Atmospheric particulate matter high-value pre-judgment method and equipment based on deep learning

Abstract

The invention provides an atmospheric particulate high-value pre-judgment method and equipment based on deep learning. A pollution condition dynamic graph is generated based on multi-source data fusion, a dynamic transmission model and a visualization technology, the time evolution of a pollution propagation path is displayed, the coverage area is dynamically adjusted along with the change of a meteorological field, and the possible PM2.5 high-value time, place and intensity within 1-24 hours in the future can be effectively predicted. The method breaks through the obvious limitations of pollution source positioning and early warning in space coverage and time resolution, achieves the quick and accurate positioning of the pollution source, and meets the business demands and daily management demands of the environmental protection supervision fields of counties, villages, streets and the like.

Description

Technical Field

[0001] The present application relates to a method and device for predicting high values of atmospheric particulate matter based on deep learning, belonging to the technical field of pollutant monitoring. Background Art

[0002] Currently, with the rapid development of the economy and industry, the emission rate of air pollution is also increasing. Air environmental pollution has become the most serious environmental safety issue, having a greater impact on people's quality of life. In order to detect and control air pollution in a timely manner, it is necessary to accurately and efficiently monitor and predict air pollutants. For this purpose, detection stations are usually arranged at various positions in the detection area to monitor the atmospheric component data in real time. Then, based on the changes in the atmospheric component data, it is judged whether air pollutants appear. When air pollutants appear, relevant stations are promptly visited to search for the pollution sources.

[0003] However, traditional air pollution source monitoring methods often rely on data collected by ground-based fixed monitoring stations. These data have obvious limitations in spatial coverage and time resolution. Moreover, since air pollutants drift continuously with the environmental wind direction, often at the downwind position of the pollution source, the characteristics of air pollutants represented by the atmospheric component data are more obvious than those at the actual pollution source position. The method of simply predicting air pollutants based on atmospheric component data has a large error, making it difficult to quickly and accurately locate the pollution source, and it also cannot meet the needs of business operations and daily management in environmental protection supervision fields such as counties, townships, and streets. Summary of the Invention

[0004] To solve the problems existing in the prior art, the present application provides a method and device for predicting high values of atmospheric particulate matter based on deep learning.

[0005] To achieve the above object, the present application provides the following technical solutions:

[0006] In a first aspect, an embodiment of the present application provides a method for predicting high values of atmospheric particulate matter based on deep learning, including:

[0007] Obtain PM2.5 concentration data, meteorological data, enterprise emission source list, and remote sensing hotspot data of a target monitoring station in a target time period, and perform cleaning and preprocessing to obtain a basic data set;

[0008] Based on the basic data set, extract features of historical high-value events to construct a high-value formation rule and a model training set; and use global wind field simulation and HYSPLIT model to analyze the diffusion and transmission paths of PM2.5, and combine with a GIS map to visually display the positions and distributions of pollution sources;

[0009] Train a pre - constructed deep - learning model using the training set and the diffusion and transmission paths;

[0010] Use the trained model to predict different points within the region to obtain prediction results.

[0011] Based on the above method, optionally, the cleaning and pre - processing includes:

[0012] Handle missing values, handle outliers, remove duplicates, correct inconsistent data, and perform standardization / normalization;

[0013] Align timestamps;

[0014] Reduce the impact of data inconsistency on the model through data assimilation technology.

[0015] Based on the above method, optionally, the feature extraction of historical high - value events based on the basic data set to construct the high - value formation rule and the model training set includes:

[0016] Use XGBoost and spatio - temporal convolutional networks to extract features from historical high - value events in the basic data set, and construct the high - value formation rule and the model training set.

[0017] Based on the above method, optionally, the pre - constructed deep - learning model includes Time Series Transformer and GNN; where

[0018] The Time Series Transformer is used to process long - time series data to improve time - series prediction performance; the GNN is used to improve the spatial correlation between monitoring stations to optimize pollutant distribution prediction.

