Dust pollution health risk early warning terminal and early warning method

Abstract

The invention discloses a sand and dust pollution health risk early warning terminal and early warning method, and the method comprises the steps: obtaining multi-source input data, such as meteorological parameters, sand and dust pollutant concentration and geographical environment, in real time, and carrying out the preprocessing through a data fusion algorithm, thereby obtaining a standardized data set; based on the data set, utilizing a trained time sequence prediction model to output a sand and dust pollution level prediction value of a future preset time length; meanwhile, an exposed crowd sensitivity index and regional health baseline data containing the disease incidence rate and the meteorological sensitive crowd proportion are constructed; calculating a comprehensive health risk index through a dynamic health risk assessment model in combination with the predicted value, the sensitivity index and the baseline data; and when the index exceeds a preset threshold dynamically adjusted based on the incidence peak range in the baseline data, generating a graded early warning signal and issuing the graded early warning signal in real time, thereby realizing accurate early warning of the dust pollution health risk.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of environmental monitoring and public health, specifically relating to early warning technology for health risks to the population caused by dust pollution, and particularly to a dust pollution health risk early warning terminal and early warning method. Background Technology

[0002] Dust pollution, a common meteorological disaster worldwide, not only leads to deteriorating air quality and reduced visibility but also poses a significant threat to public health. Studies show that particulate matter, heavy metals, and microorganisms carried in dust can trigger respiratory diseases (such as asthma and pneumonia), cardiovascular diseases, and eye inflammation through inhalation and contact, with particularly pronounced effects on vulnerable populations such as the elderly, children, and those with chronic illnesses. With climate change and accelerated urbanization, the frequency and extent of dust storms are fluctuating, making accurate early warning of the health risks posed by dust pollution a crucial issue for ensuring public health security.

[0003] In existing technologies, early warning systems for dust pollution primarily focus on environmental monitoring. These systems mainly acquire dust concentration data through ground monitoring stations and satellite remote sensing, then combine this data with meteorological forecasting models to predict dust movement paths and diffusion trends, ultimately issuing pollution level warnings. Regarding health risk associations, some studies have established statistical relationships between dust concentrations and the incidence of specific diseases by analyzing historical epidemiological data, forming preliminary health alert mechanisms. For example, some regions have attempted to issue health protection recommendations based on PM10 concentration thresholds or use simple regression models to estimate short-term health risks.

[0004] However, existing technologies still have significant limitations: First, the data dimensions are limited, relying heavily on meteorological or pollutant concentration data, without fully integrating key information such as the characteristics of exposed populations, geographical environment, and regional health baselines, resulting in a lack of targeted risk assessment; second, the prediction and assessment models are simplistic, with time series predictions often ignoring the nonlinear variation characteristics of dust pollution, health risk assessments failing to consider differences in population sensitivity, and warning thresholds often being fixed values, making it difficult to adapt to fluctuations in health baselines in different regions; third, the correlation between warning signals and health impacts is insufficient, only reflecting the degree of pollution and failing to accurately quantify the actual health threat to specific populations, thus limiting the guidance and practicality of warnings. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a dust pollution health risk early warning terminal and early warning method to solve the above-mentioned technical problems.

[0006] To achieve the above objectives, firstly, a method for early warning of health risks from sandstorm pollution is provided, which includes the following steps: Real-time acquisition of multi-source input data, including meteorological parameter data, dust pollutant concentration data, and geographic environment data; The multi-source input data is preprocessed using a data fusion algorithm to obtain a standardized dataset; Based on the standardized dataset, a trained time series prediction model is used to output a predicted value of the dust pollution level for a preset future duration. Real-time acquisition of exposed population characteristic data in the target area, and construction of an exposed population sensitivity index based on the exposed population characteristic data and dynamic feedback data in the target area; Based on historical epidemiological statistics, the incidence rate of dust-related diseases and the proportion of meteorologically sensitive populations were extracted, and regional health baseline data were constructed using the aforementioned incidence rate and proportion of meteorologically sensitive populations. Based on the predicted dust pollution level, the sensitivity index of the exposed population, and the regional health baseline data, a comprehensive health risk index is calculated using a dynamic health risk assessment model. When the comprehensive health risk index exceeds a preset dynamic threshold, a graded early warning signal is generated. The preset dynamic threshold is dynamically adjusted according to the peak range of the incidence rate in the regional health baseline data and is published to the user terminal or early warning platform in real time through the communication interface.

[0007] Secondly, a dust pollution health risk early warning terminal is provided, which includes: The data acquisition module is used to acquire multi-source input data in real time, including meteorological parameter data, dust pollutant concentration data, and geographical environment data. The data preprocessing module is used to preprocess the multi-source input data using a data fusion algorithm to obtain a standardized dataset; The prediction module is used to output a predicted value of the dust pollution level for a preset time period in the future, based on the standardized dataset and a trained time series prediction model. The sensitivity index construction module is used to acquire the exposure population characteristic data of the target area in real time, and construct the exposure population sensitivity index based on the exposure population characteristic data and dynamic feedback data of the target area; The baseline data construction module is used to extract the incidence rate of dust-related diseases and the proportion of meteorologically sensitive populations based on historical epidemiological statistics, and to construct regional health baseline data based on the incidence rate and the proportion of meteorologically sensitive populations. The risk index calculation module is used to calculate a comprehensive health risk index based on the predicted dust pollution level, the sensitivity index of the exposed population, and the regional health baseline data, using a dynamic health risk assessment model. The early warning module is used to generate a graded early warning signal when the comprehensive health risk index exceeds a preset dynamic threshold. The preset dynamic threshold is dynamically adjusted according to the historical peak fluctuation range in the regional health baseline data, and the graded early warning signal is published to the user terminal or early warning platform in real time through the communication interface.

[0008] The above technical solution has the following beneficial technical effects: The integration and standardization of multi-source input data in this technical solution avoids the limitations of a single data dimension. By combining regional health baseline data with differences in population sensitivity, the health risk assessment is more closely aligned with the actual characteristics of the target area, improving the accuracy of the assessment. Based on a time series prediction model, the predicted value of dust pollution level is obtained in advance, and a tiered early warning signal is generated through a dynamic threshold adjustment mechanism. This allows for targeted warnings before health risks manifest, providing sufficient time for the public to take protective measures and for relevant departments to deploy emergency responses, effectively enhancing the foresight and timeliness of the early warning. The deep correlation between the early warning signal and the regional health baseline and population characteristics ensures the practicality of the early warning information. It avoids over-warning or under-warning caused by fixed thresholds, and accurately targets high-risk groups, thereby minimizing the potential harm of dust pollution to public health. Attached Figure Description

[0009] Figure 1 This is a flowchart of a method for early warning of health risks from sand and dust pollution according to an embodiment of the present invention; Figure 2 This is a detailed flowchart of step S10 in an embodiment of the present invention; Figure 3 This is a detailed flowchart of step S20 in an embodiment of the present invention; Figure 4 This is a detailed flowchart of step S30 in an embodiment of the present invention; Figure 5 This is a flowchart illustrating the training process of the time series prediction model according to an embodiment of the present invention. Figure 6 This is a detailed flowchart of step S40 in an embodiment of the present invention; Figure 7 This is a detailed flowchart of step S50 in an embodiment of the present invention; Figure 8 This is a detailed flowchart of step S60 in an embodiment of the present invention; Figure 9 This is a detailed flowchart of step S70 in an embodiment of the present invention; Figure 10 This is a functional block diagram of a dust pollution health risk early warning terminal according to an embodiment of the present invention. Detailed Implementation

[0010] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details of these embodiments to aid understanding, and should be considered merely exemplary.

[0011] Example 1 like Figure 1 As shown in the figure, this embodiment provides a method for early warning of health risks from sandstorm pollution, which includes the following steps: S10: Real-time acquisition of multi-source input data, including meteorological parameter data, dust pollutant concentration data, and geographical environment data.

[0012] In step S10, meteorological parameters such as wind speed, wind direction, temperature, and humidity in the target area are collected through meteorological satellites and ground monitoring stations, with a sampling frequency of once per hour; PM10, PM2.5, and other dust pollutant concentration data are collected through environmental monitoring points distributed in various urban areas, and the data collection time and latitude and longitude of the monitoring points are obtained simultaneously; geographic environmental data such as topography, vegetation coverage, and building density of the target area are obtained through a Geographic Information System (GIS), and the above data are transmitted to the data processing center in real time through an API interface.

[0013] S20: Use a data fusion algorithm to preprocess the multi-source input data to obtain a standardized dataset.

[0014] In step S20, the multi-source data is first cleaned to remove outliers in meteorological parameters caused by sensor malfunctions. For dust pollutant concentration data, interpolation is used to fill in missing values ​​in sparse areas of the monitoring points. Then, the Z-score normalization method is used to normalize parameters of different magnitudes (e.g., wind speed in m / s, pollutant concentration in m / s, etc.). The data is transformed to the [-1,1] interval; finally, an improved Bayesian fusion algorithm is used to perform weighted fusion of heterogeneous data from the same source (such as pollutant concentration data from satellite remote sensing and ground monitoring). The weights are dynamically adjusted according to the data error variance, and finally a standardized dataset with a time granularity of 1 hour and a spatial granularity of 1km×1km is formed.

[0015] S30: Based on the standardized dataset, the trained time series prediction model is used to output the predicted value of the dust pollution level for a preset time period in the future.

