An environmental pollution risk assessment method, system, computing device, and storage medium

Abstract

The application provides an environmental pollution risk assessment method, system, computing device and storage medium. The method comprises the following steps: obtaining geographical space data and environmental pollution data of a target area; determining a first pollution risk distribution map and a pollutant diffusion path by using a target pollutant diffusion model according to the environmental pollution data; generating a second pollution risk distribution map by using a pre-trained graph convolutional neural network in combination with the geographical space data and the pollutant diffusion path; and performing fusion analysis according to the first pollution risk distribution map and the second pollution risk distribution map to obtain an environmental pollution risk assessment result of the target area. The technical solution provided by the application improves the accuracy of the environmental pollution risk assessment result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of environmental pollution risk assessment, and in particular to an environmental pollution risk assessment method and system, a computing device and a storage medium. BACKGROUND

[0002] In complex environmental scenarios such as industrial parks and urban clusters, environmental pollution risk assessment faces multiple challenges. With the intensification of industrial activities and the acceleration of urbanization, there are various types of pollutants and complex migration paths, and their diffusion is strongly coupled with meteorological conditions and geographical features. For example, pollutants may form local enrichment due to topographic barriers, or spread to downstream areas along water systems, while meteorological mutations can significantly change the diffusion direction and range. However, existing technologies are difficult to efficiently integrate multi-source dynamic data, resulting in deviations in the spatio-temporal correlation of risk assessment results.

[0003] To solve the above problems, the existing technology uses a hybrid method of diffusion model and Kriging interpolation. The specific process includes: filling in the missing pollutant concentration data of the monitoring points through Kriging interpolation to enhance data coverage, using a diffusion model (such as the Gaussian plume model) to simulate the diffusion path of the pollutant and generate a risk distribution map. This existing technology has certain advantages in handling data missing and simulating pollutant diffusion.

[0004] However, diffusion models are based on certain idealized assumptions, such as stable atmospheric conditions and uniform terrain, which are often difficult to meet in actual complex environmental scenarios. Therefore, factors such as complex changes in terrain, differences in building heights, and rapid changes in local meteorological conditions can affect the actual diffusion path of pollutants, ultimately causing a gap between the predicted environmental pollution risk assessment results and the actual situation. SUMMARY

[0005] The present application provides an environmental pollution risk assessment method, system, computing device and storage medium to solve the problem of poor accuracy of environmental pollution risk assessment results in the prior art.

[0006] In a first aspect, the embodiments of the present application provide an environmental pollution risk assessment method, comprising:

[0007] Obtaining geographical spatial data and environmental pollution data of a target area;

[0008] According to the environmental pollution data, determining a first pollution risk distribution map and a pollutant diffusion path using a target pollutant diffusion model;

[0009] Combining the geographical spatial data and the pollutant diffusion path, generating a second pollution risk distribution map using a pre-trained graph convolutional neural network;

[0010] performing fusion analysis according to the first pollution risk distribution map and the second pollution risk distribution map to obtain an environmental pollution risk assessment result of the target region.

[0011] Optionally, the determining, according to the environmental pollution data, of the first pollution risk distribution map and the pollution diffusion path by using a target pollution diffusion model comprises:

[0012] performing preprocessing on the environmental pollution data to obtain preprocessed environmental pollution data;

[0013] selecting, from a preset diffusion model library, an initial pollution diffusion model that matches both geographical environmental information of the target region and a pollution type, and adjusting parameters of the initial pollution diffusion model according to meteorological data in the preprocessed environmental pollution data to obtain a target pollution diffusion model;

[0014] simulating, according to pollution concentration distribution data in the preprocessed environmental pollution data, a diffusion process of the pollution in the target region by using the target pollution diffusion model to generate the first pollution risk distribution map and the pollution diffusion path.

[0015] Optionally, the simulating, according to pollution concentration distribution data in the preprocessed environmental pollution data, of a diffusion process of the pollution in the target region by using the target pollution diffusion model to generate the first pollution risk distribution map and the pollution diffusion path comprises:

[0016] dividing the target region into a plurality of spatial grid units based on the geospatial data, and obtaining initial pollution concentration values corresponding to the spatial grid units from the pollution concentration distribution data;

[0017] iteratively solving the target pollution diffusion model according to the initial pollution concentration values and a preset time step, calculating pollution concentration variation amounts of each spatial grid unit at a plurality of time nodes, and updating pollution concentration values according to the pollution concentration variation amounts;

[0018] mapping the updated pollution concentration values to corresponding preset risk level color scales to generate the first pollution risk distribution map;

[0019] determining, based on a spatial gradient distribution of the updated pollution concentration values, a pollution diffusion path by using a reverse particle tracking algorithm, the diffusion path comprising a pollution source position, a dynamic diffusion direction identifier, a concentration contour line, and a diffusion range boundary.

[0020] Optionally, the adjusting, according to meteorological data in the preprocessed environmental pollution data, of parameters of the initial pollution diffusion model to obtain the target pollution diffusion model comprises:

[0021] combining the pollutant diffusion equation and the settling equation into an initial pollutant diffusion model, parameters of the initial pollutant diffusion model including a benchmark diffusion coefficient and a benchmark settling rate;

[0022] generating an adjusted diffusion coefficient according to a ratio of wind speed data in meteorological data to a benchmark wind speed, in combination with a wind speed influence factor and the benchmark diffusion coefficient;

[0023] generating an adjusted settling rate according to a ratio of humidity data to temperature data in meteorological data, in combination with a temperature and humidity coupling coefficient and the benchmark settling rate;

[0024] combining the adjusted diffusion coefficient with the pollutant diffusion equation and combining the adjusted settling rate with the settling equation to obtain a target pollutant diffusion model.

[0025] Optionally, the combining the geographic space data and the pollutant diffusion path, generating a second pollution risk distribution map by using a pre-trained graph convolutional neural network, includes:

[0026] constructing a graph structure containing nodes and edges according to the geographic space data and the pollutant diffusion path, one node representing one spatial grid cell of the target area, and one edge being one pollutant diffusion path, node features of the nodes including pollutant concentration values of corresponding grid cells, and edge features of the edges including path weight coefficients;

[0027] inputting the graph structure into a pre-trained graph convolutional neural network, iteratively updating node features by multiple layers of graph convolutional layers of the pre-trained graph convolutional neural network, and a weight matrix of a convolution kernel corresponding to each layer of graph convolutional layers being obtained according to a dynamic diffusion direction identifier of the pollutant diffusion path;

[0028] mapping node features output by a last layer of graph convolutional layers to a preset pollution risk probability space to generate a second pollution risk distribution map containing a spatiotemporal correlation risk heat map and a pollutant diffusion path confidence.

[0029] Optionally, the dynamic adjustment of the weight matrix of the convolution kernel corresponding to the multiple layers of graph convolutional layers according to the dynamic diffusion direction identifier of the pollutant diffusion path includes:

[0030] decomposing a corresponding diffusion direction into a horizontal direction angle and a vertical direction angle according to the dynamic diffusion direction identifier of the pollutant diffusion path, and generating a direction vector by sinusoidal position encoding;

[0031] calculating an attention coefficient of the edge according to the direction vector and the node features;

[0032] combining the attention coefficient with a benchmark weight matrix item to generate a dynamic weight matrix;

[0033] update the weight matrix of the convolution kernel corresponding to each layer of the graph convolution layer based on the dynamic weight matrix.