[0019] Based on the above method, optionally, the using the trained model to predict different points within the region to obtain prediction results includes:

[0020] Based on the trained model, combined with the time - series anomaly detection algorithm Prophet, early warning is realized by detecting the mutation trend of PM2.5 concentration, high - value points are identified, and high - value areas are determined.

[0021] Based on the above method, optionally, it further includes:

[0022] Combine the influence of composite wind field data and online monitoring data of pollution sources to determine the direction of pollution sources, then give control tips, record high - value events, and add them to the database.

[0023] Based on the above method, optionally, it further includes:

[0024] Iteratively optimize the trained model using the updated data in the database.

[0025] In a second aspect, an embodiment of the present application further provides an electronic device, which includes a memory and a processor. When the processor calls and executes the computer program stored in the memory, the method for predicting high values of atmospheric particulate matter based on deep learning as described in any item of the first aspect is implemented.

[0026] In the method and device for predicting high values of atmospheric particulate matter based on deep learning provided by the present application, a dynamic pollution map is generated based on multi-source data fusion, a dynamic transmission model, and visualization technology, showing the temporal evolution of the pollution propagation path, and the coverage range is dynamically adjusted according to the change of the meteorological field, which can effectively predict the possible PM2.5 high value time, location, and intensity within 1 hour to 24 hours in the future. It breaks the obvious limitations in spatial coverage and time resolution of pollution source location and early warning, realizes the rapid and accurate location of pollution sources, and meets the needs of environmental protection supervision fields such as counties, townships, and streets in business requirements and daily management. Description of the Drawings

[0027] The drawings here are incorporated into the specification and constitute a part of this specification, showing embodiments that conform to the present application, and are used together with the specification to explain the principles of the present application. In addition, these drawings and the text description are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments.

[0028] Figure 1 It is a schematic flowchart of the method for predicting high values of atmospheric particulate matter based on deep learning provided by an embodiment of the present application;

[0029] Figure 2 It is a schematic structural diagram of the electronic device provided by an embodiment of the present application. Detailed Embodiments

[0030] To make the objectives, technical solutions, and advantages of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application. Without conflict, the following embodiments and the features in the embodiments may be combined with each other.

[0031] Through the analysis of the prior art, the current methods for predicting air pollution and their disadvantages are as follows:

[0032] 1. Atmospheric pollution prediction method based on spatio-temporal neural network base vector expansion analysis: A graph model is constructed using the static geographical attributes of the target area, and graph convolution operations are used to extract the spatial correlation of the air pollution sequence to obtain a spatial feature sequence. To extract the long-term change characteristics of air pollution data, base vectors are used to extract trend, periodic, and covariate features from the spatial feature sequence layer by layer. The above time features are input into a fully connected layer to obtain a large-scale and long-term air pollution prediction. Online air pollution prediction is carried out on the trained model, and the final air pollution prediction value is obtained by anti-normalizing the model output. In the field of air pollution prediction, this method has certain advantages. The proposed model can fully consider the spatial interaction of the atmospheric environment at each site and analyze the characteristics of the slow change of the atmosphere around the site, thus improving the prediction accuracy of long-term and large-scale air pollution.

[0033] The disadvantages of this method are as follows: Air pollutants will continuously drift with the environmental wind direction, and the characteristics of air pollutants characterized by its atmospheric component data are more obvious than the actual pollution source location. Due to the influence of wind direction and terrain, there will be pollution contribution correlations between pollution sources. Simply locating the emission location of air pollutants based on atmospheric component data and then predicting air pollutants has a large error, which affects the prediction efficiency and accuracy of air pollutants.