[0016] In step S30, the time series prediction model employs an attention-based LSTM neural network. The model training dataset consists of standardized historical data from the past three years. Input features include meteorological parameters, pollutant concentrations, and geographical environmental characteristics from the previous 72 hours. The output is the predicted dust pollution levels (with PM10 concentration as the core indicator) for the next 24, 48, and 72 hours. During training, a sliding window method is used to generate samples, and the Adam (Adaptive Moment Estimation) optimizer minimizes the mean square error between the predicted and actual values. After model deployment, the prediction results are updated every 6 hours to ensure timeliness.

[0017] S40: Acquire the characteristic data of the exposed population in the target area in real time, and construct the sensitivity index of the exposed population based on the characteristic data of the exposed population in the target area and the dynamic feedback data.

[0018] This step extracts characteristic data of the exposed population from the regional demographic database and community health records, including age distribution, prevalence of chronic diseases, and occupational type. In step S40, the population health characteristic data includes basic data such as the proportion of children aged 0-14 years, the proportion of elderly people aged 65 years and above, and the number of asthma / heart disease patients in the target area; the dynamic feedback data is the year-on-year change rate of respiratory disease visits to regional medical institutions in the past 7 days; after quantifying the above data, the Analytic Hierarchy Process (AHP) is used to assign weights, with the proportion of chronic disease patients having the highest weight (30%), followed by the change rate of visits (25%), and finally a sensitivity index of 0-10 is calculated, with 8-10 points being high sensitivity, 4-7 points being medium sensitivity, and 0-3 points being low sensitivity.

[0019] S50: Extract the incidence rate of dust-related diseases and the proportion of meteorologically sensitive populations based on historical epidemiological statistics, and construct regional health baseline data using the aforementioned incidence rate and proportion of meteorologically sensitive populations.

[0020] In step S50, epidemiological statistics on dust-related diseases (such as acute upper respiratory tract infection and pneumonia) over the past 5 years are extracted from the regional CDC, and the average incidence rate and the peak incidence rate range within the 95% confidence interval for the same period each year are calculated. The proportion of meteorologically sensitive populations (such as children, the elderly, and patients with chronic diseases) in the target area is statistically analyzed using census data and community health records, and a heat map of the spatial distribution of sensitive populations is generated in conjunction with the GIS system. The above data are integrated into a regional health baseline database that includes a time dimension (monthly baseline) and a spatial dimension (street-level distribution).

[0021] S60: Based on the predicted dust pollution level, the sensitivity index of the exposed population, and the regional health baseline data, calculate the comprehensive health risk index using a dynamic health risk assessment model.

[0022] In step S60, the predicted dust pollution level (PM10 concentration), the sensitivity index of the exposed population, and the regional health baseline data are first matched at a spatial granularity of 1 km × 1 km and a temporal granularity of 1 hour. Based on the mean incidence rate in the baseline data, the PM10 concentration is calculated for each increase using multiple linear regression. The corresponding increase in disease incidence risk (basic correlation coefficient); the correction coefficients corresponding to the sensitivity indicators (high sensitivity 1.2, medium sensitivity 1.0, low sensitivity 0.8) are multiplied by the basic correlation coefficients to obtain the dynamic correlation coefficients of each spatial unit; a weighted summation algorithm with population density as the weight is used to spatially aggregate the dynamic correlation coefficients of each unit, and finally the aggregation results are mapped to a comprehensive health risk index of 0-100 through an S-shaped function.

[0023] S70: When the comprehensive health risk index exceeds the preset dynamic threshold, a graded early warning signal is generated. The preset dynamic threshold is dynamically adjusted according to the peak range of the incidence rate in the regional health baseline data, and is published to the user terminal or early warning platform in real time through the communication interface.

[0024] In step S70, the preset dynamic threshold is determined based on the peak range of the incidence rate in the regional health baseline data. 70% and 85% of the peak value are taken as the yellow and orange warning thresholds, respectively, and the peak value itself is taken as the red warning threshold. When the comprehensive health risk index exceeds the corresponding threshold, a warning message is automatically generated, which includes the warning level, the scope of impact (accurate to the street level), the type of high-risk group, and protection recommendations (such as reducing going out and wearing N95 masks). The message is pushed to the user's mobile APP through the mobile communication network, and is also synchronized to the community electronic screen, hospital information system, and local emergency management platform. The release frequency is increased with the warning level (once every 2 hours for yellow, once every 1 hour for orange, and once every 30 minutes for red).

[0025] This technical solution integrates multi-source data, including meteorological, pollutant, population characteristics, and geographical environment data, and after standardized processing, achieves accurate prediction of dust pollution levels. It combines the sensitivity differences of exposed populations with regional health baseline data to construct a dynamic assessment model, ultimately generating tiered early warnings based on dynamic thresholds. This approach avoids the limitations of traditional early warning systems, which rely on single data sources, simplified assessments, and fixed thresholds. It improves the accuracy and relevance of health risk assessments and enhances the foresight and timeliness of early warnings through advance prediction and dynamic adjustment. Furthermore, specific tiered early warning signals and protective recommendations can precisely target high-risk populations and areas, effectively guiding public protection and departmental emergency response, thereby minimizing the potential harm of dust pollution to public health and providing strong protection for public health security.

[0026] like Figure 2 As shown, in some embodiments, step S10 may specifically include: S11: Real-time acquisition of meteorological parameter data, including wind speed, wind direction, temperature, relative humidity and air pressure, wherein the meteorological parameter data comes from the meteorological monitoring system of the target area.

[0027] Specifically, this step involves deploying meteorological monitoring stations in the target area at a grid density of 10 km × 10 km. Each station is equipped with an ultrasonic anemometer, wind vane, temperature and humidity sensor, and barometric pressure sensor to collect data on wind speed, wind direction, temperature, relative humidity, and barometric pressure, respectively. The monitoring system uses BeiDou satellite synchronization and collects raw data every 5 minutes. After preprocessing at the station (removing outliers exceeding reasonable ranges), the data is uploaded in real time to the central server via a 4G / 5G wireless communication module. The server then performs mean fusion on the data from multiple stations within the same grid to generate hourly meteorological parameter datasets.

[0028] S12: Collect data on the concentration of dust pollutants, including the mass concentrations of PM10, PM2.5, and total suspended particulate matter, through multi-point synchronous collection via ground environmental monitoring stations and mobile monitoring vehicles.

[0029] Specifically, when collecting data on dust pollutant concentrations, ground-based environmental monitoring stations are deployed at a standard of one station per 5 square kilometers in urban built-up areas and one station per 20 square kilometers in suburban areas. Each station is equipped with a beta-ray absorption spectrometer (used for PM10 and PM2.5 concentration detection, with a detection range of...). precision ) and oscillating microbalance method for total suspended particulates (TSP) analyzer (detection range) precision Simultaneously, 3-5 mobile monitoring vehicles are deployed, moving along a pre-set route (covering major traffic arteries, residential areas, and industrial zones) to collect samples once per hour. Sampling points are located and marked in real-time using a multi-positioning system (e.g., GPS, BDS, GLONASS). Both ground stations and mobile vehicles employ unified quality control standards (daily zero-point calibration, weekly standard membrane verification). All data is encrypted and transmitted to the data center, where spatiotemporal interpolation algorithms fill in monitoring gaps, forming a grid of pollutant concentration data (spatial resolution 1km × 1km) updated hourly.

[0030] S13: Obtain geographic environment data through remote sensing image interpretation and geographic information system vector data, including terrain type, vegetation coverage, surface roughness and land use type.

[0031] Specifically, when collecting geographic environmental data, the remote sensing images mainly used were Landsat-8 satellite imagery (spatial resolution 30 meters) and Sentinel-2 satellite imagery (spatial resolution 10 meters). The latest imagery of the target area was acquired quarterly and interpreted using ENVI (Environment for Visualizing Images) remote sensing image processing software. Supervised classification was used to identify terrain types (e.g., plains, mountains, basins). Vegetation coverage was calculated using the Normalized Difference Vegetation Index (NDVI). Surface roughness (quantified by the standard deviation of surface height) was extracted using Digital Elevation Model (DEM) data. Land use type data was obtained from the GIS vector database of the natural resources department, including classifications such as cultivated land, forest land, construction land, and water areas, with a spatial resolution of 1:10,000. All geographic environmental data are rasterized into 1km×1km spatial units after projection transformation (unified to WGS84 coordinate system) to achieve spatial matching with meteorological and pollutant data. The data is updated once a year, and is temporarily updated if major geomorphological changes occur in the region (such as large-scale construction or vegetation destruction).

[0032] By ensuring that meteorological parameter data (including wind speed, wind direction, temperature, etc.) originates from the meteorological monitoring system of the target area and is collected in real time, and that dust pollutant concentration data (including PM10, PM2.5, and total suspended particulate matter) is collected synchronously at multiple points through ground environmental monitoring stations and mobile monitoring vehicles, and that exposed population characteristic data (including age distribution, chronic disease prevalence, etc.) is obtained from regional medical and health databases and demographic data, and that geographical environmental data (including terrain type, vegetation coverage, etc.) is acquired through remote sensing image interpretation and geographic information system vector data, a multi-dimensional and multi-source data collection system has been formed. This system covers not only the meteorological conditions for dust pollution diffusion and the concentration characteristics of the pollutants themselves, but also incorporates information on the health vulnerability of exposed populations and geographical environmental factors affecting dust diffusion. Through unified collection standards and source assurance, the system ensures the matching of various data in the spatiotemporal dimensions, the accuracy of the data itself, and the comprehensiveness of information coverage, providing high-quality data support for subsequent data fusion preprocessing, pollution level prediction, and health risk assessment.