[0034] Optionally, the fusion analysis based on the first pollution risk distribution map and the second pollution risk distribution map to obtain the environmental pollution risk assessment result of the target area comprises:

[0035] Based on the pollutant concentration fluctuation value, the fusion weight of the first pollution risk distribution map and the second pollution risk distribution map is dynamically adjusted to obtain comprehensive risk assessment distribution data of the target area.

[0036] According to the corresponding relationship between the preset risk level and the risk assessment interval, the target area is divided into a plurality of sub-regions of different risk levels according to the comprehensive risk assessment distribution data.

[0037] A dynamic report containing the geographical boundary of each sub-region, the risk level and the corresponding processing scheme is generated, and the dynamic report is credibly stored and traced through the blockchain technology.

[0038] In a second aspect, the embodiments of the present application provide an environmental pollution risk assessment system, comprising:

[0039] The acquisition module is configured to acquire geographical space data and environmental pollution data of a target area.

[0040] The determination module is configured to determine a first pollution risk distribution map and a pollutant diffusion path by using a target pollutant diffusion model according to the environmental pollution data.

[0041] The generation module is configured to generate a second pollution risk distribution map by using a pre-trained graph convolution neural network in combination with the geographical space data and the pollutant diffusion path.

[0042] The fusion analysis module is configured to perform fusion analysis based on the first pollution risk distribution map and the second pollution risk distribution map to obtain an environmental pollution risk assessment result of the target area.

[0043] In a third aspect, the embodiments of the present application provide a computing device, comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement the environmental pollution risk assessment method according to any one of the first aspect.

[0044] In a fourth aspect, the embodiments of the present application provide a computer storage medium storing a computer program, and the computer program is executed by a computer to implement the environmental pollution risk assessment method according to any one of the first aspect.

[0045] In the present application, a method for environmental pollution risk assessment is provided, including: obtaining geographic spatial data and environmental pollution data of a target area; determining a first pollution risk distribution map and a pollutant diffusion path using a target pollutant diffusion model according to the environmental pollution data; combining the geographic spatial data and the pollutant diffusion path, and generating a second pollution risk distribution map using a pre-trained graph convolutional neural network; and performing fusion analysis according to the first pollution risk distribution map and the second pollution risk distribution map to obtain an environmental pollution risk assessment result of the target area.

[0046] The technical solution of the present application has the following beneficial effects: efficiently integrating multi-source dynamic data to achieve accurate deduction of pollutant diffusion paths and risk classification. By using the target pollutant diffusion model and the graph convolutional neural network, not only the pollutant diffusion process can be accurately simulated, but also the spatio-temporal correlation under complex environments can be captured, thereby improving the accuracy and reliability of the environmental pollution risk assessment result.

[0047] Further, the present application improves the prediction accuracy and enhances the adaptability to complex conditions (such as instantaneous strong wind or terrain obstruction) in the actual environment by fine preprocessing of environmental pollution data, targeted selection and adjustment of diffusion model parameters, and comprehensive analysis using the graph convolutional neural network. This method effectively solves the dynamic change problem that the traditional static model cannot handle, and provides a more scientific and reasonable environmental pollution risk assessment result. In addition, by introducing the reverse particle tracking algorithm to determine the pollutant diffusion path, the pollution source positioning and its diffusion characteristics are further refined, which helps to develop more effective pollution control measures.

[0048] These aspects or other aspects of the present application will be more apparent in the following description of the embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0050] Figure 1 A flowchart of an environmental pollution risk assessment method provided by an embodiment of the present application;

[0051] Figure 2 A structural schematic diagram of an environmental pollution risk assessment system provided by an embodiment of the present application;

[0052] Figure 3 A structural schematic diagram of a computing device provided by an embodiment of the present application. Detailed Implementation

[0053] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0054] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 11, 12, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0055] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0056] To address the issue that factors such as complex terrain variations, differences in building heights, and rapid changes in local meteorological conditions can all influence the actual diffusion paths of pollutants, ultimately leading to discrepancies between predicted environmental pollution risk assessment results and actual conditions, this application provides an environmental pollution risk assessment method. The method is conceived as follows: First, it integrates geospatial data and environmental pollution data to construct a multi-dimensional spatiotemporal database, resolving the data heterogeneity problem in traditional methods. Second, it establishes a dynamic parameter adaptation mechanism based on a pollutant diffusion model, correcting the diffusion coefficient and deposition rate using real-time wind speed, temperature, and humidity data to generate a first pollution risk distribution map and diffusion paths. Third, it encodes the geospatial topology and diffusion path features into a graph structure, employing a graph convolutional neural network to capture non-Euclidean spatial relationships, generating a second risk distribution map. Finally, it fuses the outputs of the diffusion model and the graph convolutional neural network, adjusting the weight allocation based on pollutant concentration fluctuations to improve the accuracy of environmental pollution risk assessment results for the target area.

[0057] Figure 1 A flowchart of an environmental pollution risk assessment method provided for embodiments of this application is shown below. Figure 1 As shown, the method includes:

[0058] S11, acquire geographical space data and environmental pollution data of a target region.

[0059] The target region refers to a geographical space selected for analysis, evaluation or operation according to specific research purposes, project requirements or management needs. It usually has clear geographical boundaries, and a series of activities such as data collection, model simulation, risk assessment are performed within this range. The geographical boundary can be a natural boundary (such as a river, a mountain) or a man-made boundary (such as an administrative boundary, an industrial park boundary). The target region has variable size, specifically from very small local areas (such as the surrounding area of a chemical plant) to larger areas (such as a city or part of a city). The target region also has functional characteristics, which can be defined based on the integrity of the ecosystem, the location of pollution sources, population density, etc. Examples of target regions include chemical parks, communities or cities.

[0060] The geographical space data includes topography, landform, hydrogeology, etc. These data are used to describe the spatial structure of the target region. The environmental pollution data includes pollutant concentration distribution data and meteorological data, which together with the geographical space data form multi-source data. The pollutants can be particulate matter (PM) 2.5, volatile organic compounds (VOCs), heavy metal ions, etc. The unit of pollutant concentration is μg / m 3 PM2.5 refers to particulate matter with an aerodynamic diameter less than 2.5 microns. VOCs are a class of organic chemical substances that are volatile at room temperature, including formaldehyde, benzene, toluene, xylene, etc. They are widely present in industrial emissions, automobile exhaust, paint, solvents, cleaning agents, etc. Pollutant concentration distribution data reflects the distribution of different pollutants in the environment, and meteorological data includes wind speed, humidity, temperature, etc. Key factors affecting the diffusion of pollutants.

[0061] In the embodiments of the present application, geographical space data is collected through remote sensing technology, ground monitoring stations, etc. Then, environmental pollution data is obtained using sensor networks, including the real-time concentration of various pollutants in the atmosphere and the current weather conditions. Then, the original data is cleaned and corrected to ensure its accuracy and consistency. Finally, the processed data is integrated into a unified format as input for subsequent steps.

[0062] For example, in a pollution assessment project of an industrial park, first, a high-resolution camera carried by a drone is used to take a topographic map of the park and the surrounding area, and detailed geospatial data is generated in combination with existing map data. At the same time, multiple air quality monitoring points are arranged in the park to collect real-time data on pollutant concentrations such as PM2.5 and sulfur dioxide SO2, as well as meteorological parameters such as air temperature, humidity, and wind speed.

[0063] S12, according to the environmental pollution data, using the target pollutant diffusion model to determine the first pollution risk distribution map and the pollutant diffusion path.

