion feature and physical scale fusion feature, are recorded as spatial scale fusion feature, temporal scale fusion feature and physical scale fusion feature respectively, and a causal graph between the spatial scale fusion feature, temporal scale fusion feature, physical scale fusion feature and air quality is constructed; A causal model of pollution diffusion-air quality is established based on the causal graph, and the causal model parameters are optimized.

[0013] Correspondingly, this application also provides an air quality prediction system based on multi-source spatiotemporal data fusion, including: The first module is used to obtain the pollution sources and their locations within the area to be analyzed, construct a pollution map, collect multi-source data according to the pollution map, determine the data missing areas in the pollution map, and perform interpolation processing on the data missing areas; The second module is used to analyze the pollution characteristics of pollution sources on the pollution map to perform adaptive grid processing on the spatial and temporal resolution of different areas; The third module is used to simulate the pollutant diffusion path of the pollution source, extract the pollution diffusion factor characteristics, define the comprehensive contribution of each type of pollution diffusion factor characteristics, and fuse the pollution diffusion factor characteristics based on the comprehensive contribution to obtain the fused characteristics; The fourth module is used to build a causal model of pollution diffusion-air quality based on fusion features, and to realize air quality prediction through the causal model of pollution diffusion-air quality.

[0014] The present invention has the following beneficial effects: 1. Interpolate areas with missing data. Considering data comprehensiveness, interpolate data on the pollution map to provide a reliable foundation for subsequent pollution analysis and air quality prediction. Analyze the pollution characteristics of pollution sources on the pollution map to adaptively grid the spatial and temporal resolution of different regions. Assess the pollution characteristics of pollution sources to dynamically adjust spatial and temporal resolution in a targeted manner, ensuring map description accuracy and balancing resource and simulation needs.

[0015] 2. Define the comprehensive contribution of each type of pollution diffusion factor characteristic. Using this comprehensive contribution, fuse these characteristics and describe the fused features at three scales: temporal, spatial, and physical, ensuring comprehensive analysis. Based on these fused features, a causal model for pollution diffusion and air quality is constructed, effectively integrating features from multiple sources and improving the adaptability of data fusion, thereby enhancing the accuracy of air quality predictions and meeting complex and ever-changing air quality forecasting needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of the air quality prediction method based on multi-source spatiotemporal data fusion proposed in the present invention; Figure 2 This is a structural diagram of the air quality prediction system based on multi-source spatiotemporal data fusion proposed in the present invention. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0018] Reference Figure 1 ,The air quality prediction method based on multi-source spatiotemporal data fusion, includes the following steps: Step S101 , obtaining pollution sources and pollution source locations within the area to be analyzed, constructing a pollution map, collecting multi-source data according to the pollution map, determining data missing areas in the pollution map, and performing interpolation processing on the data missing areas.

[0019] In this embodiment, the locations of pollution sources are obtained through channels such as environmental protection departments and enterprise declarations, including factories, major transportation routes, and construction sites. A pollution map is constructed using GIS software (such as QGIS) to visualize the pollution source locations into a pollution map. The map includes annotations such as pollution source symbols, names, and emission types. In addition to the monitoring stations shown on the map, there may be other key locations where pollution levels need to be determined, necessitating interpolation for areas with missing data.

[0020] In some embodiments of the present application, determining the data missing area in the pollution map includes: Mark the monitoring stations on the pollution map, calculate the spatial distribution density of the monitoring stations and the spatial correlation of the monitoring station data, and preliminarily screen out data-missing areas based on the spatial distribution density of the monitoring stations and the spatial correlation of the monitoring station data; Data missing areas are verified to determine the data missing areas in the final pollution map.

[0021] In this embodiment, the locations of all air quality monitoring stations are marked on the pollution map. Methods such as kernel density estimation are used to calculate the spatial distribution density of the monitoring stations and identify sparse areas of monitoring stations. Methods such as Moran's I are used to calculate the spatial correlation of the monitoring station data and identify the spatial aggregation or discrete characteristics of the data. Combining the spatial distribution density and spatial correlation of the monitoring stations, possible data missing areas are preliminarily screened out, such as areas with sparse monitoring stations and low data correlation. The preliminarily screened data missing areas are verified by determining whether the data in the area is actually missing, and the final data missing areas are determined.