[0033] like Figure 3 As shown, in some embodiments, step S20 specifically includes the following steps: S21: Perform outlier detection and correction on multi-source input data. For abnormal fluctuations in the meteorological parameter data and the dust pollutant concentration data, use interpolation to correct them. For missing values ​​in the geographic environment data, use mean filling to fill them.

[0034] In this step, the 3σ principle (i.e., values ​​exceeding the mean ± 3 standard deviations are considered outliers) is used to identify outliers in meteorological parameter data (e.g., wind speed, temperature) and dust pollutant concentration data (e.g., PM10). This is combined with professional threshold ranges (e.g., PM10 concentrations are unlikely to actually exceed a certain threshold). Secondary verification is performed. For detected abnormal fluctuations, a correction method is selected based on their time series characteristics. If the outlier is an isolated value, linear interpolation of data from the preceding and following hours is used for correction. If there are multiple consecutive outliers, a weighted average interpolation of data from neighboring monitoring stations during the same period is used (weights are inversely proportional to distance). For missing values ​​in geographical environment data (e.g., vegetation coverage), the mean of data of the same region and type is first calculated (e.g., the average prevalence rate of the same street, the average vegetation coverage rate of the same terrain), and then the mean is used to fill in the missing positions to ensure data integrity.

[0035] S22: Perform spatiotemporal alignment of data from different sources, unifying the temporal granularity and spatial coordinate system of the data. This ensures that the temporal resolution of the meteorological parameter data and the dust pollutant concentration data remains consistent, and that the spatial division range of the geographic environment data matches that of the meteorological parameter data and the dust pollutant concentration data. In this step, regarding the unification of temporal granularity, both the meteorological parameter data (original sampling frequency 5 minutes) and the dust pollutant concentration data (original sampling frequency 1 hour) are aggregated into one data point per hour. The meteorological data uses the hourly average, while the pollutant data directly uses the original hourly data, ensuring consistent temporal resolution. The spatial coordinate system is uniformly adopted as the WGS84 geographic coordinate system. Coordinate transformation tools are used to convert data from different sources (e.g., local coordinates of meteorological stations, UTM coordinates of remote sensing images) to this system.

[0036] In this embodiment, the specific implementation process for matching the spatial division range of the geographic environment data, the meteorological parameter data, and the dust pollutant concentration data is as follows: First, a unified standard for spatial division of the target area is determined, using a 1km×1km grid as the smallest spatial unit. The boundary of this grid unit is precisely located using the WGS84 coordinate system (the latitude and longitude of the upper left corner and the lower right corner of each grid can be automatically generated by the GIS system), ensuring that all data remain consistent in the spatial coordinate system. For the meteorological parameter data, the original data comes from ground meteorological monitoring stations deployed within the target area (each station covers an area of ​​approximately 5-10km). The point data (e.g., wind speed, temperature) collected by each monitoring station is interpolated into the 1km×1km grid using the Kriging interpolation method, so that each grid unit corresponds to a unique average meteorological parameter (the interpolation process incorporates the surrounding terrain features of the monitoring station to correct errors and ensure that the interpolation accuracy meets the requirements). For dust pollutant concentration data, the original data consisted of PM10 and PM2.5 concentration values ​​collected from environmental monitoring points distributed throughout the city (each monitoring point covering approximately 2-5 km). Inverse distance weighted interpolation was used to convert the point data into area data with a 1 km × 1 km grid. The interpolation used the straight-line distance from the monitoring point to the grid center as the weight, with closer points receiving higher weights, ensuring that the grid boundaries of the pollutant concentration data and meteorological parameter data completely overlapped. For geographic environment data, the original remote sensing images (resolution 30 m × 30 m) were resampled to 1 km × 1 km using bilinear resampling technology, ensuring that the geographic area corresponding to each remote sensing image was precisely divided into several... Create a 1km×1km grid. Simultaneously, use a vector-to-raster tool to cut and assign values ​​to the terrain type (e.g., plain, mountain), vegetation cover, and surface roughness information from the GIS vector data, ensuring that the geographic environmental characteristics (e.g., 35% vegetation cover, plain terrain) within each grid cell correspond one-to-one with the meteorological parameters and pollutant concentration data for that grid. Ultimately, achieve complete spatial matching of the three types of data; that is, any 1km×1km grid cell simultaneously contains the unit's meteorological parameters (e.g., wind speed 2.3m / s, relative humidity 45%) and dust pollutant concentration (e.g., PM10 concentration). The data, including the geographical features (e.g., 35% vegetation coverage and 0.03m surface roughness), does not have any data mismatch issues caused by spatial misalignment, thus providing a unified spatial data foundation for the subsequent standardization process in step S23 and the input of the pollution prediction model in step S30.

[0037] S23: Standardize the spatiotemporally aligned data by mapping various types of data to a preset numerical range using the range standardization method, thereby eliminating dimensional differences between different parameters.

[0038] In this step, the data standardization process employs the range standardization method, specifically using the formula: Standardized value = (Original value - Minimum value of the parameter) / (Maximum value of the parameter - Minimum value of the parameter), where the maximum and minimum values ​​are the historical extreme values ​​of the target area over the past 3 years. This method is used to standardize meteorological parameters (e.g., wind speed 0-20 m / s) and dust pollutant concentrations (e.g., ... The data on exposed population characteristics (e.g., prevalence rate 0-50%) and geographical environment data (e.g., vegetation coverage 0-100%) are all mapped to the preset numerical range of [0,1]. For some parameters with negative values ​​(e.g., NDVI may be -1 to 1), the adjustment formula is: Standardized value = (original value - minimum value) / (maximum value - minimum value) × 2 - 1, mapped to the range of [-1,1], to ensure that parameters of different dimensions and magnitudes can be fused and calculated in subsequent calculations.

[0039] S24: Use data fusion algorithms to perform feature fusion on the standardized multi-source data to generate a standardized dataset that includes meteorological parameter features, dust pollutant concentration features, and geographical environment features.

[0040] In this step, the data fusion algorithm employs an improved combination of feature stitching and weighted fusion. First, it extracts derived features such as wind speed and direction, temperature and humidity coupling, pressure gradient, and vertical wind shear from the standardized meteorological parameter data. For dust pollutant concentration data, it calculates the PM10 to PM2.5 ratio and pollutant concentration change rate. For geographical environment data, it generates features such as terrain roughness, vegetation cover, altitude, and soil moisture. Then, it uses Principal Component Analysis (PCA) to reduce the dimensionality of similar features, retaining principal components with a cumulative contribution rate of 95%. Finally, it stitches the principal component features of different categories together in a time-space unit and assigns dynamic weights based on the correlation between each feature and dust pollution diffusion (e.g., wind speed features have a higher weight than temperature features). This results in a standardized dataset containing timestamps, spatial grid coordinates, and 28 features across four categories, providing structured input for subsequent prediction models.

[0041] The process of stitching together data by time-space units involves: first, dividing continuous time into fixed intervals (e.g., 1 hour per time unit), and dividing geographic space into regular grids (e.g., 1km × 1km per spatial unit). Each time-space unit corresponds to a unique timestamp and spatial coordinates. Then, the principal components of meteorological parameters, dust pollutant concentrations, exposed populations, and geographic environment, after PCA dimensionality reduction, are each mapped one-to-one to these time-space units (ensuring that each unit contains complete data of all four types of features). Finally, the four types of principal component features are combined into a complete record by unit. The resulting dataset is a structured dataset, where each data record contains a unique timestamp, unique spatial grid coordinates, and 28 principal component feature values ​​across the four categories. Each record fully represents the comprehensive characteristics of a specific time and region and can be directly used as input samples for subsequent prediction models, allowing the models to learn the correlation between dust pollution diffusion and multi-dimensional features.

[0042] The aforementioned technical solution first effectively eliminates noise interference in meteorological parameters, pollutant concentrations, and other data by outlier detection and interpolation correction, and fills information gaps in geographical environment data, ensuring data integrity and accuracy. Next, by unifying the temporal granularity and spatial coordinate system, it achieves precise alignment of data from different sources in the spatiotemporal dimensions, ensuring consistent temporal resolution of meteorological and pollutant data and matching spatial range of geographical environment and population characteristic data, thus solving the heterogeneity problem of multi-source data. Then, by eliminating dimensional differences through range standardization, it makes parameters of different magnitudes comparable and fusionable. Finally, by feature fusion, it generates a standardized dataset containing multiple types of features, achieving deep integration of data information.

[0043] like Figure 4 As shown, in some embodiments, step S30 may specifically include: S31: Extract the associated features for dust pollution prediction from the standardized dataset. The associated features include meteorological parameter features, dust pollutant concentration features, and geographical environment features.

[0044] In this embodiment, for meteorological parameters, parameters closely related to dust diffusion, such as absolute wind speed, cosine of wind direction angle, and the product of relative humidity and air pressure, are selected. Dynamic features such as the wind speed change rate and wind direction stability index over the past 3 hours are also calculated. For dust pollutant concentration characteristics, temporal features such as the current PM10 / PM2.5 ratio, the moving average of pollutant concentration over the past 6 hours, and the time of peak concentration occurrence are extracted. For geographical environmental characteristics, spatial features such as the coupling coefficient between surface roughness and wind speed, the correlation between vegetation cover and pollutant deposition rate, and the airflow obstruction coefficient caused by topographic elevation differences are selected. Features with an absolute correlation value greater than 0.6 with dust pollution levels are selected using the Pearson correlation coefficient method, and finally, 23 core correlated features are retained to form the input feature set of the prediction model.