[0064] The target pollutant diffusion model can simulate the migration and transformation rules of pollutants in the environment medium through physical equations. The parameters of the target pollutant diffusion model are dynamically adjusted according to the meteorological data. The first pollution risk distribution map is a map that visualizes the concentration distribution and risk level of pollutants in the environment medium such as air and water through color gradient (such as red to yellow to green), representing the migration and transformation rules of pollutants in the environment medium. The pollutant diffusion path is determined by a reverse tracking algorithm, including the location of the pollution source, the dominant diffusion direction, the concentration gradient contour, and other elements.

[0065] S13, combining the geospatial data and the pollutant diffusion path, using a pre-trained graph convolutional neural network to generate a second pollution risk distribution map.

[0066] The geospatial data and the diffusion path can be integrated into a graph structure in this step and input into the graph convolutional neural network. The graph structure takes spatial grids as nodes and diffusion paths as edges, and the node features include pollutant concentration and elevation data. The edge features can be composed of the direction and concentration gradient weight of the diffusion path. The graph convolutional neural network is a deep learning model that processes graph structure data, which can learn the nonlinear correlation features of multi-source data by aggregating the features of adjacent nodes to capture spatial dependencies. The second pollution risk distribution map can be a spatiotemporal correlation risk heat map and path confidence formed after fusing the nonlinear correlation features of geospatial and diffusion path.

[0067] S14, according to the fusion analysis of the first pollution risk distribution map and the second pollution risk distribution map, obtaining the environmental pollution risk assessment result of the target area.

[0068] It should be understood that, compared with independent analysis of the first pollution risk distribution map, fusion analysis can improve the accuracy of the environmental pollution risk assessment result.

[0069] The following is a specific example. In the scenario of a VOCs leak in a certain chemical industrial park, the risk of the surrounding 10km 2 area needs to be assessed. The embodiments of the present application perform the following processes:

[0070] In S11, drone aerial photography generates a 0.5m resolution DEM, identifying elevation differences (20-30m) within the factory area; VOCs concentrations (0-500ppb) and southeast wind speeds (6m / s) are collected in real-time at 50 monitoring points; Kriging interpolation completes the data for 5 fault locations, generating a 100×100 grid dataset. In S12, a Gaussian plume model is selected, with an initial diffusion coefficient D0=1.2m. 2 / s, corrected diffusion coefficient D=1.8m 2 / s; Simulation showed that after 12 hours, VOCs diffused to a certain community (concentration 120 ppb, exceeding the standard by 2.4 times); Reverse tracing identified the leak source as reactor No. 3 (confidence level 95%). In S13, a graph structure was constructed involving 10,000 nodes, encoding the diffusion direction (horizontal angle). θ =150°) corresponds to the vector [0.48,−0.87,0.09]; the pre-trained graph convolutional neural network identifies the localized accumulation of pollutant concentrations caused by adsorption in coastal wetlands. The comprehensive risk value C final =0.72. Since 0.72 is greater than the risk threshold of 0.6 corresponding to high risk, it is set as a red alert, triggering a level 1 response. The corresponding handling method is to stop work and guide residents to evacuate.

[0071] By executing steps S11-S14, this embodiment of the application constructs a high-resolution, high-coverage standardized dataset by fusing geospatial data and multi-dimensional environmental pollution data, providing accurate input for subsequent models. The target pollutant diffusion model adapts to changes in meteorological conditions by dynamically adjusting parameters (diffusion coefficient, deposition rate), improving the accuracy of pollutant migration simulation. A pre-trained graph convolutional neural network constructs a nonlinear relationship between geospatial data and diffusion paths, identifying hidden risks that are difficult to detect by traditional physical models, such as underground seepage paths. Comprehensive analysis of the first and second pollution risk distribution maps can compensate for the limitations of a single model, thereby obtaining more accurate and reliable environmental pollution risk assessment results. Furthermore, this embodiment of the application can implement corresponding early warning and emergency response strategies for risk areas of different levels, effectively reducing the impact of environmental pollution events on human health.

[0072] In one possible embodiment, S12, based on environmental pollution data, determining a first pollution risk distribution map and pollutant diffusion paths using a target pollutant diffusion model, including:

[0073] Step 121: Preprocess the environmental pollution data to obtain preprocessed environmental pollution data.

[0074] The preprocessing includes noise removal, missing value filling, and standardization processing. Therefore, the preprocessed environmental pollution data refers to the structured data set after noise removal, missing value filling, and standardization processing, including pollutant concentration (such as PM2.5, VOCs), meteorological parameters (wind speed, temperature, humidity), and the like. Among them, noise removal is to eliminate outliers caused by abnormal fluctuations of sensors or transmission interference; missing value filling is to complete the lost data due to equipment failure or network interruption by interpolation or prediction method; standardization processing is to map different dimensional data (such as concentration μg / m 3 and wind speed m / s) to a unified dimension (such as [0, 1] interval), eliminating the influence of numerical difference on the model.

[0075] As a possible implementation, the embodiment can use a sliding window mean filtering algorithm to calculate the moving average of pollutant concentration in a time window (such as 10 minutes), and replace the abnormal values exceeding the threshold (such as ±3 times the standard deviation) in the window. For example, the PM2.5 concentration of a certain monitoring point suddenly increases to 500 μg / m 3 (normal range 0-200 μg / m 3 ) within 10 minutes, and the mean value 150 μg / m 3 of other data in the window is used to replace 500 μg / m 3 . Using the Kriging interpolation algorithm, the missing data is completed based on spatial correlation. Specifically, assuming that a monitoring point is missing PM2.5 data due to failure, the weight is calculated according to the variation function (such as spherical model) of the concentration and distance of the adjacent monitoring points to generate the interpolation result. The maximum and minimum normalization method is used to map the pollutant concentration and meteorological parameters to the [0, 1] interval. For example, the PM2.5 concentration range 0-200 μg / m 3 is calculated to be 150 μg / m 3 , which is mapped to 0.75.

[0076] For example, among 20 monitoring points in a chemical industrial park, point No. 3 is missing PM2.5 data due to equipment failure. The preprocessing process is as follows: point No. 2 records 300 μg / m3 (normal value 80-120 μg / m 3 ) due to instantaneous interference, which is corrected to 110 μg / m 3 by using the sliding window mean filtering (window = 10 minutes). For the missing PM2.5 data of point No. 3, based on the concentrations (100, 105, 95 μg / m 3 ) and distance weight of points No. 1, 4, and 5, the Kriging interpolation gives the concentration of point No. 3 as 98 μg / m 3 . The PM2.5 concentration (range 0-200 μg / m 3 ) is normalized to [0, 1], such as 150 μg / m 3 to 0.75.

[0077] Step 122: Select an initial pollutant diffusion model from the preset diffusion model library that matches the geographical environment information and pollutant type of the target area. Adjust the parameters of the initial pollutant diffusion model based on the meteorological data in the preprocessed environmental pollution data to obtain the target pollutant diffusion model.

[0078] The initial pollutant diffusion model is a baseline model (such as the Gaussian plume model or the Lagrange particle model) that matches the geographical environment (such as plains or mountains) with the pollutant type (such as gas or particulate matter) from a pre-set model library. The diffusion coefficient reflects the ability of pollutants to diffuse in the medium, while the settling rate characterizes the settling speed of particulate matter due to gravity or adsorption.

[0079] This step matches an initial pollutant diffusion model from a pre-set model library containing models such as Gaussian and Lagrange models, based on the geographical environment of the target area (e.g., selecting a model that considers terrain barriers in mountainous areas) and the type of pollutant (e.g., selecting a particulate matter diffusion model for PM2.5).