[0022] In some embodiments of the present application, interpolation processing is performed on the data missing area, including: The multi-source data in the area to be analyzed are divided into main variable data and auxiliary variable data. The main variable data is pollution-related data, and the auxiliary variable data is elevation-related data. The areas other than the data missing areas in the area to be analyzed are the data known areas. The similarity of geographical features between the data known areas and the data missing areas is calculated to find the data known area closest to the data missing area. Calculate the semivariogram of the known data area based on the primary variable data and the auxiliary variable data of the data known area closest to the data missing area, and calculate the semivariogram of the entire area to be analyzed based on the primary variable data and the auxiliary variable data of the entire area to be analyzed, and record the two semivariograms as the first semivariogram and the second semivariogram respectively; The third semivariogram of the data missing area is obtained based on the first and second semivariograms; The data-missing area is divided into multiple cells through the third semivariogram, and the third semivariogram of each cell is re-evaluated. A co-kriging model is constructed for each cell. The co-kriging model introduces the directional effect and scale effect of the third semivariogram under different cells, and interpolates each cell, thereby completing the interpolation of the main variable data in the data-missing area.

[0023] In this embodiment, conventional cokriging improves interpolation accuracy by introducing auxiliary variables. Its core approach is to exploit the spatial correlation between the primary and auxiliary variables. However, traditional cokriging assumes that spatial correlation is homogeneous, meaning that the structure and parameters of spatial correlation remain constant across the entire study area. However, in real-world applications, spatial heterogeneity is ubiquitous, meaning that spatial correlation in different regions may exhibit different characteristics. Therefore, correcting for spatial correlation by taking spatial heterogeneity into account is crucial for improving cokriging interpolation accuracy. Primary variable data is collected, such as concentration data for pollutants like PM2.5 and NO2, typically from ground-based monitoring stations. Auxiliary variable data is collected, such as DEMs (digital elevation models) and road network density, which can be obtained from open-source platforms (such as OpenStreetMap and USGS) or relevant organizations.

[0024] To determine the semivariogram (third semivariogram) for areas with missing data, the semivariogram of the entire area under analysis (second semivariogram, such as a Gaussian or exponential model) and the semivariogram of similar areas (first semivariogram, such as a Gaussian or exponential model) are combined. The first and second semivariograms are weighted and summed to obtain the third semivariogram. Semivariograms of the primary and auxiliary variables are calculated to analyze the directionality and scale effects of spatial variability. The semivariogram reveals the variability and correlation of data across distances and directions, providing a basis for quantifying spatial heterogeneity. Based on the results of spatial autocorrelation and semivariogram analysis, the study area is divided into several subregions. Each subregion should have relatively homogeneous spatial characteristics, meaning that the structure and parameters of spatial correlation can be assumed to be constant within the subregion. A cokriging model is constructed within each subregion. For each subregion, the variogram model parameters are reestimated to reflect the spatial correlation characteristics within the subregion. Directional effects (such as anisotropy) and scale effects are considered when constructing the cokriging model. Directional effects can be modeled by introducing anisotropic variograms to reflect differences in spatial correlation in different directions. Scale effects can be modeled by multiscale analysis to capture spatial heterogeneity at different scales.

[0025] Construct a covariance matrix: Construct a covariance matrix between the primary and auxiliary variables based on the variogram model and model parameters.

[0026] Solve the Kriging equations: Use the covariance matrix and the data of the known points to solve the Kriging equations and obtain the predicted values ​​of the unknown points.

[0027] Output interpolation results: including predicted values ​​and prediction variances. The prediction variance is used to assess the uncertainty of the interpolation results.

[0028] Step S102 : Analyze the pollution characteristics of the pollution sources on the pollution map to perform adaptive grid processing on the spatial and temporal resolutions of different areas.