[0045] S32: Input the extracted correlation features into the trained time series prediction model. The time series prediction model calculates the dust pollution level for different preset durations in the future based on the correlation between meteorological conditions and dust pollution in historical data.

[0046] In this embodiment, the time series prediction model employs a bidirectional LSTM (Long Short-Term Memory) neural network based on an attention mechanism. The model input is the associated feature sequence of the past 72 hours (time step of 1 hour), and the output is the dust pollution level for the next 24, 48, and 72 hours, three preset time periods. Model training uses historical data from the target area over the past 5 years, divided into training, validation, and test sets in a 7:2:1 ratio. The Adam optimizer minimizes the mean absolute error (MAE) between the predicted and actual concentrations, and an early stopping mechanism is introduced (training stops if the error on the validation set does not improve after 10 consecutive rounds). The model learns the correlation patterns in historical data, such as strong winds and low humidity leading to pollutant diffusion, and basin topography leading to pollutant accumulation. It dynamically adjusts the prediction weights based on real-time input associated features to achieve a stepped calculation of pollution levels for different time periods.

[0047] S33: Output the predicted dust pollution level for each preset time period. The predicted value includes the dust pollutant concentration and distribution status of different sub-regions within the target area.

[0048] In this embodiment, the output dust pollution level prediction is based on a 1km × 1km grid as the spatial unit. Each grid contains the predicted mass concentrations (unit: μg / m³) of PM10, PM2.5, and total suspended particulate matter for three time nodes: the next 24 hours, 48 ​​hours, and 72 hours. Simultaneously, concentration grading is performed... For excellence For good A spatial distribution map of dust pollution in the target area is generated for areas with light pollution levels. Different colors are used to indicate the pollution level of each sub-region, and the center coordinates and expected duration of high-pollution areas are marked. The prediction results are stored in JSON format, including timestamp, spatial grid ID, pollutant concentration value, and distribution level fields, which facilitates subsequent steps for spatial matching.

[0049] This technical solution provides targeted input for the prediction model by accurately extracting meteorological, pollutant concentration, and geographical environmental characteristics related to dust pollution. Combined with a pre-trained time-series prediction model that learns the correlation between meteorological conditions and dust pollution in historical data, it can scientifically calculate pollution levels at different time points in the future, improving the logic and accuracy of the predictions. The final output includes predicted values ​​of pollutant concentrations and distribution in different sub-regions of the target area, achieving multi-period prediction in the time dimension and covering refined spatial distribution. This provides accurate and comprehensive pollution level data support for subsequent health risk assessments based on population characteristics and health baselines, effectively enhancing the overall early warning method's ability to predict dynamic changes in dust pollution.

[0050] like Figure 5 As shown, in some embodiments, the training process of the time series prediction model includes: S321: Select standardized historical data from historical periods, including meteorological parameter sequences, dust pollutant concentration sequences, and static geographical environmental characteristics collected at fixed time intervals, and divide the standardized historical data into training sets and validation sets according to a preset ratio.

[0051] In practice, the selected standardized historical data covers continuous monitoring data of the target area over the past five years. This includes meteorological parameter sequences such as hourly wind speed, wind direction, and temperature (at 1-hour intervals), dust pollutant concentration sequences including synchronized hourly concentrations of PM10 and PM2.5, and static geographical environmental features such as terrain type and vegetation cover (updated quarterly). This data is divided into training and validation sets in a predetermined 7:3 ratio, maintaining the continuity of the time series during the division (e.g., the first 70% of the data is for the training set, and the last 30% for the validation set) to avoid fragmentation of time-series information caused by random partitioning. Simultaneously, data augmentation is performed on the training set by adding ±5% Gaussian noise to simulate sensor errors and improve the model's adaptability to data fluctuations.

[0052] S322: A time series prediction model is constructed using an architecture that combines a bidirectional long short-term memory network with an attention mechanism. The input layer receives the associated feature sequence, extracts the time series features through a multi-layer stacked bidirectional long short-term memory network, assigns dynamic weights to features at different time steps through an attention layer, and then maps them to the output layer through a fully connected layer containing several neurons to output the predicted value of the dust pollution level for a preset time period in the future.

[0053] In specific implementation, the architecture of the time series prediction model (BiLSTM-Attention) in S322 can be as follows: The input layer receives a 72-hour sequence of associated features (containing 23 features, with an input dimension of 72×23); it is then connected to a three-layer stacked bidirectional long short-term memory network (Bi-LSTM), with the first layer containing 128 neurons, the second layer containing 64 neurons, and the third layer containing 32 neurons. Each layer uses dropout regularization (dropout rate of 0.2) to prevent overfitting, and captures the bidirectional dependencies of the time series data through LSTM units in both the forward and backward directions; the attention layer assigns dynamic weights (weights sum to 1) to the 72 time-step features output by the Bi-LSTM, giving higher weights to periods with significant impact from dust pollution (e.g., the initial stage of pollution spread); the fully connected layers (FC) first integrate these high-dimensional time series features into a more compact feature vector. For example, the fully connected layer (32 neurons) converts the output of the attention layer into 32-dimensional features, achieving feature dimensionality reduction and key information extraction. During the key information extraction process, the fully connected layer introduces a nonlinear transformation through the ReLU activation function to capture the complex nonlinear relationships between features. For example, in dust pollution prediction, the relationship between meteorological factors, pollution sources, and PM10 concentration is often nonlinear; ReLU activation can enhance the model's ability to fit such relationships. The final fully connected layer (16 neurons) further refines the features, outputting predicted dust pollution levels (with PM10 concentration as the core indicator) for the next 24, 48, and 72 hours through the output layer (3 neurons).

[0054] S323: During model training, the root mean square error is used as the loss function, and an adaptive momentum optimizer is used for iterative training. After each round of training, the generalization ability is evaluated through the validation set. When the loss on the validation set does not decrease for a consecutive preset number of rounds, training is stopped and the optimal model parameters are saved.

[0055] In specific implementation, the model training process is set as follows: Root Mean Square Error (RMSE) is used as the loss function, calculated as the square root of the mean of the squared differences between the predicted and actual values; an adaptive momentum optimizer (Adam) is used, with an initial learning rate of 0.001, which automatically decays to 1 / 10 of its original rate if the validation set loss does not decrease for three consecutive rounds; the maximum number of iterations is set to 50 rounds, and the loss value is calculated using the validation set after each round of training. If the validation set loss does not decrease for five consecutive rounds (fluctuation range less than...), the model is considered lost. If the model fails to reach the minimum loss on the validation set, the early stopping mechanism will be triggered to terminate the training. Finally, the model parameters (including the weights of each layer, bias terms, and attention mechanism parameters) will be saved when the loss on the validation set is minimized. The model will be stored as a binary file and the training log (including the loss curve and the ranking of feature importance) will be recorded to ensure that the model is reproducible and easy to tune later.

[0056] This technical solution provides a high-quality data foundation that fits real-world scenarios for model training by selecting standardized historical data containing meteorological parameters, pollutant concentrations, and spatiotemporal characteristics, and rationally dividing the training and validation sets. It employs an architecture combining a bidirectional long short-term memory network and an attention mechanism, which not only captures the bidirectional dependencies in time-series data but also highlights key period features through dynamic weights, enhancing the model's ability to extract complex correlation patterns. The training process, using root mean square error as the loss function and incorporating an adaptive momentum optimizer and early stopping mechanism, effectively avoids model overfitting, ensuring its generalization ability and predictive stability. This training process enables the time-series prediction model to accurately learn the correlation between meteorological conditions and dust pollution in historical data, facilitating the output of accurate future dust pollution level predictions and improving the reliability and accuracy of the prediction model.

[0057] like Figure 6 As shown, in some embodiments, step S40 may include: S41: Extract the characteristic data of the exposed population in the target area from the regional medical and health database. The characteristic data of the exposed population includes the age distribution data of the population in the target area, the proportion of patients with respiratory diseases, and the proportion of patients with cardiovascular diseases.

[0058] Specifically, this step involves establishing encrypted data interfaces with regional health commissions, disease control centers, statistics bureaus, and designated hospitals to extract characteristic data of the exposed population in the target area from medical and health databases and population statistics databases. On the one hand, it extracts the prevalence of chronic respiratory diseases (such as asthma and chronic bronchitis) and the number of cardiovascular disease patients (statistically calculated per 100,000 people) by street for the past three years. Simultaneously, it obtains the latest population distribution by age group (subdivided into four intervals: 0-6 years, 7-14 years, 15-64 years, and 65 years and above) and population density data per square kilometer. Disease data is updated monthly, and population data is updated quarterly. On the other hand, it extracts the age distribution of the population by street for the past year, calculates the proportion of each age group (accurate to two decimal places), and converts the number of respiratory and cardiovascular disease patients into the number of patients per 10,000 people (the calculation formula is: number of patients in a certain area / total population of the area × 10,000). All data, after being anonymized (removing personally identifiable information), is first geocoded and mapped to corresponding spatial units, then weighted and allocated according to the population ratio of a 1km×1km grid to the street population, forming a population characteristic dataset with spatial granularity consistent with the pollutant concentration data; basic data is updated monthly, with temporary encrypted updates during major public health events. S42: Real-time physical sensation feedback data of the target area population is collected through a pre-set environmental response questionnaire system, and dynamic weighting factors are generated based on the severity of the feedback data.