[0080] For example, since the target area is a flat industrial park and the pollutants are particulate matter, a Gaussian plume model is selected as the initial pollutant diffusion model. Based on the ratio of real-time wind speed (v) to the reference wind speed (v0 = 2 m / s), the dispersion is calculated using the formula D = D0 × (v / v0). 1.2 Adjust the diffusion coefficient. For example, the initial diffusion coefficient D0 = 0.8m. 2 / s, when the wind speed v=3m / s, the corrected diffusion coefficient D=0.8×(3 / 2) 1.2 ≈1.12m 2 / s. Based on the ratio of temperature (T) to humidity (RH), the formula is S=S0×e 0.05×(RH / T) Adjust the settlement rate. For example, with an initial S0 = 0.1 m / s, when RH = 70% and T = 25℃, the corrected settlement rate S = 0.1 × e (0.05×70 / 25) ≈0.13m / s.

[0081] Step 123: Based on the pollutant concentration distribution data in the preprocessed environmental pollution data, use the target pollutant diffusion model to simulate the diffusion process of pollutants in the target area, and generate the first pollution risk distribution map and pollutant diffusion path.

[0082] The first pollution risk distribution map is a visual map of risk levels based on grid cells. Color gradients (such as red, yellow, and green) correspond to different concentration ranges; for example, red corresponds to >100 μg / m³. 3 The concentration range. The pollutant diffusion path is a contour line generated by the inverse particle tracing algorithm, showing the location of the pollution source, the dominant diffusion direction, and the concentration gradient.

[0083] In this step, the target area is divided into a 100x100 grid, and the pre-processed pollutant concentration data (e.g., PM2.5 at point 3 = 98 μg / m 3 ) is loaded. The finite difference method is used to iteratively solve the diffusion equation ∂C / ∂t = D∇ 2 C−∇×(v×C) with a time step Δt = 1 minute to simulate the 24-hour concentration spatiotemporal variation. In the above diffusion equation, ∂C / ∂t represents the rate of change of pollutant concentration C with respect to time t. If ∂C / ∂t > 0, it indicates that the pollutant concentration at a certain location is increasing. D is the diffusion coefficient, which characterizes the diffusion ability of a substance in a medium. The diffusion coefficient of PM2.5 in air is approximately 10 −5 m 2 / s, and the diffusion coefficient of ions in water is approximately 10 −9 m 2 / s. ∇ 2 C represents the Laplacian of concentration, i.e., the second-order spatial derivative of concentration, which describes the diffusion flux caused by the spatial non-uniformity of concentration. For example, the diffusion rate from a high-concentration area to a low-concentration area is determined by ∇ 2 C. v is the velocity field vector, representing the flow velocity of a fluid (such as air or water). For example, a wind speed v = 3 m / s indicates that the pollutant migrates at a speed of 3 meters per second with the airflow. ∇×(v×C) is the divergence of the velocity field and concentration product, which represents the mass transport caused by fluid flow,

[0084] and 10 3 virtual particles are released from the high-concentration area (e.g., C > 100 μg / m 5 ), and the particle trajectory is calculated in reverse according to the concentration gradient to determine the location of the pollution source (e.g., reactor 3). The final concentration is mapped to a pre-set color scale (e.g., 0-50 μg / m 3 is green, 50-100 is yellow, and >100 is red), generating the first pollution risk distribution map. The Gaussian model simulation shows that PM2.5 diffuses to the downwind 1.2 km residential area within 6 hours, with a concentration of 85 μg / m 3 (35% over standard). The reverse tracking locks the leakage source to reactor 3 (confidence 92%).

[0085] By performing steps 121-123, the embodiments of the present application eliminate noise and missing interference through data preprocessing, dynamically adjust model parameters to adapt to real-time weather conditions, and accurately simulate pollutant dispersion based on physical simulation and reverse tracking to ultimately generate high-precision risk maps and pollution source paths. In the case of PM2.5 leakage in a chemical industrial park, pre-processing reduces data errors by 15%, and the dynamically adjusted target pollutant dispersion model reduces concentration prediction errors from 25% to 8%, accurately locates the leakage source through reverse tracking, and supports a 40% increase in emergency response efficiency. The method improves the reliability and timeliness of risk assessment in complex environments.

[0086] In the above embodiment, in step 122, the parameters of the initial pollutant dispersion model are adjusted according to the weather data in the pre-processed environmental pollution data to obtain a target pollutant dispersion model, including:

[0087] Step a1, combining a pollutant dispersion equation and a settling equation into an initial pollutant dispersion model, the parameters of the initial pollutant dispersion model including a benchmark diffusion coefficient and a benchmark settling rate.

[0088] It should be understood that the pollutant dispersion equation can be ∂C / ∂t=D∇ 2 C−∇×(v×C), which can also be other expressions other than this expression, and the embodiments of the present application do not make specific limitations on the form of the expression of the pollutant dispersion equation. The settling equation is used to describe the settling behavior of particulate matter due to gravity or adsorption, and the form is ∂C / ∂t=−S×C, where S is the adjusted settling rate.

[0089] It should also be understood that the benchmark diffusion coefficient can be understood as an initial diffusion coefficient, and the benchmark diffusion coefficient D0 is the diffusion capacity under ideal steady-state conditions, such as the diffusion coefficient D0=0.8m 2 / s of PM2.5 in a windless environment. The benchmark settling rate is the settling speed under standard temperature and humidity conditions, such as S0=0.1m / s of PM2.5 at a humidity of 60% and a temperature of 20℃.

[0090] For example, the geographical environment of a certain chemical industrial park is a flat industrial area, and the pollutant type is PM2.5 particulate matter. The Gaussian plume model is selected as the initial model from the model library, and the benchmark diffusion coefficient D0=0.8m 2 / s and the benchmark settling rate S0=0.1m / s are set.

[0091] Step a2, generating an adjusted diffusion coefficient according to the ratio of wind speed data in the weather data to the benchmark wind speed, combining the wind speed influence factor and the benchmark diffusion coefficient.

[0092] Wherein, the adjusted diffusion coefficient D=D0×(v / v0) α, the wind speed factor a = 1.2, v0 is the reference wind speed, and v is the wind speed data. For example, the initial diffusion coefficient D0 = 0.8 m 2 / s, when the wind speed v = 3 m / s, the corrected diffusion coefficient D = 0.8 x (3 / 2) 1.2 ≈1.12 m 2 / s.

[0093] Step a3, according to the ratio of humidity data to temperature data in meteorological data, combined with the temperature and humidity coupling coefficient and the reference settling rate, the adjusted settling rate is generated.

[0094] The adjusted settling rate S = S0 x e β×(RH / T) , β The synergistic effect of humidity and temperature on settling is generally β = 0.05. For example, the corrected settling rate S = 0.1 x e (0.05×70 / 25) ≈0.13 m / s.

[0095] Step a4, the adjusted diffusion coefficient is combined with the pollutant diffusion equation, and the adjusted settling rate is combined with the settling equation to obtain the target pollutant diffusion model.

[0096] In this step, the parameters of the target pollutant diffusion model include diffusion coefficient, settling rate, etc., wherein the diffusion coefficient is corrected according to the ratio of real-time wind speed to reference wind speed, for example, when the wind speed is 3 m / s, D = 0.8 x (3 / 2) 1.2 ≈1.12 m 2 / s, the diffusion coefficient can be adjusted from the reference value 0.8 m 2 / s to 1.12 m 2 / s. The settling rate is corrected by an exponential function combined with temperature and humidity changes. The target pollutant diffusion model is solved by finite difference method to generate the spatial and temporal concentration distribution of the pollutant, i.e. the first pollution risk distribution map. At the same time, in this step, the reverse particle tracking algorithm traces the pollution source from the high concentration area to determine the diffusion path, such as the location of the leakage source, the dominant wind direction arrow mark, etc. The finite difference method and the reverse particle tracking algorithm are both conventional technologies, and the embodiments of the present application will not be described again.

