CN-121031344-B - Atmospheric monitoring distribution point optimizing and identifying method, device, equipment and storage medium

CN121031344BCN 121031344 BCN121031344 BCN 121031344BCN-121031344-B

Abstract

The application provides an atmospheric monitoring distribution point optimizing and identifying method, device, equipment and storage medium. Relates to the technical field of environmental monitoring. The method comprises the steps of calculating Voronoi area and neighborhood density based on longitude and latitude of sites to serve as node characteristics, combining with AQI correlation to construct a graph structure, adopting a double-layer GAT network training PollutionGNN model with residual connection to establish a three-objective optimization model of site quantity, coverage area and prediction error, and utilizing NSGA-III algorithm to realize monitoring efficiency pareto optimal solution set. The application solves the problems that the traditional point distribution method is difficult to balance cost, coverage and precision, provides theoretical basis for cross-regional application through the influence mechanism of geographic environment and pollution source structure on algorithm performance, finally realizes the upgrade of monitoring network from experience layout to intelligent optimization, and provides efficient and reusable solution for different cities.

Inventors

LI HUAN
MA XIN
DUAN JINHONG
LI XINXING
CHEN YANG
JIN LINFENG
GUO JIAYIN

Assignees

湖南工商大学

Dates

Publication Date: 20260508
Application Date: 20250825

Claims (9)

1. An atmospheric monitoring distribution point optimizing and identifying method, which is characterized by comprising the following steps: acquiring node characteristics of a monitoring site, wherein the node characteristics of the monitoring site comprise normalized longitude and latitude, voronoi area and space neighborhood density; constructing edge features, wherein the edge features comprise normalized inter-site distances and AQI correlation coefficients; constructing graph structure data based on node characteristics, edge characteristics and connection relations among sites; constructing a pollution propagation model, and training the pollution propagation model by using the graph structure data, wherein the pollution propagation model adopts a double-layer graph attention network with residual connection, and fuses node characteristics and edge characteristics through a dynamic mask mechanism to model a pollution propagation process and output a characteristic vector predicted value of each monitoring site; Establishing a three-target constraint optimization model by taking the minimum station number, the maximum coverage area and the minimum prediction error as three targets, solving the three-target constraint optimization model by using an NSGA-III algorithm to obtain a pareto optimal solution set, and outputting a monitoring station optimization layout scheme according to the pareto optimal solution set, wherein the prediction error is the prediction error of the training pollution propagation model; The connection relation between the stations is determined by the following method: setting a distance threshold and a correlation coefficient threshold; Traversing all site pairs by taking the pearson correlation coefficient as an AQI correlation coefficient, and determining that effective connection exists between two sites when the distance between the sites is smaller than the distance threshold and the absolute value of the AQI correlation coefficient is larger than the correlation coefficient threshold; and supplementing neighboring stations of the stations as connection stations for stations with the number of candidate connections smaller than the set number after screening so as to ensure that the number of the connection stations of the stations is not smaller than the set number.
2. The atmospheric monitoring distribution optimizing identification method according to claim 1, wherein the pollution propagation model is based on PollutionGNN models of a graph attention network and comprises two layers of attention layers and residual connection, and attention coefficients among sites are calculated in the attention layers according to the following formula: ; in the formula, For the neighbor set of site i, In order to pay attention to the weight vector, For a normalized attention coefficient for site i to site j, As a feature of the station i, As a feature of site j, As a feature of the neighbor site k, As a function of the index of the values, To activate the function.
3. The atmospheric monitoring distribution optimizing identification method according to claim 2, wherein residual connection is introduced in the quality inspection of two layers of attention layers, shallow layer characteristics and deep layer characteristics are added through the following formula, and an identity mapping path is formed: ; in the formula, As a characteristic transformation function of the current layer, In the form of an edge feature matrix, For the current layer to output characteristics, The features are output for the previous layer.
4. The atmospheric monitoring distribution optimization identification method according to claim 1, wherein when the graph structure data is used for training a pollution propagation model, part of sites are randomly covered by a dynamic mask, wherein the dynamic mask is expressed as: ; in the formula, Representing site mask proportions, e represents a natural constant and epoch represents training rounds.
5. The atmospheric monitoring distribution optimization identification method according to claim 1, wherein the three-objective constraint optimization model is expressed as: ; in the formula, For the number of sites to be the number of sites, In order for the coverage area to be a range, For prediction errors, T is the matrix transpose, In order to be a constraint condition, In order to measure the number of points of interest, For the minimum of the three objective functions sought, For the number of stations to be calculated, Indicating whether the ith station is selected; The number of sites is determined by the following formula: ; coverage is determined by the following formula: ; in the formula, Voronoi area for site i; the prediction error is determined by the following formula: ; in the formula, For the maximum allowable prediction error to be the most, For the predicted value of the pollution propagation model, y is the true value, n is the total number of samples, Is the mean square error.
6. The atmospheric monitoring distribution point optimization identification method according to claim 1, wherein solving the three-objective constraint optimization model by using NSGA-III algorithm to obtain pareto optimal solution set comprises: the population initialization, namely randomly generating an initial population P 0 and calculating the objective function value of each individual; generating a group of reference points according to the quantity and the distribution of the objective functions; generating a group of self-adaptive weight vectors according to the reference points; Non-dominant ranking and association, namely non-dominant ranking the population and associating each individual with the nearest reference point; Selecting, crossing and mutating, namely generating a child population Q 0 through a selection operator, a crossing operator and a mutation operator; The population updating comprises the steps of merging an initial population P 0 and a child population Q 0 to form a new population R 0 , and then carrying out non-dominant sorting and association on R 0 to select a next generation parent population P 1 ; Repeating the weight vector generation, non-dominant ordering and association, selection, crossover and mutation and population updating until the termination condition is met, and outputting the pareto optimal solution set.
7. An atmospheric monitoring distribution point optimizing and identifying device, characterized in that the device comprises: the node characteristic acquisition module is configured to acquire node characteristics of the monitoring site, wherein the node characteristics of the monitoring site comprise normalized longitude and latitude, voronoi area and space neighborhood density; The system comprises an edge feature construction module, a data processing module and a data processing module, wherein the edge feature construction module is configured to construct edge features, and the edge features comprise normalized inter-site distances and AQI correlation coefficients; The system comprises a graph structure construction module, a connection module and a connection module, wherein the graph structure construction module is configured to construct graph structure data based on node characteristics, edge characteristics and connection relations among stations, the connection relation among stations is determined by setting a distance threshold and a correlation coefficient threshold, traversing all station pairs by taking a Pearson correlation coefficient as an AQI correlation coefficient, and determining that effective connection exists between two stations when the distance among stations is smaller than the distance threshold and the absolute value of the AQI correlation coefficient is larger than the correlation coefficient threshold; The model training module is configured to construct a pollution propagation model and train the pollution propagation model by utilizing the graph structure data, wherein the pollution propagation model adopts a double-layer graph attention network with residual connection, fuses node characteristics and edge characteristics through a dynamic mask mechanism to model a pollution propagation process, and outputs a characteristic vector predicted value of each monitoring site; The optimization solving module is configured to take the minimum station number, the maximum coverage area and the minimum prediction error as three targets, establish a three-target constraint optimization model, solve the three-target constraint optimization model by using an NSGA-III algorithm to obtain a pareto optimal solution set, and output a monitoring station optimization layout scheme according to the pareto optimal solution set, wherein the prediction error is the prediction error of the training pollution propagation model.
8. An electronic device comprising a processor and a memory communicatively coupled to the processor; The memory stores computer-executable instructions; the processor executes computer-executable instructions stored in the memory to implement the atmospheric monitoring point placement optimization identification method of any one of claims 1-6.
9. A computer readable storage medium having stored therein computer executable instructions which when executed by a processor are for implementing the atmospheric monitoring point placement optimization identification method of any one of claims 1-6.