[0034] 2. Prediction method based on numerical simulation wind field: Combining a numerical weather prediction model and an air quality model, by simulating the emission, transmission, transformation, and deposition processes of air pollutants, predicting the PM2.5 concentration distribution in the future for a period of time, including collecting and processing site observation wind field data and multiple characteristic variable data simulated by numerical models, using the Random Forest model to screen important characteristic variables that have a significant impact on the wind field, dividing the data sets of the observed wind field and important characteristic variables into training sets, validation sets, and test sets, constructing and training a BP (Back Propagation) neural network model, and finding the optimal model by adjusting parameters, evaluating the performance of the optimal BP neural network model, and performing visualization processing; Based on the observed wind field data and numerical model data, using the Random Forest model and combining with the BP neural network, constructing an intelligent correction method for the numerical simulation wind field. This method can further improve the accuracy of wind field simulation on the basis of the dynamic numerical model simulation wind field, providing a new method for the correction of the numerical simulation wind field.

[0035] The disadvantages of this method are as follows: The modeling process based on numerical analysis usually has difficulty integrating the physical mechanism of the data itself, and there is still room for improvement in aspects such as the difficulty in capturing local and global information caused by the model lag.

[0036] 3. PM2.5 Spatiotemporal Variation Prediction System and Prediction Method Based on Neural Network: It includes a graph convolutional neural network module, a temporal convolutional neural network module, and a linear module. The graph convolutional neural network module includes a graph convolutional layer and a residual connection layer. The graph convolutional layer is used for information propagation and feature extraction in image data, and the residual connection layer performs an addition operation by adding the input features to the features processed by the graph convolutional layer; the temporal convolutional neural network includes a causal convolutional layer, a dilated convolutional layer, and a second-order residual connection layer. The causal convolutional layer is used to capture the relationship between the length of historical data and the depth of the network, the dilated convolutional layer is used to expand the receptive field of the network, and the second-order residual connection layer adds the features after the convolutional operation to the input features; the linear module is used to capture the predicted change in PM2.5 concentration.

[0037] The disadvantages of this method are as follows: Taking the time-series data of air pollution as an example, the transmission and diffusion of pollutants are extremely complex non-linear processes. In its dilution and dispersion changes, in addition to the physicochemical reactions that occur between various pollution factors, meteorology, terrain, and the underlying surface conditions are also important influencing factors. The atmospheric diffusion process under different conditions will present different change patterns, bringing very great difficulties to accurately predicting air quality.

[0038] In view of the problems existing in the above-mentioned various methods, the present invention provides a high-value prediction scheme for atmospheric particulate matter based on deep learning. This method combines the wind field data obtained by a wind profiler radar with an artificial intelligence algorithm, and uses technologies such as deep learning to analyze and mine the formation rules of high-value pollution events, and can more accurately predict the possible PM2.5 high-value time, location, and intensity within the next 1 hour to 24 hours. Early warning of possible high-value pollution events, providing a control basis for relevant departments, dynamically monitoring the pollution transmission path, and assisting emergency response. The following provides a non-limiting description of the specific implementation solutions through several examples or embodiments.

[0039] Some embodiments of the present application provide a high-value prediction method for atmospheric particulate matter based on deep learning. Refer to Figure 1 , Figure 1 is a schematic flowchart of a high-value prediction method for atmospheric particulate matter based on deep learning provided by an embodiment of the present application. As Figure 1 shown, the high-value prediction method for atmospheric particulate matter based on deep learning in this embodiment includes the following steps:

[0040] S101: Obtain the PM2.5 concentration data, meteorological data, enterprise emission source list, and remote sensing hotspot data of the target monitoring site in the target time period, and perform cleaning and preprocessing to obtain a basic data set.

[0041] Among them, meteorological data includes but is not limited to data such as wind speed, wind direction, temperature and humidity, air pressure, etc. In practice, PM2.5 concentration data and meteorological data are collected through corresponding detection equipment. The enterprise emission source list and remote sensing hotspot data are directly obtained through corresponding channels.

[0042] After obtaining the corresponding data, cleaning and preprocessing are carried out to improve the accuracy and effectiveness of the data, and finally a basic data set is obtained.

[0043] In some embodiments, the cleaning and preprocessing include:

[0044] Handling missing values, handling outliers, removing duplicates, correcting inconsistent data, and performing standardization / normalization; aligning timestamps; reducing the impact of data inconsistency on the model through data assimilation technology.