[0029] In this embodiment, the area to be analyzed is divided into multiple sub-areas based on the location of pollution sources on the pollution map, each of which contains one or more pollution sources. The pollution situation of each area is statistically integrated to dynamically set adaptive grid processing in a targeted time and space manner.

[0030] In some embodiments of the present application, the pollution characteristics of the pollution sources on the pollution map are analyzed to perform adaptive grid processing on the spatial and temporal resolution of different areas, including: The pollution map is divided into multiple regions according to the location of pollution sources. The average distance between monitoring stations, terrain complexity, number of pollution sources, and pollutant parameters in each region are counted. Pollutant parameters include pollutant toxicity parameters, emission parameters, meteorological parameters, and flow range. The pollution intensity in the region is calculated based on the pollutant toxicity parameters, emission parameters, meteorological parameters, flow range, and number of pollution sources. The initial value of the spatial resolution of the region is set based on the average distance between monitoring stations in the region, the complexity of the terrain, and the real-time value of the pollution intensity. The change value of the pollution intensity in the region is monitored according to the first preset monitoring period, and the initial value of the spatial resolution of the region is dynamically adjusted. The pollution source types, meteorological parameters, and pollutant categories and pollutant change rates generated by all pollution sources in each area are counted to set the initial value of the regional time resolution. The meteorological parameters and pollutant change rates in the area are monitored according to the second preset monitoring period, and the initial value of the regional time resolution is dynamically adjusted. Adaptive grid processing is performed for each region by dynamically adjusting the spatial and temporal resolution within the region.

[0031] In this example, the average distance between monitoring stations, terrain complexity (e.g., calculated using DEM data), number of pollution sources, and pollutant parameters (e.g., pollutant toxicity, emissions, meteorological conditions, and flow range) within each sub-region are calculated. Based on pollutant toxicity, emissions, meteorological conditions, flow range, and number of pollution sources, the pollution intensity of each sub-region is calculated using the following formula: ; in, For the The pollution intensity of a region, 、 are the conversion factors for pollutant toxicity and emission, is the number of toxicity categories of pollutants in the region, For the The pollution weight of the toxicity of each pollutant, For the Region No. The size of the parameters of the toxicity of each pollutant, is the number of pollution sources in the region, For the The pollution weight of each pollution source, For the The first The emissions from pollution sources, For the The flow effect conversion coefficient of each region is: For the The flow range of the region, For the The degree of impact after analyzing the meteorological conditions in each region, Indicates the combined influence of flow range and meteorological conditions, For the The second constant of the region, It represents the correction of the combined influence of flow range and meteorological conditions on the sum of pollutant toxicity and emissions. The second constant is to balance the size of the correction function.

[0032] Set the initial spatial resolution for each subregion based on the average distance between monitoring stations, terrain complexity, and pollution intensity. A higher spatial resolution should be set for areas with high pollution intensity, complex terrain, and sparse monitoring stations. Monitor changes in pollution intensity within each subregion according to a first preset monitoring period (e.g., hourly). If the change in pollution intensity exceeds a preset threshold, dynamically adjust the spatial resolution for that region to more precisely capture pollution changes. Because pollutant intensity varies with the operating conditions of pollution sources, monitoring is necessary. Statistics are collected for each subregion regarding pollution source type, meteorological parameters (e.g., wind speed, wind direction, temperature), and the types of pollutants generated by all pollution sources, as well as the rate of change of pollutants. Set the initial temporal resolution for each subregion based on pollution source type, meteorological parameters, and pollutant change rate. A higher temporal resolution should be set for areas with rapid pollutant change rates and variable meteorological conditions. Monitor meteorological parameters and the rate of change of pollutants within each subregion according to a second preset monitoring period (e.g., every half hour). If the rate of change exceeds a preset threshold, dynamically adjust the temporal resolution for that region to more promptly reflect pollution changes.

[0033] Adaptive meshing is performed on each sub-region based on the dynamically adjusted spatial and temporal resolutions. Regions with higher spatial resolutions have denser meshing, while regions with higher temporal resolutions have shorter time steps.