Description

Atmospheric monitoring distribution point optimizing and identifying method, device, equipment and storage medium Technical Field The present application relates to the field of environmental monitoring technologies, and in particular, to a method, an apparatus, a device, and a storage medium for optimizing and identifying atmospheric monitoring points. Background Traditional atmosphere monitoring site layout often depends on expert experience judgment or a single-target optimization method, and such layout strategies are worry about dealing with nonlinear diffusion characteristics of atmospheric pollution and differences among different areas. Because atmospheric pollution is affected by various complex factors such as meteorological conditions, topography, pollution source distribution and the like, the interaction among the factors enables the diffusion of pollutants to show high nonlinearity and space-time dynamic change. Thus, conventional monitoring site placement has significant limitations in capturing these complex interactions, especially poor performance in space-time dynamic fusion and multi-objective collaborative optimization. In addition, conventional atmospheric monitoring site placement has difficulty finding a balance between monitoring cost, coverage, and prediction accuracy. In order to improve the coverage and prediction accuracy of monitoring, the number of monitoring stations is often required to be increased, which clearly increases the monitoring cost greatly. Meanwhile, the specificity of different areas is not fully considered in the construction process of a single model, so that the model lacks versatility and cannot be directly applied to atmosphere monitoring of other areas. This limitation makes the conventional atmosphere monitoring site layout face many challenges in practical application, and it is difficult to meet the requirements of modern atmospheric environment monitoring. Therefore, there is a need to develop a more scientific and reasonable atmosphere monitoring site layout method to adapt to complex and variable atmosphere pollution conditions. Disclosure of Invention The application provides an atmospheric monitoring distribution point optimizing and identifying method, device, equipment and storage medium, which can improve the efficiency, increase the prediction precision, better capture the spatial relationship among stations and the pollutant diffusion characteristic, and promote the universality of a model for different scenes. In a first aspect, the present application provides an atmospheric monitoring distribution point optimizing and identifying method, including: acquiring node characteristics of a monitoring site, wherein the node characteristics of the monitoring site comprise normalized longitude and latitude, voronoi area and space neighborhood density; constructing edge features, wherein the edge features comprise normalized inter-site distances and AQI correlation coefficients; constructing graph structure data based on node characteristics, edge characteristics and connection relations among sites; constructing a pollution propagation model, and training the pollution propagation model by using the graph structure data, wherein the pollution propagation model adopts a double-layer graph attention network with residual connection, and fuses node characteristics and edge characteristics through a dynamic mask mechanism to model a pollution propagation process and output a characteristic vector predicted value of each monitoring site; And taking the minimum station number, the maximum coverage area and the minimum prediction error as three targets, establishing a three-target constraint optimization model, solving the three-target constraint optimization model by using an NSGA-III algorithm to obtain a pareto optimal solution set, and outputting a monitoring station optimization layout scheme according to the pareto optimal solution set, wherein the prediction error is the prediction error of the training pollution propagation model. In one possible design, the inter-site connection relationship is determined by: setting a distance threshold and a correlation coefficient threshold; Traversing all site pairs by taking the pearson correlation coefficient as an AQI correlation coefficient, and determining that effective connection exists between two sites when the distance between the sites is smaller than the distance threshold and the absolute value of the AQI correlation coefficient is larger than the correlation coefficient threshold; and supplementing neighboring stations of the stations as connection stations for stations with the number of candidate connections smaller than the set number after screening so as to ensure that the number of the connection stations of the stations is not smaller than the set number. In one possible design, the pollution propagation model is based on PollutionGNN model of a graph attention network, and comprises two