[0097] For example, in a PM2.5 leakage event of a chemical plant, the target pollutant diffusion model simulation shows that the pollutant spreads to the downwind 1.2 km residential area within 6 hours, with a concentration of 85 μg / m 3 , the concentration exceeds 35%, generating a red high-risk area, and locking the leakage source as No. 3 reactor through reverse tracking.

[0098] By performing steps a1~a4, the embodiments of the present application can adapt to changes in meteorological conditions by dynamically adjusting the diffusion coefficient and the settling rate. In the case of a PM2.5 leakage in a chemical industrial park, the diffusion coefficient is modified from the baseline value of 0.8 to 1.12 m / s, and the settling rate is modified from 0.1 to 0.12 m / s. The model prediction accuracy is improved by 60%, the leakage source positioning error is less than 50 meters, and the emergency response efficiency is improved by 40%. This method solves the problem of insufficient adaptability of traditional static models in complex meteorological environments, and provides a reliable technical foundation for dynamic risk assessment. 2

[0099] In the above embodiment, step 123, according to the pollutant concentration distribution data in the pre-processed environmental pollution data, a target pollutant diffusion model is used to simulate the diffusion process of the pollutant in the target area, to generate a first pollution risk distribution map and a pollutant diffusion path, which includes:

[0100] Step b1, based on geographic spatial data, the target area is divided into a plurality of spatial grid units, and the initial pollutant concentration value corresponding to each spatial grid unit is obtained from the pollutant concentration distribution data.

[0101] In this step, the embodiments can divide the target area into equal-sized rectangular or hexagonal grids according to geographic spatial data (such as latitude and longitude or planar coordinate system), and each grid represents an independent calculation unit. For example, a chemical industrial park is divided into 100x100 grids, and each grid has a side length of 10 meters. The initial pollutant concentration value can refer to the pollutant concentration (such as PM2.5 concentration) of the center point of each grid extracted from the pre-processed data.

[0102] Specifically, based on geographic spatial data, the target area is divided into grids of a specified precision using geographic information system tools. For example, a 1km x 1km area is divided into 100x100 grids, each with a size of 10m x 10m. From the pollutant concentration distribution data, the discrete monitoring point data is mapped to the center point of each grid using a bilinear interpolation method. For example, the PM2.5 concentration of a monitoring point is 150μg / m 3 , and the concentrations of the grid and its adjacent grids are generated by distance-weighted interpolation.

[0103] For example, a chemical industrial park (1km x 1km) is divided into 100x100 grids (each grid is 10m x 10m). Through the pre-processed monitoring data, the initial PM2.5 concentration of the grid where the No. 3 reactor is located is 200μg / m 3 , and the adjacent grids are interpolated to 180μg / m 3 (east side), 160μg / m 3 (south side), etc.

[0104] ​Step b2, according to the initial pollutant concentration value, the target pollutant diffusion model is solved iteratively according to the preset time step, the concentration change of each spatial grid unit at multiple time nodes is calculated, and the concentration value is updated according to the concentration change.

[0105] Wherein, the preset time step is the time interval of each iteration in the exponential value simulation, which needs to meet the stability condition. For example, Δt = 1 minute, 1440 iterations are needed for 24 hours simulation. The concentration change (ΔC) is the concentration change of each grid in Δt due to diffusion, convection and sedimentation. The calculation formula of the concentration change can be: ΔC = (D∇ 2 C - ∇ × (v × C) - S × C) Δt, where the symbols in the formula are as described above, which will not be repeated here.

[0106] Specifically, the Laplacian term ∇2C and the convection term ∇ × (v × C) are calculated for each grid, ΔC is calculated according to the formula, and the concentration value Cnew = Cold + ΔC is updated. Repeat the above process until the preset simulation time (such as 24 hours) is reached. For example, in the PM2.5 diffusion simulation of the chemical park, the initial concentration of the No. 3 reactor grid is 200 μg / m 3 , after Δt = 1 minute of iteration calculation, the concentration change ΔC = -5 μg / m 3 (diffusion to adjacent grids), the updated concentration is 195 μg / m 3 . After 24 hours, the concentration of the grid 1.2 km downwind is 85 μg / m 3 .

[0107] Step b3, map the updated pollutant concentration value to the corresponding preset risk level color scale to generate a first pollution risk distribution map.

[0108] It should be understood that the preset risk level color scale can map the concentration interval to a color gradient, such as red: > 100 μg / m³ (high risk), yellow: 50-100 μg / m³ (medium risk), and green: < 50 μg / m³ (low risk). The final concentration value of each grid is converted to red, green and blue color values according to the preset color scale. For example, 85 μg / m³ corresponds to yellow (255, 255, 0). The color mapping result is rendered into a heat map using software with rendering function. For example, the simulation result of the chemical park shows that the grids around the No. 3 reactor are red (concentration > 100 μg / m 3 ), and the grids of the residential area 1.2 km away are yellow (85 μg / m 3 ), and the generated first risk distribution map is published in real time through the preset platform.

[0109] Step b4: Based on the spatial gradient distribution of the updated pollutant concentration values, the inverse particle tracing algorithm is used to determine the pollutant diffusion path. The diffusion path includes the location of the pollution source, the dynamic diffusion direction marker, the concentration contour lines, and the boundary of the diffusion range.

[0110] The reverse particle tracing algorithm releases virtual particles from high-concentration areas and calculates their trajectories in reverse based on the concentration gradient to determine the pollution source and diffusion path. Concentration contour lines are closed curves formed by connecting grids with the same concentration value, used to identify the pollution range.

[0111] For example, 10 are released in the high-risk area (red grid) of the risk distribution map. 5 Each particle is tracked. Based on the concentration gradient field, the particle's position at the previous moment is calculated using the formula corresponding to the reverse particle tracing algorithm, until the initial moment is reached. Grids through which more than 80% of the particles pass are marked as pollution sources, generating diffusion path arrows and concentration contour lines. For example, particles are released from the yellow grid (85 μg / m³) in a residential area. Reverse tracing shows that 80% of the particles pass through reactor No. 3, the diffusion path is marked with a southeast arrow, and the concentration contour lines show a pollution radius of 1.2 km.

[0112] By executing steps b1 to b4, this embodiment of the application achieves accurate prediction and source tracing of pollutant diffusion through gridded simulation, dynamic iteration, visualization mapping, and reverse tracing. In the case of a chemical industrial park, the grid division accuracy is 10m×10m, the simulation concentration error is ≤8%, the reverse tracing error in locating the pollution source is <50 meters, and the generated risk distribution map and diffusion path support improved emergency response efficiency.

[0113] In one possible embodiment, S13, combining geospatial data and pollutant diffusion paths, a second pollution risk distribution map is generated using a pre-trained graph convolutional neural network, including:

[0114] Step 131: Based on geospatial data and pollutant diffusion paths, construct a graph structure containing nodes and edges. A node represents a spatial grid cell of the target area, and an edge represents a pollutant diffusion path. The node characteristics include the pollutant concentration value of the corresponding grid cell, and the edge characteristics include the path weight coefficient.

[0115] The node features include the pollutant concentration value (e.g., PM2.5 concentration), elevation data (meters), and land use type (classification code) for that grid. Edges connect pollutant diffusion paths between two adjacent grids, and edge features include path weight coefficients (reflecting diffusion intensity) and dynamic diffusion direction indicators (e.g., a southeast wind direction angle of 135°). The path weight coefficients can be calculated based on the concentration gradient and distance along the diffusion path, using the formula w. ij =ΔC ij / d ij where ΔC ij is the concentration difference between adjacent grids i and j, d ij is the grid spacing.