[0059] In this step, the environmental response questionnaire system operates by sending questionnaires to permanent residents of the target area via WeChat mini-programs, community health apps, and other channels. The questionnaires include five core questions: "Do you have respiratory symptoms such as cough or sore throat?", "Do you feel discomfort such as chest tightness or dizziness?", and "Severity of symptoms (mild / moderate / severe)?". A submission location option is set to associate the feedback location with the grid where the respondent is located. The questionnaire system reminds users who have not yet submitted feedback three times daily at 9:00 AM, 3:00 PM, and 9:00 PM. The collection period is 24 hours, automatically removing duplicate submissions (deduplicated by user ID) and obviously abnormal data (e.g., feedback from locations outside the target area). A dynamic weighting factor is generated based on the severity level of the feedback data, including: the percentage of people reporting severe symptoms (denoted as a) and the percentage of people reporting moderate symptoms (denoted as b) within each grid, and the dynamic weighting factor is calculated using a formula. The calculation takes values ​​ranging from 0 to 1 and is updated daily.

[0060] S43: The age sensitivity coefficient corresponding to the age distribution data of the population, the proportion data of patients with respiratory diseases, the proportion data of patients with cardiovascular diseases, and the dynamic weighting factor are weighted and fused to generate the sensitivity index of the exposed population.

[0061] In this step, firstly, age sensitivity coefficients are set: 1.5 for 0-6 years old, 1.2 for 7-14 years old, 1.3 for 65 years old and above, and 0.8 for 15-64 years old. The age sensitivity index is calculated by weighting the proportion of each age group (formula: age sensitivity index = Σ (age group proportion × corresponding coefficient)). Secondly, the sum of the proportion of patients with respiratory diseases × 1.1 + the proportion of patients with cardiovascular diseases × 1.0 is used as the disease sensitivity index. Finally, the sensitivity index of the exposed population is weighted and fused according to the formula: "Sensitivity index of exposed population = 0.4 × age sensitivity index + 0.3 × disease sensitivity index + 0.3 × dynamic weighting factor". The calculation results are mapped to a score of 0-10 (0-3 points are low sensitivity, 4-7 points are medium sensitivity, and 8-10 points are high sensitivity) and associated with the corresponding grid to form spatial sensitivity distribution data, providing a basis for population vulnerability assessment in subsequent risk assessments.

[0062] This technical solution captures the inherent health vulnerabilities of different population groups by extracting data on age distribution and disease prevalence from regional healthcare databases. It collects real-time physical feedback using an environmental response questionnaire system and generates dynamic weighting factors to reflect the immediate sensitivity of the population to the current environment. Furthermore, by weighted fusion, it combines age sensitivity coefficients, disease prevalence data, and dynamic weighting factors to generate an exposure population sensitivity index that encompasses both the inherent age and disease-related sensitivity characteristics of the population and incorporates real-time dynamic physical feedback. This achieves a comprehensive, dynamic, and accurate quantification of the sensitivity of the exposed population, providing a highly targeted basis for subsequent health risk assessments combining dust pollution predictions and regional health baseline data, effectively improving the personalization and accuracy of health risk assessments.

[0063] like Figure 7 As shown, in some embodiments, step S50 may specifically include: S51: Collect historical epidemiological statistics within a preset number of years in the target area. The historical epidemiological statistics include medical institution visit records, disease reporting data, and public health monitoring data.

[0064] This step involves collecting historical epidemiological statistics for a predetermined period (e.g., the past 5 years) within the target area. Specifically, this includes: extracting patient records (including outpatient and emergency room diagnoses, consultation times, and patient addresses) from the regional health commission's electronic medical record system; obtaining data on dust-related diseases (e.g., allergic pneumonia) from the CDC's infectious disease reporting system; and retrieving chronic disease management data (e.g., the frequency of acute asthma attacks) from the public health monitoring platform. Data collection is conducted on an annual cycle, with annual data verification at the end of each year to remove duplicate records (using unique patient identifiers) and erroneous diagnoses (e.g., non-dust-related underlying diseases), ultimately forming a continuous historical database covering the entire region.

[0065] S52: Screen target disease types related to dust pollution from the historical epidemiological statistics, including respiratory diseases, cardiovascular diseases and allergic diseases.

[0066] In this step, when screening for target disease types related to dust pollution, authoritative domestic and international research literature is used to clearly define respiratory diseases as including acute upper respiratory tract infections, bronchial asthma, and acute exacerbations of chronic obstructive pulmonary disease; cardiovascular diseases as including hypertensive emergencies, angina attacks due to coronary heart disease, and stroke; and allergic diseases as including allergic rhinitis and allergic conjunctivitis. Precise matching using the International Classification of Diseases (ICD-10) codes is employed; for example, asthma corresponds to J45, and acute upper respiratory tract infections correspond to J06. This ensures that the screened disease types have a clear causal relationship with dust pollution, excluding diseases primarily caused by genetic, dietary, or other factors.

[0067] S53: Extract the incidence rate of each target disease type at preset time intervals, including the frequency of disease occurrence under different seasons and different levels of dust pollution.

[0068] When implementing this step, the statistical unit is 100,000 people, and the calculation formula is "Incidence rate of a disease in a certain period = (Number of cases of the disease in that period / Total population of the region in the same period) × 100,000". At the same time, the incidence rate is statistically analyzed according to the season (spring March-May, summer June-August, etc.), with special emphasis on spring as it is the peak season for sandstorms; the degree of sandstorm pollution is also divided into mild (… ), moderate ( ), severe ( The three levels are used to calculate the frequency of disease occurrence under the corresponding level, forming a three-dimensional correlation table of time, pollution level and incidence rate.

[0069] S54: Based on the historical epidemiological statistics, identify and count the proportion of people in the target area who are sensitive to weather changes. The weather-sensitive population includes the elderly, children, patients with chronic diseases, and people with sensitive constitutions.

[0070] In this step, the elderly population is defined as those aged 65 and above, and the children population is defined as those aged 0-14, with data sourced from the regional population census database. The chronic disease patient population includes diagnosed patients with asthma, coronary heart disease, and stage III hypertension, statistically analyzed through chronic disease management records from the medical insurance system. The population with sensitive constitutions is defined by medical records showing a history of allergic reactions following dust exposure within the past three years (e.g., anaphylactic shock, severe rashes). The proportion of each group to the total population of each street-level administrative unit is calculated, for example, "Percentage of the population aged 65 and above in a certain street = (Number of people aged 65 and above in the street / Total population of the street) × 100%", with the result accurate to one decimal place and linked to GIS coordinates to form spatial distribution data.

[0071] S55: Standardize the extracted incidence rate data and the proportion of weather-sensitive population data to eliminate differences between different statistical periods and data sources.

[0072] In this step, for incidence rate data, different statistical periods (e.g., some hospitals use weekly data, some use monthly data) are uniformly converted into monthly averages. The range standardization formula "Standardized Incidence Rate = (Original Incidence Rate - Lowest Regional Incidence Rate) / (Highest Regional Incidence Rate - Lowest Regional Incidence Rate)" is used to map it to the [0,1] interval. For the proportion of weather-sensitive populations, statistical differences between data sources are eliminated (e.g., census data is year-end data, medical insurance records are dynamic data), and corrections are made based on the population snapshot on July 1st each year to ensure data comparability across different years in the same region. The standardized data must pass a consistency test (e.g., Kappa coefficient ≥ 0.8) to verify the processing effect.

[0073] S56: Integrate the standardized incidence rate data and the proportion data of meteorologically sensitive populations to construct regional health baseline data, which includes the mean incidence rate, the peak range of incidence rate, and the distribution characteristics of meteorologically sensitive populations.

[0074] In this step, the average annual incidence rate and 95% confidence interval for each target disease over the past 5 years are calculated (e.g., the average annual incidence rate of acute upper respiratory tract infection is 3500 / 100,000, with a 95% confidence interval of 3200-3800). The peak incidence rate and range for the same period each year are extracted (e.g., the peak in March 2023 is 5200 / 100,000, with a range of 4800-5600). Street-level proportion data of meteorologically sensitive populations are summarized to generate a spatial distribution heat map (red indicates a proportion >30%, blue indicates <10%). Finally, a structured database containing time dimension (monthly baseline), spatial dimension (street-level distribution), and disease type dimension is formed. The data format uses a hybrid storage of CSV and shapefiles to support spatial matching with subsequent prediction models.

[0075] This technical solution systematically collects historical epidemiological data and screens for dust-related target diseases. It extracts the incidence rates and proportions of meteorologically sensitive populations under different seasons and pollution levels. After standardization, the data is integrated into regional health baseline data that includes the average incidence rate, peak range, and distribution of sensitive populations. This eliminates data discrepancies to ensure comparability and comprehensively reflects the historical health status and sensitive population characteristics related to dust pollution in the target area. It provides a scientific historical benchmark and population distribution basis for subsequent dynamic health risk assessment, improving the targeting and accuracy of risk assessment.