[0034] Step S103: simulate the pollutant diffusion path of the pollution source, extract the pollution diffusion factor characteristics, define the comprehensive contribution of each type of pollution diffusion factor characteristics, and fuse the pollution diffusion factor characteristics based on the comprehensive contribution to obtain the fused characteristics.

[0035] In some embodiments of the present application, the pollutant diffusion path of the pollution source is simulated to extract the pollution diffusion characteristics, including: Establish a diffusion model for the pollution sources on the pollution map, and establish different types of diffusion models for different pollution sources to simulate the diffusion path of pollutants from the pollution sources; In the process of simulating the pollutant diffusion path of the pollution source, the pollution diffusion characteristics at the spatial scale, time scale and physical scale are extracted.

[0036] In this embodiment, the diffusion of pollutants from pollution sources is an important factor affecting air quality and requires simulation analysis. For different types of pollution sources (such as point sources, line sources, and surface sources), select an appropriate diffusion model. For example, for factory point sources, you can choose the Gaussian diffusion model; for traffic line sources, you can choose the line source diffusion model. Set the parameters of the diffusion model based on the characteristics of the pollution source (such as emission rate, emission height, emission temperature) and meteorological conditions (such as wind speed, wind direction, and atmospheric stability). Run the diffusion model: Use the set parameters to run the diffusion model to simulate the diffusion path of pollutants from the pollution source. Output simulation results: Output simulation results such as the pollutant concentration distribution map, the coordinates of the maximum concentration point, and the diffusion range.

[0037] Spatial feature extraction: Extract spatial features such as diffusion range, maximum concentration point coordinates, and concentration gradient from the pollutant concentration distribution map.

[0038] Time feature extraction: Extract time features such as peak occurrence time, duration, and change rate from the time series of simulation results.

[0039] Physical feature extraction: Calculate physical features such as turbulent diffusion coefficient, dry deposition rate, and chemical reaction rate based on the diffusion model and meteorological conditions.

[0040] In some embodiments of the present application, the comprehensive contribution of each type of pollution diffusion factor characteristic is defined, including: Calculate the original contribution of each pollution diffusion factor feature to air quality, analyze the interaction between pollution diffusion factors, and divide pollution diffusion factors into independent features and interactive features based on the interaction; For independent features, the original contribution of the independent feature is used as the comprehensive contribution of the feature to air quality; For interactive features, the original contribution of the interactive feature under the interactive relationship is adjusted according to the interaction, and the adjusted original contribution is used as the comprehensive contribution of the feature to air quality.

[0041] In this embodiment, the original contribution is calculated: the original contribution of each pollution diffusion factor characteristic to the air quality is calculated using methods such as SHAP value and sensitivity analysis. Interaction analysis: the interaction between pollution diffusion factor characteristics is analyzed using methods such as SHAP interaction value and partial correlation coefficient. Classification features: According to the analysis results of the interaction, the pollution diffusion factor characteristics are divided into independent features and interactive features. For interactive features, the original contribution is adjusted according to the interaction. For example, if there is a positive interaction between feature A and feature B, and the contribution of the interaction is C, the comprehensive contribution of feature A and feature B can be adjusted to the original contribution + C / 2 respectively. Alternatively, different correction coefficients correspond to the contribution of different interactions, and the original contribution is corrected by multiplying the correction coefficients.

[0042] Based on the comprehensive contribution of pollution diffusion factors, weights are assigned to the pollution diffusion factors at the spatial, temporal, and physical scales. Features with higher contributions are given higher weights. Within each scale (spatial, temporal, and physical), pollution diffusion factors are fused using methods such as weighted averaging and principal component analysis to generate fused features at each scale.

[0043] In some embodiments of the present application, the pollution diffusion factor features are fused by means of comprehensive contribution to obtain fused features, including: According to the comprehensive contribution of pollution diffusion characteristics, the combined weights of the pollution diffusion characteristics at the spatial scale, time scale and physical scale are assigned; The pollution diffusion characteristics are fused according to the spatial scale, temporal scale and physical scale respectively to obtain the fusion characteristics at the three scales.