[0116] In this step, the 100x100 grid (each 10m x 10m) divided in step b1 is taken as a node, and the pollutant concentration and elevation data of each node are extracted. For example, the characteristics of the node where the No. 3 reactor is located are PM2.5 = 200 μg / m 3 , and elevation = 25 m. If a diffusion path exists between two grids determined by the reverse particle tracking in step b4, an edge is established. For example, an edge is established between the node of the No. 3 reactor and the node of the adjacent grid to the east. The path weight coefficient is calculated according to the diffusion direction (e.g., southeast wind 135°) and the concentration gradient. For example, the concentration difference ΔC = 20 μg / m³, the distance d = 10 m, and the weight coefficient w = 20 / 10 = 2.0.

[0117] For example, the graph structure constructed for the chemical industrial park contains 10,000 nodes (100x100 grids), in which the node of the No. 3 reactor (node identifier 5000) establishes edges with three nodes to the east, south, and north, in which the weight coefficient of the east edge is 2.0, the direction angle is 135°; the weight coefficient of the south edge is 1.5, the direction angle is 180°; and the weight coefficient of the north edge is 0.8, the direction angle is 90°.

[0118] Step 132: inputting the graph structure into a pre-trained graph convolutional neural network, and iteratively updating the node features through multiple layers of graph convolution of the pre-trained graph convolutional neural network, wherein the weight matrix of the convolution kernel corresponding to each layer of graph convolution is obtained by dynamically adjusting according to the dynamic diffusion direction of the pollutant diffusion path.

[0119] where the node features are iteratively updated through three layers of graph convolution, and the formula of each layer is:

[0120] ;

[0121] where, is the node i feature vector in the first layer, and represents the feature of the first spatial grid unit in the target area after the first layer of graph convolution, containing information such as the pollutant diffusion risk and the geographical spatial correlation. l +1layer, and represents the feature of the first spatial grid unit in the target area after the first layer of graph convolution, containing information such as the pollutant diffusion risk and the geographical spatial correlation. i l l ​​=0) features: raw data (e.g. PM2.5 concentration 200 μg / m³, elevation 25 m) such as pollutant concentration, elevation, land use type, etc. Deep features: spatio-temporal correlation risk features learned by multiple rounds of graph convolution (e.g. risk probability 0.73). σ is the ReLU activation function, which is used to capture the non-linear correlation between pollutant diffusion paths and geographical space data, such as sudden concentration increases caused by topographic changes. Summing up the neighbor nodes j of node i aggregates the neighbor grid information associated with the diffusion paths directly connected to the current grid. N(i) includes grids connected to node i by pollutant diffusion paths (determined by backward particle tracking). Example: the neighbors of the No. 3 reactor node in the chemical park (i = 5000) include the east, south, and north grids (j1, j2, j3), corresponding to the diffusion paths, respectively. i ij is the attention coefficient between node i and neighbor j , W (l) : the learnable weight matrix of the l-th layer, which performs linear transformation on node features to extract high-order spatial correlation patterns. Example: in the chemical park model, W (l) maps the 5-dimensional input features (concentration, elevation, etc.) to 32-dimensional hidden features, capturing the coupling effect of diffusion-topography.

[0122] Example: the initial features of node 5000, 200, 25, are updated to 0.72, 0.35 (normalized risk latent features) after 3 layers of GCN, and the attention coefficient of the east side edge, 0.85, dominates feature propagation.

[0123] Step 133: Map the node features output by the last layer of graph convolution to the pre-set pollution risk probability space to generate a second pollution risk distribution map containing a spatio-temporal correlation risk heat map and a pollutant diffusion path confidence.

[0124] where the pollution risk probability space can map node features to the [0, 1] interval, representing risk probability, such as 0.7 indicating a 70% probability of high risk. The diffusion path confidence is based on the edge attention coefficient output by the graph convolutional neural network, calculating the confidence degree of the path (e.g. α = 0.85 corresponds to a confidence of 85%).

[0125] ​In this step, the node features output by the graph convolutional neural network (such as 0.72, 0.35) are input into a fully connected layer and Softmax to generate a risk probability. For example, the risk probability of node 5000 is 0.72 (high risk). The risk probability is mapped to a color gradient (red, yellow, green) to generate a spatiotemporal correlation risk heat map. The confidence of the diffusion path is marked according to the attention coefficient of the edge (such as 85% confidence on the east side). The second risk distribution map of the chemical industrial park shows that the area around the No. 3 reactor is red (risk probability 0.72), which is consistent with the results of the traditional model; an orange area (risk probability 0.68) appears in the southeast wetland, which is unique to the graph convolutional neural network (the traditional model did not detect it), and the path confidence is 78%.

[0126] By performing steps 131-133, the embodiment of the present application identifies hidden risks (such as wetland adsorption of pollution) missed by traditional models by modeling the complex correlation between geographical space and diffusion path through graph structure, combining dynamic direction encoding and attention mechanism. In the case of a chemical industrial park, the second risk distribution map adds three high-risk areas (confidence > 70%), reducing the evaluation result error by 32% compared to a single physical model, supporting the development of precise control measures.

[0127] In the above embodiment, in step 132, the weight matrix of the convolution kernel corresponding to the multiple graph convolutional layers is dynamically adjusted according to the dynamic diffusion direction of the pollutant diffusion path, including:

[0128] Step c1, according to the dynamic diffusion direction of the pollutant diffusion path, the corresponding diffusion direction is decomposed into a horizontal direction angle and a vertical direction angle, and a direction vector is generated through sinusoidal position encoding.

[0129] In this step, the graph convolutional neural network can decompose the diffusion direction into a horizontal angle θ and a vertical angle ϕ to generate a 3D direction vector [sin θ , cos θ , sin ϕ ]. For example, the southeast direction is decomposed into a horizontal angle of 135° and a vertical angle of 0°, and a vector is generated. The graph convolutional neural network can dynamically adjust the convolution kernel weight through the introduced attention mechanism, thereby capturing the nonlinear correlation of multi-source data.

[0130] For example, through the analysis of the graph convolutional neural network, it is found that the underground anti-seepage layer of a certain chemical industrial park is damaged, causing the pollutant to diffuse to the wetland area along a hidden path, which is not detected by the traditional model, but the risk probability output by the graph convolutional neural network is 0.73, which can be orange. For example, the dominant direction of the diffusion path in a certain chemical industrial park is southeast wind ( θ = 135°, ϕ= 0°), the encoding direction vector [0.707, -0.707, 0] is generated, which is used for subsequent attention calculation.

[0131] Step c2, according to the direction vector and the node feature, the attention coefficient of the edge is calculated.

[0132] According to the direction vector and the node feature, the attention weight of the edge is calculated, and the formula is a ij = softmax(W x [h i ||d ij ]), where h i is the node feature, d ij is the direction vector, and || represents the splicing operation.

[0133] Specifically, the node feature (such as PM2.5 = 200 μg / m 3 , elevation = 25 m) and the direction vector 0.707, -0.707, 0 are spliced to form an extended feature vector. The spliced features are mapped to a scalar through a learnable weight matrix W: eij =W x [h i ||d ij ], and softmax is used to normalize the attention score of the adjacent edge. For example, the extended feature of node 5000 (No. 3 reactor) and the east side edge is 200, 25, 0.707, -0.707, 0, and e ij = 1.2 is calculated through the fully connected layer, and after softmax normalization, a ij = 0.85, indicating that the contribution weight of this edge to the node update is 85%.