[0076] like Figure 8 As shown, in some embodiments, step S60 may include: S61: Perform spatiotemporal matching processing on the predicted dust pollution level, the sensitivity index of the exposed population, and the regional health baseline data, align the time granularity of the three to a preset evaluation period, and unify the spatial range of the three to the smallest spatial unit of the target area. The regional health baseline data includes the mean incidence rate of diseases, the peak range of incidence rate, and the distribution characteristics of meteorologically sensitive populations.

[0077] Specifically, this step aligns the predicted dust pollution level (1-hour temporal granularity, 1km×1km spatial unit), the sensitivity index of exposed populations (daily updated, 1km×1km spatial unit), and the regional health baseline data (originally monthly temporal granularity, street-level spatial unit). Temporally, monthly baseline data is refined to a 1-hour assessment period through linear interpolation to ensure consistency in time stamps across the three. Spatially, street-level baseline data is allocated to the smallest 1km×1km spatial unit according to population weight (for example, if a street contains 5 grids, the baseline data is split according to the proportion of each grid's population to the street's total population). Ultimately, this ensures a one-to-one correspondence between the three types of data at the "1-hour×1km×1km" spatiotemporal scale, providing a unified input dimension for subsequent calculations.

[0078] S62: Based on the historical mean and peak range of incidence rates in the regional health baseline data, the basic correlation coefficient between dust pollution levels and health risks is calculated using a regression model.

[0079] Specifically, the basic correlation coefficient is calculated using a multiple linear regression model. This model uses the historical mean incidence rate from the regional health baseline data as the benchmark, different dust pollution levels (PM10 concentration) as independent variables, and the deviation of the incidence rate from the mean (increase in incidence rate) as the dependent variable, all incorporated into the regression model. The model training data consists of paired "PM10 concentration-incidence rate" data from the same period over the past five years. The regression coefficient, i.e., the basic correlation coefficient (unit: increase in incidence rate), is obtained through least squares fitting. ), indicating that for every increase in PM10 concentration The corresponding increase in health risk. For example, if the regression result is 0.02, it indicates that for every increase in PM10, the health risk increases. The incidence rate of the target disease increased by 0.02 per 100,000 compared to the mean.

[0080] S63: Using the sensitivity index of the exposed population as a correction factor, assign corresponding correction coefficients according to the sensitivity level, and multiply the correction coefficients by the basic correlation coefficients to obtain the dynamic correlation coefficients.

[0081] Specifically, this step involves classifying the sensitivity index of the exposed population (0-3 points for low sensitivity, 4-7 points for moderate sensitivity, and 8-10 points for high sensitivity), with corresponding correction coefficients of 0.8, 1.0, and 1.2, respectively. For each 1km × 1km spatial unit, the basic correlation coefficient of that unit is multiplied by the correction coefficient corresponding to its own sensitivity level to obtain the dynamic correlation coefficient. For example, if a grid has a basic correlation coefficient of 0.02 and a sensitivity index of 9 points (high sensitivity), then its dynamic correlation coefficient = 0.02 × 1.2 = 0.024, thus reflecting the differentiated impact of different population sensitivities on the strength of the association with health risks.

[0082] S64: The dynamic correlation coefficients of each smallest spatial unit are spatially aggregated using a weighted summation algorithm to obtain the aggregation result. The weight values ​​of the weighted summation algorithm are determined based on the population density ratio of each smallest spatial unit.

[0083] Specifically, this step first calculates the population density percentage of each smallest spatial unit (1km × 1km), i.e., the population density of a unit divided by the total average population density of the target area (weight values ​​range from 0 to 1); then, it multiplies the dynamic correlation coefficient of each unit by its corresponding weight value, and finally sums the products of all units to obtain the aggregated result of the dynamic correlation coefficients for the entire target area. For example, if the dynamic correlation coefficients of the three grids in the area are 0.024, 0.018, and 0.020, and the population density percentages are 0.4, 0.3, and 0.3, then the aggregated result is... .

[0084] S65: Substitute the aggregation result into the dynamic health risk assessment model to calculate the comprehensive health risk index within the preset interval.

[0085] In S65, the comprehensive health risk index is calculated using a dynamic health risk assessment model: the model uses a sigmoid function (logistic function) to map the spatial aggregation results to a preset range of 0-100. The formula for the comprehensive health risk index is as follows: , where k is the shape parameter (value 2.5, controlling the steepness of the curve), and θ is the threshold parameter (value 0.015, corresponding to a medium risk level). For example, when the aggregation result is 0.0204, substituting it into the formula yields an index of approximately 68 points, which intuitively reflects the overall health risk level of the target area within the preset assessment period; a higher value indicates a higher risk.

[0086] This technical solution ensures the spatiotemporal consistency of the assessment by spatiotemporally matching predicted dust pollution values, population sensitivity indicators, and health baseline data. It calculates a basic correlation coefficient based on historical data and assigns a correction coefficient based on population sensitivity, dynamically reflecting the correlation between pollution and health risks and population differences. Finally, it obtains a comprehensive health risk index through population density-weighted aggregation and model calculation. This integrates multi-dimensional information and highlights the overall risk characteristics of the region, ultimately achieving a precise quantitative assessment of the comprehensive health risk within a preset range for the target area, providing a scientific basis for subsequent tiered early warning systems.

[0087] like Figure 9 As shown, in some embodiments, step S70 includes: S71: The comprehensive health risk index is divided into multiple warning levels, each warning level corresponds to a preset index range, and different warning levels are matched with different emergency response measures.

[0088] In this step, the comprehensive health risk index is divided into four levels of warning: Level 1 (blue warning) corresponds to an index of 0-25 points, which is low risk, and the corresponding emergency response measures are to release popular science information on sand and dust pollution and daily protection tips through public platforms; Level 2 (yellow warning) corresponds to an index of 26-50 points, which is low to medium risk, and the measures include strengthening ventilation guidance for key places such as schools and nursing homes, and increasing respiratory disease consultation windows in community health service centers; Level 3 (orange warning) corresponds to an index of 51-75 points, which is medium to high risk, and the measures include advising children, the elderly and patients with chronic diseases to reduce outdoor activities, and medical institutions to activate emergency plans and increase the number of medical staff; Level 4 (red warning) corresponds to an index of 76-100 points, which is high risk, and the measures include suspending large-scale outdoor activities, temporarily closing primary and secondary schools and kindergartens, and increasing the number of ambulances on standby at emergency centers. The index range and measures for each level are all written into the regional emergency management regulations to ensure uniform implementation standards.

[0089] S72: During the dynamic adjustment of the preset dynamic threshold, in addition to the historical peak fluctuation range in the regional health baseline data, it is also corrected by combining the current seasonal characteristics of the target area and the trend of the incidence of dust-related diseases in the most recent preset number of days.

[0090] In this step, the adjustment of the preset dynamic threshold adopts the calculation method of "basic threshold and correction coefficient": the basic threshold is determined based on the fluctuation range of historical incidence rate peaks in the regional health baseline data (for example, the basic threshold for a red alert is 75 points, corresponding to the 90th percentile of historical peaks); the seasonal characteristic correction coefficient is set according to the intensity of seasonal dust storm activity in the target area, taking 0.9 in spring (the peak dust storm season) (threshold lowered by 10%) and 1.1 in winter (threshold raised by 10%); the incidence rate trend correction over the past 14 days is determined by calculating the month-on-month change rate of the incidence rate. If the incidence rate shows an upward trend for 7 consecutive days (daily average increase ≥ 5%), the correction coefficient is lowered by 0.05; conversely, if it shows a downward trend, it is raised by 0.05. The final dynamic threshold = basic threshold × seasonal coefficient × trend coefficient, ensuring that the threshold can dynamically adapt to actual risk changes.

[0091] S73: When the comprehensive health risk index exceeds the preset dynamic threshold of the corresponding warning level, a graded warning signal is generated, which includes the warning level, the scope of risk impact, and targeted protection recommendations.

[0092] In this step, the generation of tiered early warning signals includes: the warning level (clearly marked as Level 1 to Level 4 and their corresponding colors); the scope of risk impact (high-risk areas are marked on a 1km×1km grid using a GIS map, accurate to the street-level administrative unit); targeted protection recommendations (divided into two categories: the general population and sensitive populations; the general population is advised to wear masks when going out and wash their mouth and nose after returning home; sensitive populations are advised to stay at home and wear N95 masks and carry emergency medications when necessary to go out); and a 24-hour pollution trend forecast and 24-hour consultation telephone numbers for medical institutions are also included. After the signal is generated, it is reviewed and confirmed by the regional emergency command center to ensure the accuracy of the information.

[0093] S74: The graded early warning signals are released in real time to user terminals, community public information display platforms, medical institution terminals and meteorological service platforms through communication interfaces, wherein user terminals include mobile smart devices and wearable health monitoring devices.

[0094] In this step, the issuance of warning signals is achieved through multi-channel communication interfaces. It connects to mobile operator API interfaces to push SMS warnings (including risk levels and simplified suggestions) to mobile phone users within the target area; through the IoT interface of the community public information display platform, detailed warning maps and protection guidelines are displayed on community electronic screens and bulletin boards; it calls the medical institution information system interface to push high-risk area distribution and patient reception prompts to hospital emergency and respiratory departments; and it shares data interfaces with the meteorological service platform to integrate warning signals into the health risk module of the meteorological APP. For wearable health monitoring devices, personalized warnings are pushed via Bluetooth protocol (e.g., sending reminders to chronic disease patients to reduce outdoor activities), achieving accurate delivery and multi-scenario coverage of warning information.