[0044] In this embodiment, the pollution diffusion characteristics are fused at three scales: spatial scale, temporal scale, and physical scale, as follows: ; in, For the The fusion features of different scales (spatial scale, temporal scale or physical scale) are is the number of characteristics of pollution diffusion at this scale, For the The combined weight of the comprehensive contribution mapping of each pollution diffusion factor characteristic, For the On the scale The pollution spread is due to the characteristic size, for The maximum value in For the The first constant corresponding to the scale, Used to correct the sum of the features by the maximum value on this scale. The first constant is used to balance the size of the correction function. Step S104: constructing a pollution diffusion-air quality causal model based on the fusion features, and realizing air quality prediction through the pollution diffusion-air quality causal model.

[0045] In some embodiments of the present application, a causal model of pollution diffusion-air quality is constructed based on fusion features, including: The fusion features of the three scales, namely spatial scale fusion feature, temporal scale fusion feature and physical scale fusion feature, are recorded as spatial scale fusion feature, temporal scale fusion feature and physical scale fusion feature respectively, and a causal graph between the spatial scale fusion feature, temporal scale fusion feature, physical scale fusion feature and air quality is constructed; A causal model of pollution diffusion-air quality is established based on the causal graph, and the causal model parameters are optimized.

[0046] In this example, spatial scale fusion features, temporal scale fusion features, physical scale fusion features, and air quality indicators (such as PM2.5 concentration) are used as nodes to construct a causal graph between them. This causal graph should be developed based on domain knowledge and data-driven methods, reflecting both direct and indirect causal relationships between variables. Causal modeling methods such as structural equation modeling (SEM) and Bayesian networks (BN) are used to establish a causal model of pollution diffusion and air quality based on the causal graph. The causal model is trained using historical data, and model parameters are optimized to ensure that the model accurately reflects the causal relationships between variables and achieves good predictive performance.

[0047] The process of achieving air quality prediction through the pollution diffusion-air quality causal model: Input new data: Collect new fusion feature data (including fusion features of space, time, and physical scales). These data should reflect the latest pollution source emissions, meteorological conditions, etc.

[0048] Model prediction: The new fused feature data is input into the trained causal model, and the model outputs the predicted air quality indicator (such as PM2.5 concentration).

[0049] Result output: Output air quality prediction results, including predicted air quality index values, prediction time, prediction area and other information.

[0050] Correspondingly, this application also provides an air quality prediction system that integrates multi-source spatiotemporal data, such as Figure 2 Shown, including, The first module is used to obtain the pollution sources and their locations within the area to be analyzed, construct a pollution map, collect multi-source data according to the pollution map, determine the data missing areas in the pollution map, and perform interpolation processing on the data missing areas; The second module is used to analyze the pollution characteristics of pollution sources on the pollution map to perform adaptive grid processing on the spatial and temporal resolution of different areas; The third module is used to simulate the pollutant diffusion path of the pollution source, extract the pollution diffusion factor characteristics, define the comprehensive contribution of each type of pollution diffusion factor characteristics, and fuse the pollution diffusion factor characteristics based on the comprehensive contribution to obtain the fused characteristics; The fourth module is used to build a causal model of pollution diffusion-air quality based on fusion features, and to realize air quality prediction through the causal model of pollution diffusion-air quality.

[0051] The present invention has the following beneficial effects: 1. Interpolate areas with missing data. Considering data comprehensiveness, interpolate data on the pollution map to provide a reliable foundation for subsequent pollution analysis and air quality prediction. Analyze the pollution characteristics of pollution sources on the pollution map to adaptively grid the spatial and temporal resolution of different regions. Assess the pollution characteristics of pollution sources to dynamically adjust spatial and temporal resolution in a targeted manner, ensuring map description accuracy and balancing resource and simulation needs.