[0045] The specific content of each processing method includes:

[0046] 1. Handling missing values:

[0047] Deletion: Directly delete samples or features with too many missing values.

[0048] Filling: Fill in missing values with the mean, median, mode, fixed value, or model prediction (such as KNN, regression).

[0049] 2. Handling outliers:

[0050] Detection: Identify outliers through methods such as box plots (IQR), Z-score, DBSCAN clustering, etc.

[0051] Handling: Delete, truncate (Winsorization), replace with statistical values, or binning.

[0052] 3. Removing duplicates:

[0053] Delete exactly duplicate samples or features.

[0054] Handle approximate duplicates (such as through similarity matching).

[0055] 4. Correcting inconsistent data:

[0056] Unify the format (such as date, unit, text case).

[0057] Correct spelling mistakes (such as "USA" vs. "U.S.A").

[0058] 5. Standardization / normalization:

[0059] Standardization (Z-score): Convert the data into a distribution with a mean of 0 and a variance of 1 (suitable for linear models such as SVM, logistic regression).

[0060] Normalization (Min - Max): Scale the data to the range [0, 1] (suitable for neural networks, KNN).

[0061] Robust Scaling: Scale using the median and quartiles (suitable for data with outliers).

[0062] In addition, data assimilation techniques such as Kalman filtering are not specifically restricted here.

[0063] By cleaning and preprocessing the data, the accuracy and effectiveness of the data can be improved.

[0064] S102: Based on the basic dataset, extract features from historical high - value events to construct the high - value formation pattern and the model training set; and, analyze the diffusion and transmission paths of PM2.5 using global wind field simulation and the HYSPLIT model, and combine with the GIS map to visually display the location and distribution of pollution sources.

[0065] Specifically, extract and process features from the basic dataset to be used as model training data.

[0066] Furthermore, based on the basic dataset, extract features from historical high - value events to construct the high - value formation pattern and the model training set, including:

[0067] Use XGBoost and spatio - temporal convolutional networks to extract features from historical high - value events in the basic dataset to construct the high - value formation pattern and the model training set.

[0068] XGBoost (eXtreme Gradient Boosting) is a machine - learning algorithm based on Gradient Boosting and belongs to the Ensemble Learning method. It iteratively trains multiple weak learners (usually decision trees) and combines their results with weights to finally form a strong learner.

[0069] The core principle of XGBoost:

[0070] 1. Gradient Boosting:

[0071] Iteratively train decision trees, and each tree learns the residuals (prediction errors) of the previous tree to gradually reduce the loss function.

[0072] 2. Objective function optimization:

[0073] Objective function = Loss function (such as mean squared error)+Regularization term (to control model complexity).

[0074] Use the second-order Taylor expansion to approximate the objective function and improve the optimization efficiency.

[0075] 3. Decision tree construction:

[0076] Adopt the greedy algorithm to select the optimal splitting point and decide whether to split by Gain:

[0077] 4. Regularization:

[0078] Control the depth of the tree (max_depth), the weight of the leaf nodes (min_child_weight), etc. to prevent overfitting.

[0079] The Spatio-Temporal Convolutional Network (STCN) is a deep learning model specifically designed to process data with both spatial and temporal dimensions, and is widely used in video analysis, traffic prediction, weather forecasting, action recognition and other fields. It models the spatio-temporal dependence of data by combining spatial convolution (capturing spatial features) and temporal convolution (capturing temporal dynamics).

[0080] Through XGBoost and the spatio-temporal convolutional network, historical high-value events can be effectively extracted and the high-value formation rules can be formed, and finally the feature data required for model training can be obtained.