[0095] This technical solution divides the comprehensive health risk index into multiple levels and matches them with corresponding emergency measures. It dynamically adjusts the warning threshold based on seasonal characteristics and recent incidence trends, ensuring the pertinence and adaptability of the warnings. It generates graded signals that include the warning level, the scope of impact, and protective recommendations, and releases them to various terminals in real time through multiple channels. This achieves accurate reach and comprehensive coverage of warning information, enabling the public, communities, medical institutions, and other stakeholders to take timely and appropriate measures, effectively reducing the health risks caused by sandstorm pollution and improving the timeliness and effectiveness of risk response.

[0096] Example 2 like Figure 10 As shown, this embodiment provides a dust pollution health risk early warning terminal, which includes: The data acquisition module is used to acquire multi-source input data in real time, including meteorological parameter data, dust pollutant concentration data, and geographical environment data. The data preprocessing module is used to preprocess the multi-source input data using a data fusion algorithm to obtain a standardized dataset; The prediction module is used to output a predicted value of the dust pollution level for a preset time period in the future, based on the standardized dataset and a trained time series prediction model. The sensitivity index construction module is used to acquire the exposure population characteristic data of the target area in real time, and construct the exposure population sensitivity index based on the exposure population characteristic data and dynamic feedback data of the target area; The baseline data construction module is used to extract the incidence rate of dust-related diseases and the proportion of meteorologically sensitive populations based on historical epidemiological statistics, and to construct regional health baseline data based on the incidence rate and the proportion of meteorologically sensitive populations. The risk index calculation module is used to calculate a comprehensive health risk index based on the predicted dust pollution level, the sensitivity index of the exposed population, and the regional health baseline data, using a dynamic health risk assessment model. The early warning module is used to generate a graded early warning signal when the comprehensive health risk index exceeds a preset dynamic threshold. The preset dynamic threshold is dynamically adjusted according to the historical peak fluctuation range in the regional health baseline data, and the graded early warning signal is published to the user terminal or early warning platform in real time through the communication interface.

[0097] In some embodiments, the data acquisition module includes: The meteorological parameter acquisition unit is used to collect meteorological parameter data in real time. The meteorological parameter data includes wind speed, wind direction, temperature, relative humidity and air pressure. The data comes from the meteorological monitoring system of the target area. The pollutant concentration acquisition unit is used to collect dust pollutant concentration data, including the mass concentrations of PM10, PM2.5 and total suspended particulate matter, and achieves multi-point synchronous acquisition through ground environmental monitoring stations and mobile monitoring vehicles; The geographic data acquisition unit acquires geographic environmental data through remote sensing image interpretation and geographic information system vector data. The geographic environmental data includes terrain type, vegetation coverage, surface roughness, and land use type.

[0098] In some embodiments, the data preprocessing module includes: The outlier processing unit is used to detect and correct outliers in multi-source input data. It uses interpolation to correct abnormal fluctuations in meteorological parameter data and dust pollutant concentration data, and uses mean filling to handle missing values ​​in exposed population characteristic data and geographic environment data. The spatiotemporal alignment unit is used to align data from different sources in a spatiotemporal manner, unify the temporal granularity and spatial coordinate system of the data, and ensure that the temporal resolution of meteorological parameter data and dust pollutant concentration data remains consistent. The spatial division range of the geographic environment data is matched with that of the meteorological parameter data and dust pollutant concentration data. The standardization processing unit is used to standardize the spatiotemporally aligned data. It maps various types of data to a preset numerical range through the range standardization method, eliminating the dimensional differences between different parameters. The feature fusion unit is used to perform feature fusion on standardized multi-source data using data fusion algorithms to generate a standardized dataset that includes meteorological parameter features, dust pollutant concentration features, and geographical environment features.

[0099] In some embodiments, the prediction module includes: The feature extraction unit is used to extract relevant features for dust pollution prediction from a standardized dataset. These relevant features include meteorological parameter features, dust pollutant concentration features, and geographical environment features. The model calculation unit is used to input the extracted correlation features into the trained time series prediction model. The time series prediction model calculates the dust pollution level for different preset durations in the future based on the correlation between meteorological conditions and dust pollution in historical data. The prediction output unit is used to output the predicted value of the dust pollution level for each preset time period. The predicted value includes the dust pollutant concentration and distribution status of different sub-regions within the target area.

[0100] In some embodiments, the prediction module further includes a model training unit, the model training unit comprising: The data is divided into sub-units to select standardized historical data from historical periods, including meteorological parameter sequences, dust pollutant concentration sequences, and static geographical environmental characteristics collected at fixed time intervals, which are divided into training sets and validation sets according to a preset ratio. The architecture building subunit is used to build a time series prediction model using an architecture that combines bidirectional long short-term memory network and attention mechanism. The input layer receives the associated feature sequence, extracts the time series features through a multi-layer stacked bidirectional long short-term memory network layer, assigns dynamic weights to the features at different time steps through the attention layer, and then maps them to the output layer through a fully connected layer containing several neurons to output the predicted value of the dust pollution level at a preset time in the future. The training optimization subunit is used to iteratively train the model during training, using the root mean square error as the loss function and an adaptive momentum optimizer. After each training round, the generalization ability is evaluated through the validation set. Training stops when the loss on the validation set does not decrease for a set of consecutive rounds, and the optimal model parameters are saved.

[0101] In some embodiments, the sensitivity index construction module includes: The health data extraction unit is used to extract the characteristic data of the exposed population in the target area from the regional medical and health database. The characteristic data of the exposed population includes the age distribution data of the population in the target area, the proportion of patients with respiratory diseases, and the proportion of patients with cardiovascular diseases. The dynamic feedback unit is used to collect real-time physical sensation feedback data of the target area population through a preset environmental response questionnaire system, and generate dynamic weighting factors based on the severity of the feedback data. The weighted calculation unit is used to perform weighted fusion calculation on the age sensitivity coefficient corresponding to the age distribution data of the population, the proportion data of patients with respiratory diseases, the proportion data of patients with cardiovascular diseases, and the dynamic weighting factor to generate the sensitivity index of the exposed population.

[0102] In some embodiments, the baseline data construction module includes: An epidemiological data collection unit is used to collect historical epidemiological statistics within a predetermined period of time in the target area. The historical epidemiological statistics include medical institution visit records, disease reporting data, and public health monitoring data. The disease type screening unit is used to screen target disease types related to dust pollution from historical epidemiological statistics. The target disease types include respiratory diseases, cardiovascular diseases, and allergic diseases. The incidence rate extraction unit is used to extract the incidence rate of each target disease type at preset time intervals. The incidence rate includes the frequency of disease occurrence under different seasons and different levels of dust pollution. The sensitive population statistics unit is used to identify and count the proportion of people in the target area who are sensitive to weather changes based on historical epidemiological statistics. The weather-sensitive population includes the elderly, children, patients with chronic diseases, and people with sensitive constitutions. The data standardization unit is used to standardize the extracted incidence rate data and the proportion of weather-sensitive population data to eliminate differences between different statistical periods and data sources; The baseline integration unit is used to integrate standardized incidence rate data and meteorologically sensitive population proportion data to construct regional health baseline data, which includes the mean incidence rate, the peak range of incidence rate, and the distribution characteristics of meteorologically sensitive population.

[0103] In some embodiments, the risk index calculation module includes: The spatiotemporal matching unit is used to perform spatiotemporal matching processing on the predicted value of dust pollution level, the sensitivity index of exposed population and regional health baseline data, aligning the time granularity of the three to the preset evaluation period, and unifying the spatial range of the three to the smallest spatial unit of the target area. The regional health baseline data includes the mean incidence rate of disease, the peak range of incidence rate and the distribution characteristics of meteorologically sensitive populations. The basic correlation calculation unit is used to calculate the basic correlation coefficient between dust pollution level and health risk based on the historical mean and peak range of incidence rates in regional health baseline data through a regression model. The dynamic correlation calculation unit is used to use the sensitivity index of the exposed population as a correction factor, assign corresponding correction coefficients according to the sensitivity level, and multiply the correction coefficients by the basic correlation coefficients to obtain the dynamic correlation coefficients. Spatial aggregation unit is used to spatially aggregate the dynamic correlation coefficients of each smallest spatial unit using a weighted summation algorithm to obtain the aggregation result. The weight value is determined according to the population density ratio of each smallest spatial unit. The index calculation unit is used to input the aggregated results into the dynamic health risk assessment model to calculate the comprehensive health risk index within a preset range.

[0104] In some embodiments, the early warning module includes: The level division unit is used to divide the comprehensive health risk index into multiple warning levels. Each warning level corresponds to a preset index range, and different warning levels are matched with different emergency response measures. The threshold adjustment unit is used to make corrections during the dynamic adjustment of the preset dynamic threshold, in addition to the historical peak fluctuation range in the regional health baseline data, by combining the current seasonal characteristics of the target area and the trend of the incidence of dust-related diseases in the most recent preset number of days. The signal generation unit is used to generate a graded warning signal containing the warning level, the scope of risk impact, and targeted protection recommendations when the comprehensive health risk index exceeds the preset dynamic threshold of the corresponding warning level. The signal dissemination unit is used to disseminate graded early warning signals in real time to user terminals, community public information display platforms, medical institution terminals, and meteorological service platforms via communication interfaces. The user terminals include mobile smart devices and wearable health monitoring devices.