[0134] Step c3, combine the attention coefficient with the reference weight matrix item to generate a dynamic weight matrix. The reference weight matrix W0 is the initial learnable parameter of the graph convolution layer, which is used for feature transformation. The dynamic weight matrix W dyn is the weight adjusted by the attention coefficient, which enhances the direction-sensitive feature propagation. The attention coefficient and the reference weight matrix are combined according to the channel dimension to realize attention weighting, and the dynamic weight is expanded according to the network depth (number of layers) to ensure that each layer learns the direction mode independently. The first layer reference weight, the input feature is 5-dimensional, and the output is 32-dimensional. After the dynamic weight adjustment, the feature propagation weight of the east side edge is improved, and the weights of other edges (south, north) are reduced.

[0135] Step c4, update the weight matrix of the convolution kernel corresponding to each layer of the graph convolution layer based on the dynamic weight matrix.

[0136] By performing steps c1-c4, the embodiments of the present application enable the graph convolutional neural network to capture the direction-sensitive features of pollutant diffusion through direction encoding and dynamic weight adjustment. In the case of a chemical industrial park, the attention coefficient of 0.85 for the southeast wind direction drives the model to identify the wetland leakage path that is missed by the traditional physical model (confidence 78%), and the risk probability of 0.73 triggers an orange warning, supporting accurate investigation of the damage to the underground anti-seepage layer. The dynamic weight mechanism reduces the modeling error of the complex diffusion mode for the graph convolutional neural network, verifying the technical advantages of the direction-adaptive modeling.

[0137] In one possible embodiment, S14 performs fusion analysis on the first and second pollution risk distribution maps to obtain an environmental pollution risk assessment result of the target area, including:

[0138] Step 141: Based on the pollutant concentration fluctuation value, dynamically adjusting the fusion weights of the first and second pollution risk distribution maps to obtain comprehensive risk assessment distribution data of the target area.

[0139] It should be understood that the pollutant concentration fluctuation value σ² refers to the variance of the pollutant concentration in the target area, reflecting the degree of fluctuation in time or space. The larger σ² is, the more dramatic the concentration change is, and the higher the data uncertainty is. The variance is calculated for the concentration time series data (such as 24-hour monitoring values) of each grid cell to obtain the pollutant concentration fluctuation value σ².

[0140] In the fusion analysis process, a dynamic fusion weight can be introduced, which includes σ 2 , the contribution weight of the diffusion model, i.e., the weight of the first pollution risk distribution map w1=1 / (1+σ 2 ), and the contribution weight of the graph convolutional neural network, i.e., the weight of the second pollution risk distribution map w2=σ 2 / (1+σ 2 ).

[0141] For example, σ 2 =0.18 around the No. 3 reactor, w1=0.45, and w2=0.55 are calculated. After fusion, the concentration C final =0.45×0.85+0.55×0.73=0.79, which is determined as high risk.

[0142] Step 142: According to the corresponding relationship between the preset risk level and the risk assessment interval, the target area is divided into multiple sub-areas of different risk levels according to the comprehensive risk assessment distribution data.

[0143] For example, the risk levels are low, medium and high, where low risk (<0.3) corresponds to green, no intervention is needed; medium risk (0.3-0.6) corresponds to yellow, monitoring is recommended; high risk (>0.6) corresponds to red, immediate disposal. Optionally, adjacent grids with the same risk level are merged into continuous sub-regions using a clustering algorithm. The clustering results are converted into vector polygons.

[0144] For example, in a chemical park, the area around the No. 3 reactor C final =0.79 (red), area 1.2 km²; coastal wetland C final =0.68 (red), area 0.5 km²; other areas C final <0.3 (green). Three risk sub-regions are generated, with geographical boundaries marked.

[0145] In step 143, a dynamic report containing the geographical boundaries of each sub-region, risk level and corresponding treatment scheme is generated, and the dynamic report is credibly stored and traced through blockchain technology.

[0146] In step 143, the report hash value is calculated, and the hash value and timestamp are written into the blockchain. A unique data fingerprint is generated through a hash algorithm to ensure that the report cannot be tampered with.

[0147] By performing steps 141-143, the embodiments of the present application reduce the error of the fusion model, and the blockchain storage ensures that the evaluation results cannot be tampered with, supporting accountability and auditing; providing efficient and reliable technical support for risk assessment in complex environments.

[0148] Figure 2 The structure diagram of an environmental pollution risk assessment system provided by the embodiments of the present application is shown in Figure 2 The system comprises:

[0149] The acquisition module 21 is configured to acquire geographical space data and environmental pollution data of a target area.

[0150] The determination module 22 is configured to determine a first pollution risk distribution map and a pollution diffusion path by using a target pollutant diffusion model according to the environmental pollution data.

[0151] The generation module 23 is configured to generate a second pollution risk distribution map by using a pre-trained graph convolutional neural network in combination with the geographical space data and the pollution diffusion path.

[0152] The fusion analysis module 24 is configured to perform fusion analysis according to the first pollution risk distribution map and the second pollution risk distribution map to obtain an environmental pollution risk assessment result of the target area.

[0153] Figure 2The environmental pollution risk assessment system can perform Figure 1 The environmental pollution risk assessment method of the embodiments described above has the same implementation principles and technical effects. The specific operation modes of each module and unit of the environmental pollution risk assessment system in the above embodiments have been described in detail in the embodiments of the method, and will not be described in detail here.

[0154] In one possible design, Figure 2 The environmental pollution risk assessment system of the embodiments described above can be implemented as a computing device, such as a server. Figure 3 As shown, the computing device can include a storage component 31 and a processing component 32.

[0155] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32.

[0156] The processing component 32 is configured to: obtain geographic space data and environmental pollution data of a target area; determine a first pollution risk distribution map and a pollutant diffusion path by using a target pollutant diffusion model according to the environmental pollution data; generate a second pollution risk distribution map by using a pre-trained graph convolutional neural network in combination with the geographic space data and the pollutant diffusion path; and perform fusion analysis according to the first pollution risk distribution map and the second pollution risk distribution map to obtain an environmental pollution risk assessment result of the target area.

[0157] The processing component 32 can include one or more processors to execute the computer instructions to complete all or part of the steps in the above method. Of course, the processing component can also be one or more application-specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components, which are configured to execute the above method.

[0158] The storage component 31 is configured to store various types of data to support the operation of the terminal. The storage component can be implemented by any type of volatile or nonvolatile storage devices or a combination thereof, such as a random access memory (RAM), a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk, or a compact disk.

[0159] Of course, the computing device can also include other components, such as an input / output interface, a display component, a communication component, etc.

[0160] The input / output interface provides an interface between the processing component and peripheral interface modules, which can be output devices, input devices, etc.

[0161] The communication component is configured to facilitate wired or wireless communication between the computing device and other devices, etc.

[0162] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform, and the processing component, the storage component, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0163] The embodiments of the present application also provide a computer storage medium storing a computer program, and the computer program can implement the above-mentioned Figure 1 The environmental pollution risk assessment method of the embodiments.

[0164] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system and unit can refer to the corresponding process in the foregoing method embodiments, which will not be described herein.