[0081] In addition, global wind field simulation refers to the high-precision prediction and reconstruction of the wind field (wind speed, wind direction) on the Earth's surface and in the atmosphere through numerical models or machine learning methods, and is widely used in weather forecasting, climate research, wind energy assessment, aviation and navigation and other fields. The HYSPLIT (Hybrid Single-Particle Lagrangian Integrated Trajectory) model is an atmospheric transport and diffusion model jointly developed by the US NOAA (National Oceanic and Atmospheric Administration) and ARL (Air Resources Laboratory), and is widely used in pollutant tracking, air mass movement simulation, nuclear radiation diffusion prediction and other fields. It combines the Lagrangian method (particle tracking) and the Eulerian method (grid calculation) and can simulate air mass trajectories, diffusion processes and chemical transformations.

[0082] Using global wind field simulation and the HYSPLIT model, the diffusion and transmission paths of PM2.5 can be analyzed, and then combined with the GIS map to intuitively display the location and distribution of pollution sources for users to view.

[0083] S103: Train the pre-constructed deep learning model using the training set and the diffusion and transmission paths.

[0084] Specifically, through a deep learning model, effective monitoring and early warning of environmental pollution can be achieved. Therefore, in this embodiment, a pre-constructed deep learning model is trained using a training set and the pollutant diffusion and transmission paths to obtain a corresponding prediction model.

[0085] In some embodiments, the pre-constructed deep learning model includes Time Series Transformer and GNN; among them,

[0086] Time Series Transformer (a large time series model) is a deep learning model based on the Transformer architecture, specifically designed to process time series data (such as sensor data, meteorological data, etc.). It captures long-term dependencies in the sequence through the self-attention mechanism, overcomes the gradient vanishing problem of traditional RNN / LSTM, and performs well in many tasks. In this embodiment, Time Series Transformer is used to process long-term meteorological sequence data to improve time series prediction performance.

[0087] GNN (Graph Neural Network) is a class of deep learning models specifically designed to process graph-structured data, capable of modeling the features of nodes, edges, and the global graph. Its core idea is to aggregate neighbor information through message passing and iteratively update the node representation. In this embodiment, GNN is used to improve the spatial correlation between monitoring stations to optimize the prediction of pollutant distribution.

[0088] During the model training process, performance evaluation can be carried out through a loss function to ensure that a prediction model with sufficient accuracy is obtained.

[0089] S104: Use the trained model to predict different points in the region to obtain prediction results.

[0090] Specifically, after training the required model, it can be applied to predict the concentration of pollutants (PM2.5) at different points and regions.

[0091] Furthermore, the step of using the trained model to predict different points in the region to obtain prediction results includes: based on the trained model, combined with the time series anomaly detection algorithm Prophet, early warning is realized by detecting the mutation trend of PM2.5 concentration, high-value points are identified, and high-value areas are determined.

[0092] Prophet is an open-source time series forecasting tool developed by Facebook (Meta), which is suitable for the forecasting and analysis of time series data with strong seasonality and holiday effects (such as daily active users, sales). In this embodiment, combined with the Prophet algorithm, early warning can be achieved by detecting the mutation trend of PM2.5 concentration, identifying high-value points and determining high-value areas, so as to meet the needs of business requirements and daily management.

[0093] In addition, in some embodiments, the above method further includes: determining the source direction by combining the influence of composite wind field data and online monitoring data of pollution sources, then giving control tips, and recording high-value events and adding them to the database.

[0094] Furthermore, in some embodiments, the above method further includes: iteratively optimizing the trained model using the updated data in the database.

[0095] Through the above method, it aims to achieve effective monitoring and early warning of environmental pollution through scientific methods and advanced technologies, and provide support for environmental protection and public health.

[0096] Through the above solution, a dynamic pollution status map is generated based on multi-source data fusion, dynamic transmission model and visualization technology, which shows the temporal evolution of the pollution propagation path, and the coverage range is dynamically adjusted according to the change of the meteorological field. It can effectively predict the possible PM2.5 high-value time, location and intensity within 1 hour to 24 hours in the future. It breaks the obvious limitations in the spatial coverage and time resolution of pollution source positioning and early warning, realizes the rapid and accurate positioning of pollution sources, and meets the needs of business requirements and daily management in the environmental protection supervision fields such as counties, townships and streets.