[0052] 2. Define the comprehensive contribution of each type of pollution diffusion factor characteristic. Using this comprehensive contribution, fuse these characteristics and describe the fused features at three scales: temporal, spatial, and physical, ensuring comprehensive analysis. Based on these fused features, a causal model for pollution diffusion and air quality is constructed, effectively integrating features from multiple sources and improving the adaptability of data fusion, thereby enhancing the accuracy of air quality predictions and meeting complex and ever-changing air quality forecasting needs.

[0053] Through the above description of the embodiments, those skilled in the art will clearly understand that the present invention can be implemented via hardware or via software combined with a necessary general-purpose hardware platform. Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product. This software product can be stored on a non-volatile storage medium (such as a CD-ROM, USB flash drive, or external hard drive) and includes instructions for enabling a computer device (such as a personal computer, server, or network device) to execute the methods described in various implementation scenarios of the present invention.

[0054] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily required to implement the present invention.

[0055] Those skilled in the art will appreciate that the modules in the system of the implementation scenario can be distributed in the system of the implementation scenario according to the implementation scenario description, or can be modified accordingly and located in one or more systems different from the implementation scenario. The modules of the above implementation scenario can be combined into one module or further divided into multiple submodules.

[0056] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. An air quality prediction method based on multi-source spatiotemporal data fusion, characterized by: include, Obtain the pollution sources and their locations within the area to be analyzed, construct a pollution map, collect multi-source data according to the pollution map, determine the data missing areas in the pollution map, and perform interpolation processing on the data missing areas; Analyze the pollution characteristics of pollution sources on the pollution map to perform adaptive grid processing on the spatial and temporal resolution of different areas; Simulate the pollutant diffusion path of the pollution source, extract the pollution diffusion characteristics, define the comprehensive contribution of each type of pollution diffusion characteristics, and fuse the pollution diffusion characteristics based on the comprehensive contribution to obtain the fusion characteristics; Based on the fusion features, a causal model of pollution diffusion-air quality is constructed, and air quality prediction is achieved through the causal model of pollution diffusion-air quality.

2. The air quality prediction method based on multi-source spatiotemporal data fusion according to claim 1 is characterized in that: Identify areas with missing data in pollution maps, including: Mark the monitoring stations on the pollution map, calculate the spatial distribution density of the monitoring stations and the spatial correlation of the monitoring station data, and preliminarily screen out data-missing areas based on the spatial distribution density of the monitoring stations and the spatial correlation of the monitoring station data; Data missing areas are verified to determine the data missing areas in the final pollution map.

3. The air quality prediction method based on multi-source spatiotemporal data fusion according to claim 2 is characterized in that: Interpolation is performed on areas where data is missing, including: The multi-source data in the area to be analyzed are divided into main variable data and auxiliary variable data. The main variable data is pollution-related data, and the auxiliary variable data is elevation-related data. The areas other than the data missing areas in the area to be analyzed are the data known areas. The similarity of geographical features between the data known areas and the data missing areas is calculated to find the data known area closest to the data missing area. Calculate the semivariogram of the known data area based on the primary variable data and the auxiliary variable data of the data known area closest to the data missing area, and calculate the semivariogram of the entire area to be analyzed based on the primary variable data and the auxiliary variable data of the entire area to be analyzed, and record the two semivariograms as the first semivariogram and the second semivariogram respectively; The third semivariogram of the data missing area is obtained based on the first and second semivariograms; The data-missing area is divided into multiple cells through the third semivariogram, and the third semivariogram of each cell is re-evaluated. A co-kriging model is constructed for each cell. The co-kriging model introduces the directional effect and scale effect of the third semivariogram under different cells, and interpolates each cell, thereby completing the interpolation of the main variable data in the data-missing area.