[0105] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for early warning of health risks from sandstorm pollution, characterized in that, Includes the following steps: S10: Real-time acquisition of multi-source input data, including meteorological parameter data, dust pollutant concentration data, and geographical environment data; S20: Preprocess the multi-source input data using a data fusion algorithm to obtain a standardized dataset; S30: Based on the standardized dataset, the trained time series prediction model is used to output the predicted value of the dust pollution level for a preset time period in the future. S40: Real-time acquisition of exposed population characteristic data in the target area, and construction of an exposed population sensitivity index based on the exposed population characteristic data and dynamic feedback data in the target area; S50: Extract the incidence rate of dust-related diseases and the proportion of meteorologically sensitive populations based on historical epidemiological statistics, and construct regional health baseline data using the aforementioned incidence rate and proportion of meteorologically sensitive populations; S60: Based on the predicted dust pollution level, the sensitivity index of the exposed population, and the regional health baseline data, calculate the comprehensive health risk index using a dynamic health risk assessment model; S70: When the comprehensive health risk index exceeds the preset dynamic threshold, a graded early warning signal is generated. The preset dynamic threshold is dynamically adjusted according to the peak range of the incidence rate in the regional health baseline data, and is published to the user terminal or early warning platform in real time through the communication interface.

2. The method for early warning of health risks from sandstorm pollution according to claim 1, characterized in that, Step S10 includes: S11: Real-time collection of meteorological parameter data, including wind speed, wind direction, temperature, relative humidity and air pressure, which are obtained from the meteorological monitoring system of the target area; S12: Collect data on the concentration of dust pollutants, including the mass concentrations of PM10, PM2.5, and total suspended particulate matter; S13: Obtain geographic environment data through remote sensing image interpretation and geographic information system vector data, including terrain type, vegetation coverage, surface roughness and land use type.

3. The method for early warning of health risks from sandstorm pollution according to claim 1, characterized in that, Step S20 includes: S21: Perform outlier detection and correction on the multi-source input data. For abnormal fluctuations in the meteorological parameter data and the dust pollutant concentration data, use interpolation to correct them. For missing values ​​in the geographic environment data, use mean filling to process them. S22: Perform spatiotemporal alignment on data from different sources, unify the temporal granularity and spatial coordinate system of the data, so that the temporal resolution of the meteorological parameter data and the dust pollutant concentration data is consistent, and the spatial division range of the geographic environment data matches that of the meteorological parameter data and the dust pollutant concentration data. S23: Standardize the spatiotemporally aligned data by mapping various types of data to a preset numerical range using the range standardization method to eliminate the dimensional differences between different parameters. S24: Use data fusion algorithms to perform feature fusion on the standardized multi-source data to generate a standardized dataset that includes meteorological parameter features, dust pollutant concentration features, and geographical environment features.

4. The method for early warning of health risks from sandstorm pollution according to claim 1, characterized in that, Step S30 includes: S31: Extract the associated features for dust pollution prediction from the standardized dataset. The associated features include meteorological parameter features, dust pollutant concentration features, and geographical environment features. S32: Input the extracted correlation features into the trained time series prediction model. The time series prediction model calculates the dust pollution level for different preset durations in the future based on the correlation between meteorological conditions and dust pollution in historical data. S33: Output the predicted dust pollution level for each preset time period. The predicted value includes the dust pollutant concentration and distribution status of different sub-regions within the target area.

5. The method for early warning of health risks from sandstorm pollution according to claim 4, characterized in that, The training process of the time series prediction model includes: S321: Select standardized historical data from historical periods, including meteorological parameter sequences, dust pollutant concentration sequences, and static geographical environmental characteristics collected at fixed time intervals, and divide the standardized historical data into training sets and validation sets according to a preset ratio; S322: A time series prediction model is constructed using an architecture that combines a bidirectional long short-term memory network with an attention mechanism. The input layer receives the associated feature sequence, extracts the time series features through a multi-layer stacked bidirectional long short-term memory network, assigns dynamic weights to the features at different time steps through the attention layer, and then maps them to the output layer through a fully connected layer containing several neurons to output the predicted value of the dust pollution level at a preset time in the future. S323: During model training, the root mean square error is used as the loss function, and an adaptive momentum optimizer is used for iterative training. After each round of training, the generalization ability is evaluated through the validation set. When the loss on the validation set does not decrease for a consecutive preset number of rounds, training is stopped and the optimal model parameters are saved.

6. The method for early warning of health risks from sandstorm pollution according to claim 1, characterized in that, Step S40 includes: S41: Extract the exposed population characteristic data from the regional medical and health database. The exposed population characteristic data includes the age distribution data of the population in the target area, the proportion of patients with respiratory diseases, and the proportion of patients with cardiovascular diseases. S42: Collect real-time physical sensation feedback data of the target area population through a preset environmental response questionnaire system, and generate dynamic weighting factors based on the severity of the feedback data. S43: The age sensitivity coefficient corresponding to the age distribution data of the population, the proportion data of patients with respiratory diseases, the proportion data of patients with cardiovascular diseases, and the dynamic weighting factor are weighted and fused to generate the sensitivity index of the exposed population.

7. The method for early warning of health risks from sandstorm pollution according to claim 1, characterized in that, Step S50 includes: S51: Collect historical epidemiological statistics within a preset number of years in the target area. The historical epidemiological statistics include medical institution visit records, disease reporting data, and public health monitoring data. S52: Screen target disease types related to dust pollution from the historical epidemiological statistics, including respiratory diseases, cardiovascular diseases and allergic diseases; S53: Extract the incidence rate of each target disease type at preset time intervals, including the frequency of disease occurrence under different seasons and different levels of dust pollution; S54: Based on the aforementioned historical epidemiological statistics, identify and statistically analyze the proportion of the population in the target area that is sensitive to weather changes; S55: Standardize the extracted incidence rate data and the proportion of weather-sensitive population data to eliminate differences between different statistical periods and data sources; S56: Integrate the standardized incidence rate data and the proportion data of meteorologically sensitive populations to construct regional health baseline data, which includes the mean incidence rate, the peak range of incidence rate, and the distribution characteristics of meteorologically sensitive populations.

8. The method for early warning of health risks from sandstorm pollution according to claim 1, characterized in that, Step S60 includes: S61: Perform spatiotemporal matching processing on the predicted dust pollution level, the sensitivity index of the exposed population, and the regional health baseline data, align the time granularity of the three to a preset evaluation period, and unify the spatial range of the three to the smallest spatial unit of the target area. S62: Based on the historical mean and peak range of incidence rates in the regional health baseline data, calculate the basic correlation coefficient between dust pollution levels and health risks using a regression model; S63: Using the sensitivity index of the exposed population as a correction factor, assign corresponding correction coefficients according to the sensitivity level, and multiply the correction coefficients by the basic correlation coefficients to obtain the dynamic correlation coefficients; S64: The dynamic correlation coefficients of each minimum spatial unit are spatially aggregated using a weighted summation algorithm to obtain the aggregation result. The weight values ​​of the weighted summation algorithm are determined according to the population density ratio of each minimum spatial unit. S65: Substitute the aggregation result into the dynamic health risk assessment model to calculate the comprehensive health risk index within the preset interval.

9. The method for early warning of health risks from sandstorm pollution according to claim 1, characterized in that, Step S70 includes: S71: The comprehensive health risk index is divided into multiple warning levels, each warning level corresponds to a preset index range, and different warning levels are matched with different emergency response measures. S72: During the dynamic adjustment of the preset dynamic threshold, in addition to the historical peak fluctuation range in the regional health baseline data, it is also corrected in combination with the current seasonal characteristics of the target area and the trend of the incidence of dust-related diseases within the preset number of days; S73: When the comprehensive health risk index exceeds the preset dynamic threshold of the corresponding warning level, a graded warning signal is generated, which includes the warning level, the scope of risk impact, and targeted protection recommendations. S74: The graded early warning signals are released in real time to user terminals, community public information display platforms, medical institution terminals and meteorological service platforms through the communication interface.

10. A dust pollution health risk early warning terminal, characterized in that, include: The data acquisition module is used to acquire multi-source input data in real time, including meteorological parameter data, dust pollutant concentration data, and geographical environment data. The data preprocessing module is used to preprocess the multi-source input data using a data fusion algorithm to obtain a standardized dataset; The prediction module is used to output a predicted value of the dust pollution level for a preset time period in the future, based on the standardized dataset and a trained time series prediction model. The sensitivity index construction module is used to acquire the exposure population characteristic data of the target area in real time, and construct the exposure population sensitivity index based on the exposure population characteristic data and dynamic feedback data of the target area; The baseline data construction module is used to extract the incidence rate of dust-related diseases and the proportion of meteorologically sensitive populations based on historical epidemiological statistics, and to construct regional health baseline data based on the incidence rate and the proportion of meteorologically sensitive populations. The risk index calculation module is used to calculate a comprehensive health risk index based on the predicted dust pollution level, the sensitivity index of the exposed population, and the regional health baseline data, using a dynamic health risk assessment model. The early warning module is used to generate a graded early warning signal when the comprehensive health risk index exceeds a preset dynamic threshold. The preset dynamic threshold is dynamically adjusted according to the historical peak fluctuation range in the regional health baseline data, and the graded early warning signal is published to the user terminal or early warning platform in real time through the communication interface.

Images