CN114638417A - Fine-grained air quality prediction method and system - Google Patents

Fine-grained air quality prediction method and system Download PDF

Info

Publication number
CN114638417A
CN114638417A CN202210276051.1A CN202210276051A CN114638417A CN 114638417 A CN114638417 A CN 114638417A CN 202210276051 A CN202210276051 A CN 202210276051A CN 114638417 A CN114638417 A CN 114638417A
Authority
CN
China
Prior art keywords
data
air quality
quality prediction
neural network
matrix
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210276051.1A
Other languages
Chinese (zh)
Inventor
徐文静
郝洁
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Aeronautics and Astronautics
Original Assignee
Nanjing University of Aeronautics and Astronautics
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Aeronautics and Astronautics filed Critical Nanjing University of Aeronautics and Astronautics
Priority to CN202210276051.1A priority Critical patent/CN114638417A/en
Publication of CN114638417A publication Critical patent/CN114638417A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Computational Linguistics (AREA)
  • Development Economics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Game Theory and Decision Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to a fine-grained air quality prediction method and a fine-grained air quality prediction system, which relate to the technical field of air quality prediction, and the method comprises the following steps: acquiring multi-source data of the atmosphere by using a satellite; carrying out numerical value conversion and projection operation on the aerosol data to obtain aerosol data with geographic information; performing space-time matching on the aerosol data with the geographic information, the ground monitoring point data, the meteorological data and the interest point data to obtain a node characteristic matrix; inputting the node characteristic matrix into an air-map neural network air quality prediction model to obtain an air quality prediction value; the method can solve the problems that the extraction capability of multi-source space-time big data in two dimensions of time and space is insufficient and the air quality information with high space-time resolution of a city is lost in the prior art.