[0165] The system embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0166] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and necessary universal hardware platforms, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of software products, and the computer software products can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and include a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0167] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An environmental pollution risk assessment method, characterized in that, include: Acquire geospatial data and environmental pollution data for the target area; Based on the environmental pollution data, the first pollution risk distribution map and pollutant diffusion path are determined using the target pollutant diffusion model; By combining the geospatial data and the pollutant diffusion paths, a second pollution risk distribution map is generated using a pre-trained graph convolutional neural network. By integrating and analyzing the first pollution risk distribution map and the second pollution risk distribution map, the environmental pollution risk assessment results of the target area are obtained. The step of combining the geospatial data and the pollutant diffusion path, and using a pre-trained graph convolutional neural network to generate a second pollution risk distribution map includes: Based on the geospatial data and the pollutant diffusion path, a graph structure containing nodes and edges is constructed. A node represents a spatial grid cell of the target area, and an edge represents a pollutant diffusion path. The node features include the pollutant concentration value of the corresponding grid cell, and the edge features include the path weight coefficient. The graph structure is input into a pre-trained graph convolutional neural network. The node features are updated iteratively through multiple graph convolutional layers of the pre-trained graph convolutional neural network. The weight matrix of the convolutional kernel corresponding to each graph convolutional layer is dynamically adjusted according to the dynamic diffusion direction identifier of the pollutant diffusion path. The node features output from the last graph convolutional layer are mapped to a preset pollution risk probability space to generate a second pollution risk distribution map that includes a spatiotemporal correlation risk heatmap and the confidence level of pollutant diffusion paths. The weight matrix of the convolution kernel corresponding to the multi-layer graph convolutional layer is dynamically adjusted based on the dynamic diffusion direction identifier of the pollutant diffusion path, including: Based on the dynamic diffusion direction identifier of the pollutant diffusion path, the corresponding diffusion direction is decomposed into horizontal and vertical direction angles, and a direction vector is generated by sinusoidal position encoding. Calculate the attention coefficient of the edge based on the direction vector and node features; The attention coefficients are combined with the baseline weight matrix terms to generate a dynamic weight matrix; The weight matrix of the convolution kernel corresponding to each graph convolutional layer is updated based on the dynamic weight matrix.

2. The method according to claim 1, characterized in that, The step of determining the first pollution risk distribution map and pollutant diffusion path using the target pollutant diffusion model based on the environmental pollution data includes: The environmental pollution data is preprocessed to obtain preprocessed environmental pollution data; Select an initial pollutant diffusion model from the preset diffusion model library that matches the geographical environment information and pollutant type of the target area. Adjust the parameters of the initial pollutant diffusion model according to the meteorological data in the preprocessed environmental pollution data to obtain the target pollutant diffusion model. Based on the pollutant concentration distribution data in the preprocessed environmental pollution data, the target pollutant diffusion model is used to simulate the diffusion process of pollutants in the target area, generating a first pollution risk distribution map and pollutant diffusion paths.

3. The method according to claim 2, characterized in that, The step of simulating the diffusion process of pollutants in the target area using the target pollutant diffusion model based on the pollutant concentration distribution data in the preprocessed environmental pollution data, and generating a first pollution risk distribution map and pollutant diffusion paths, includes: Based on the geospatial data, the target area is divided into multiple spatial grid units, and the initial pollutant concentration values ​​corresponding to each spatial grid unit are obtained from the pollutant concentration distribution data. Based on the initial pollutant concentration value, the target pollutant diffusion model is iteratively solved according to a preset time step, the pollutant concentration change of each spatial grid cell at multiple time nodes is calculated, and the pollutant concentration value is updated according to the pollutant concentration change. The updated pollutant concentration values ​​are mapped to the corresponding preset risk level color codes to generate the first pollution risk distribution map; Based on the spatial gradient distribution of the updated pollutant concentration values, the inverse particle tracing algorithm is used to determine the pollutant diffusion path, which includes the location of the pollution source, dynamic diffusion direction markers, concentration contour lines, and diffusion range boundaries.

4. The method according to claim 2, characterized in that, The step of adjusting the parameters of the initial pollutant diffusion model based on the meteorological data in the preprocessed environmental pollution data to obtain the target pollutant diffusion model includes: The pollutant diffusion equation and the sedimentation equation are combined to form an initial pollutant diffusion model, the parameters of which include the baseline diffusion coefficient and the baseline sedimentation rate. Based on the ratio of wind speed data in meteorological data to the baseline wind speed, combined with the wind speed influence factor and the baseline diffusion coefficient, the adjusted diffusion coefficient is generated. Based on the ratio of humidity to temperature data in meteorological data, combined with the temperature-humidity coupling coefficient and the baseline settlement rate, the adjusted settlement rate is generated. By combining the adjusted diffusion coefficient with the pollutant diffusion equation and the adjusted settling rate with the settling equation, a target pollutant diffusion model is obtained.

5. The method according to claim 1, characterized in that, The step of fusing and analyzing the first pollution risk distribution map and the second pollution risk distribution map to obtain the environmental pollution risk assessment results for the target area includes: Based on the pollutant concentration fluctuation values, the fusion weights of the first pollution risk distribution map and the second pollution risk distribution map are dynamically adjusted to obtain comprehensive risk assessment distribution data for the target area; Based on the comprehensive risk assessment distribution data and the correspondence between the preset risk level and the risk assessment interval, the target area is divided into multiple sub-areas with different risk levels. A dynamic report is generated that includes the geographical boundaries, risk levels, and corresponding treatment plans for each sub-region, and the dynamic report is reliably stored and traced using blockchain technology.

6. An environmental pollution risk assessment system, characterized in that, include: The acquisition module is used to acquire geospatial data and environmental pollution data of the target area. The determination module is used to determine the first pollution risk distribution map and the pollutant diffusion path based on the environmental pollution data and using the target pollutant diffusion model; The generation module is used to combine the geospatial data and the pollutant diffusion path, and generate a second pollution risk distribution map using a pre-trained graph convolutional neural network; The fusion analysis module is used to perform fusion analysis based on the first pollution risk distribution map and the second pollution risk distribution map to obtain the environmental pollution risk assessment results of the target area; The step of combining the geospatial data and the pollutant diffusion path, and using a pre-trained graph convolutional neural network to generate a second pollution risk distribution map includes: Based on the geospatial data and the pollutant diffusion path, a graph structure containing nodes and edges is constructed. A node represents a spatial grid cell of the target area, and an edge represents a pollutant diffusion path. The node features include the pollutant concentration value of the corresponding grid cell, and the edge features include the path weight coefficient. The graph structure is input into a pre-trained graph convolutional neural network. The node features are updated iteratively through multiple graph convolutional layers of the pre-trained graph convolutional neural network. The weight matrix of the convolutional kernel corresponding to each graph convolutional layer is dynamically adjusted according to the dynamic diffusion direction identifier of the pollutant diffusion path. The node features output from the last graph convolutional layer are mapped to a preset pollution risk probability space to generate a second pollution risk distribution map that includes a spatiotemporal correlation risk heatmap and the confidence level of pollutant diffusion paths. The weight matrix of the convolution kernel corresponding to the multi-layer graph convolutional layer is dynamically adjusted based on the dynamic diffusion direction identifier of the pollutant diffusion path, including: Based on the dynamic diffusion direction identifier of the pollutant diffusion path, the corresponding diffusion direction is decomposed into horizontal and vertical direction angles, and a direction vector is generated by sinusoidal position encoding. Calculate the attention coefficient of the edge based on the direction vector and node features; The attention coefficients are combined with the baseline weight matrix terms to generate a dynamic weight matrix; The weight matrix of the convolution kernel corresponding to each graph convolutional layer is updated based on the dynamic weight matrix.

7. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement an environmental pollution risk assessment method as described in any one of claims 1 to 5.

8. A computer storage medium, characterized in that, The device contains a computer program that, when executed by a computer, implements an environmental pollution risk assessment method as described in any one of claims 1 to 5.

Images