4. The air quality prediction method based on multi-source spatiotemporal data fusion according to claim 1, characterized in that: Analyze the pollution characteristics of pollution sources on the pollution map to perform adaptive grid processing on spatial and temporal resolutions for different areas, including: The pollution map is divided into multiple regions according to the location of pollution sources. The average distance between monitoring stations, terrain complexity, number of pollution sources, and pollutant parameters in each region are counted. Pollutant parameters include pollutant toxicity parameters, emission parameters, meteorological parameters, and flow range. The pollution intensity in the region is calculated based on the pollutant toxicity parameters, emission parameters, meteorological parameters, flow range, and number of pollution sources. The initial value of the spatial resolution of the region is set based on the average distance between monitoring stations in the region, the complexity of the terrain, and the real-time value of the pollution intensity. The change value of the pollution intensity in the region is monitored according to the first preset monitoring period, and the initial value of the spatial resolution of the region is dynamically adjusted. The pollution source types, meteorological parameters, and pollutant categories and pollutant change rates generated by all pollution sources in each area are counted to set the initial value of the regional time resolution. The meteorological parameters and pollutant change rates in the area are monitored according to the second preset monitoring period, and the initial value of the regional time resolution is dynamically adjusted. Adaptive grid processing is performed for each region by dynamically adjusting the spatial and temporal resolution within the region.

5. The air quality prediction method based on multi-source spatiotemporal data fusion according to claim 1 is characterized in that: Simulate the pollutant diffusion path of the pollution source and extract the pollution diffusion characteristics, including: Establish a diffusion model for the pollution sources on the pollution map, and establish different types of diffusion models for different pollution sources to simulate the diffusion path of pollutants from the pollution sources; In the process of simulating the pollutant diffusion path of the pollution source, the pollution diffusion characteristics at the spatial scale, time scale and physical scale are extracted.

6. The air quality prediction method based on multi-source spatiotemporal data fusion according to claim 5 is characterized in that: Define the comprehensive contribution of each type of pollution diffusion characteristics, include, Calculate the original contribution of each pollution diffusion factor feature to air quality, analyze the interaction between pollution diffusion factors, and divide pollution diffusion factors into independent features and interactive features based on the interaction; For independent features, the original contribution of the independent feature is used as the comprehensive contribution of the feature to air quality; For interactive features, the original contribution of the interactive feature under the interactive relationship is adjusted according to the interaction, and the adjusted original contribution is used as the comprehensive contribution of the feature to air quality.

7. The air quality prediction method based on multi-source spatiotemporal data fusion according to claim 5, characterized in that: The pollution diffusion characteristics are fused by comprehensive contribution to obtain fusion features, including: According to the comprehensive contribution of pollution diffusion characteristics, the combined weights of the pollution diffusion characteristics at the spatial scale, time scale and physical scale are assigned; The pollution diffusion characteristics are fused according to the spatial scale, temporal scale and physical scale respectively to obtain the fusion characteristics at the three scales.

8. The air quality prediction method based on multi-source spatiotemporal data fusion according to claim 7 is characterized in that: Construct a causal model of pollution diffusion-air quality based on fusion features, including: The fusion features of the three scales, namely spatial scale fusion feature, temporal scale fusion feature and physical scale fusion feature, are recorded as spatial scale fusion feature, temporal scale fusion feature and physical scale fusion feature respectively, and a causal graph between the spatial scale fusion feature, temporal scale fusion feature, physical scale fusion feature and air quality is constructed; A causal model of pollution diffusion-air quality is established based on the causal graph, and the causal model parameters are optimized.

9. The air quality prediction system based on multi-source spatiotemporal data fusion is characterized by: include, The first module is used to obtain the pollution sources and their locations within the area to be analyzed, construct a pollution map, collect multi-source data according to the pollution map, determine the data missing areas in the pollution map, and perform interpolation processing on the data missing areas; The second module is used to analyze the pollution characteristics of pollution sources on the pollution map to perform adaptive grid processing on the spatial and temporal resolution of different areas; The third module is used to simulate the pollutant diffusion path of the pollution source, extract the pollution diffusion factor characteristics, define the comprehensive contribution of each type of pollution diffusion factor characteristics, and fuse the pollution diffusion factor characteristics based on the comprehensive contribution to obtain the fused characteristics; The fourth module is used to build a causal model of pollution diffusion-air quality based on fusion features, and to realize air quality prediction through the causal model of pollution diffusion-air quality.

Images