[0097] In addition, an embodiment of the present application provides an electronic device, as Figure 2 shown, the electronic device includes a memory 21 and a processor 22; wherein, the memory 21 stores a computer program, and when the processor 22 calls and executes the computer program, it realizes the method for predicting high values of atmospheric particulate matter based on deep learning in any of the above embodiments.

[0098] Among them, the electronic device can be a computer or a server, etc.

[0099] It can be understood that the same or similar parts in the above embodiments can be referred to each other, and the content not detailed in some embodiments can be referred to the same or similar content in other embodiments.

[0100] It should be noted that in the description of the present invention, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the present invention, unless otherwise specified, the meaning of "a plurality" refers to at least two.

[0101] Any process or method description shown in the flowchart or described otherwise herein can be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a specific logical function or process. The scope of the preferred embodiments of the present invention includes additional implementations, where functions may be executed in a substantially simultaneous manner or in a reverse order according to the involved functions, rather than in the order shown or discussed, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0102] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following techniques known in the art or a combination thereof can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0103] Those of ordinary skill in the art of the present technology can understand that all or part of the steps carried by the method of the above embodiments can be completed by instructing relevant hardware through a program, and this program can be stored in a computer-readable storage medium. When this program is executed, it includes one or a combination of the steps of the method embodiments.

[0104] In addition, each functional unit in the various embodiments of the present invention can be integrated into one processing module, or each unit can exist physically alone, or two or more units can be integrated into one module. The above integrated modules can be implemented in the form of hardware or in the form of software functional modules. When the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The above-mentioned storage medium can be a read-only memory, a magnetic disk, an optical disk, etc.

[0105] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0106] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for predicting high values of atmospheric particulate matter based on deep learning, characterized in that, Including: Obtain the PM2.5 concentration data, meteorological data, enterprise emission source inventory, and remote sensing hotspot data of the target monitoring site during the target time period, and perform cleaning and preprocessing to obtain a basic dataset; Based on the basic dataset, extract features of historical high-value events, construct high-value formation rules and a model training set; and, use global wind field simulation and HYSPLIT model to analyze the diffusion and transmission paths of PM2.5, and combine with GIS maps to visually display the positions and distributions of pollution sources; Use the training set and the diffusion and transmission paths to train a pre-constructed deep learning model; Use the trained model to predict different points in the region to obtain prediction results.

2. The method according to claim 1, wherein The performing cleaning and preprocessing includes: Handling missing values, handling outliers, removing duplicates, correcting inconsistent data, and performing standardization / normalization; Aligning timestamps; Through data assimilation technology, reducing the impact of data inconsistency on the model.

3. The method according to claim 1, characterized in that The based on the basic dataset, extracting features of historical high-value events, constructing high-value formation rules and a model training set, includes: Use XGBoost and spatio-temporal convolutional network to extract features of historical high-value events in the basic dataset, and construct high-value formation rules and a model training set.

4. The method according to claim 1, wherein The pre-constructed deep learning model includes TimeSeries Transformer and GNN; wherein, The Time Series Transformer is used to process long time series data to improve time series prediction performance; the GNN is used to improve the spatial correlation between monitoring sites to optimize pollutant distribution prediction.

5. The method according to claim 1, wherein The using the trained model to predict different points in the region to obtain prediction results includes: Based on the trained model, combined with the time series anomaly detection algorithm Prophet, early warning is realized by detecting the mutation trend of PM2.5 concentration, high-value points are identified, and high-value areas are determined.

6. The method according to claim 1, characterized in that, Also including: Combined with the influence of composite wind field data and on-line monitoring data of pollution sources to determine the orientation of pollution sources, and then give control tips, record high-value events, and add them to the database.

7. The method according to claim 6, characterized in that, Also including: Use the updated data in the database to iteratively optimize the trained model.

8. An electronic device, characterized in that, Including a memory and a processor, when the memory stores a computer program and the processor calls and executes the computer program, the method for high-value prediction of atmospheric particulate matter based on deep learning as described in any one of claims 1 to 7 is implemented.

Images