Description

Fine-grained air quality prediction method and system
Technical Field
The invention relates to the technical field of air quality prediction, in particular to a fine-grained air quality prediction method and system.
Background
The rapid development of urbanization modernizes life, but also brings about a serious air pollution problem. At present, frequent haze events affect the nerves of people in the country, so that the air pollution problem is concerned unprecedentedly. The research on PM2.5 mainly starts from the beginning of the 20 th century in the national academia, and particularly, the research on urban air quality problems is remarkably advanced in recent years, so that the research on the urban air quality problems is gradually a hot spot of academic research in recent years.
Before the development of big data technology, numerical forecasting methods are mainly adopted for air quality prediction. The method requires familiarity with the principle and law of air pollutant change in the air and needs a large amount of numerical calculation, which has higher requirements on research personnel and needs more professional field experts to participate in modeling.
With the popularization of mobile terminals and various position sensors, city space-time big data is generated at ubiquitous and unprecedented speed, the daily life aspect of people is influenced, and meanwhile, important data support is provided for governments and enterprises to implement smart city construction. Big data means abundant urban knowledge, if the data is used properly, the data can help to deal with urban environment calculation, and new breakthroughs are brought to high-time and high-time prediction of air quality by researching the time-space correlation between ground air quality monitoring site data and other data sources (remote observation remote sensing satellite aerosol AOD data, meteorological data and POI (point of interest)).
At present, the relevant research of an air quality prediction model is mainly based on a Recurrent Neural Network (RNN) and a variant long-short term memory (LSTM) of the RNN and a gate control unit (GRU), and a neural network method which is good at processing serialized data can well learn the time dimension correlation of the data. However, the air quality has variation diversity in different regions, and changes in the same region along with the time, so that the air quality has strong space-time nonlinear correlation, and neglecting the spatial correlation among data can cause great errors on the model prediction result. In addition, considering that the ground monitoring points are rare, the number of air quality monitoring stations deployed in each city by the government is only dozens, so that the problems of serious sparseness, uneven spatial distribution and the like exist, and air quality prediction information with high space-time resolution cannot be provided. Although, with the development of deep learning algorithms, in recent years, the complex spatio-temporal relationship among the influencing factors of the air quality has been noticed more and more, and a spatio-temporal model is appeared to predict the air quality, the existing research, which is only based on the existing ground monitoring points to predict the air quality concentration, still cannot provide air quality information with high spatio-temporal resolution for resident trip (when to go out, when to close doors and windows) and government environmental governance decision (traffic control).
Disclosure of Invention
The invention aims to provide a fine-grained air quality prediction method and a fine-grained air quality prediction system, which are used for solving the problems that in the prior art, the extraction capability of multi-source space-time big data in two dimensions of time and space is insufficient, and the air quality information with high space-time resolution of a city is lost.
In order to achieve the purpose, the invention provides the following scheme:
a fine-grained air quality prediction method, comprising:
acquiring multi-source data of the atmosphere by using a satellite; the multi-source data comprises aerosol data, ground monitoring point data, meteorological data and interest point data;
carrying out numerical value conversion and projection operation on the aerosol data to obtain aerosol data with geographic information;
performing space-time matching on the aerosol data with the geographic information, the ground monitoring point data, the meteorological data and the interest point data to obtain a node characteristic matrix;
inputting the node characteristic matrix into an air-map neural network air quality prediction model to obtain an air quality prediction value; the air quality prediction model of the space-time diagram neural network comprises a first gate-controlled cyclic neural network, a graph convolution neural network and a second gate-controlled cyclic neural network which are connected in sequence.
Optionally, before performing space-time matching on the aerosol data with geographic information, the ground monitoring point data, the meteorological data, and the interest point data to obtain a node feature matrix, the method further includes:
and respectively carrying out abnormal value elimination, missing value filling and normalization processing on the aerosol data with geographic information, the ground monitoring point data, the meteorological data and the interest point data in sequence.
Optionally, the construction process of the graph convolution neural network specifically includes:
constructing a graph data structure according to the ground monitoring point data to obtain a spatial relationship topological graph and an adjacency weight matrix among the ground monitoring points;
determining a degree matrix of the ground monitoring point according to the spatial relationship topological graph and the adjacency weight matrix;
and constructing a graph convolution neural network according to the degree matrix and the node characteristic matrix.
Optionally, the expression of the adjacency weight matrix is:
Figure BDA0003555784750000031
wherein, wijAs weights in the adjacency weight matrix, dijIs the Euclidean distance, σ, between the station numbered i and the station j2The method comprises the steps of setting a first threshold for controlling the distribution of an initial weight matrix, setting epsilon as a second threshold for controlling the sparsity of the initial weight matrix, setting i as an ith ground monitoring point and setting j as a jth ground monitoring point.
Optionally, the loss function expression of the air quality prediction model of the space-time diagram neural network is as follows:
Figure BDA0003555784750000032
wherein the content of the first and second substances,
Figure BDA0003555784750000033
for the loss function, n is the total number of samples, yiThe value is monitored for the current site,
Figure BDA0003555784750000034
and i is the predicted value of the current station, and the ith ground monitoring point.
Optionally, the expression of the graph convolutional neural network is:
Figure BDA0003555784750000035
wherein H(l+1)For the l +1 th layer of data,
Figure BDA0003555784750000036
is a contiguous matrix with self-connection,
Figure BDA0003555784750000037
is a degree matrix, sigma is a nonlinear model Relu activation function, theta is a training parameter matrix, HlIs the l-th layer data.
A fine-grained air quality prediction system comprising:
the acquisition module is used for acquiring multi-source data of the atmosphere by using a satellite; the multi-source data comprises aerosol data, ground monitoring point data, meteorological data and interest point data;
the numerical value conversion and projection giving operation module is used for carrying out numerical value conversion and projection giving operation on the aerosol data to obtain aerosol data with geographic information;
the space-time matching module is used for performing space-time matching on the aerosol data with the geographic information, the ground monitoring point data, the meteorological data and the interest point data to obtain a node characteristic matrix;
the prediction module is used for inputting the node characteristic matrix into an air-map neural network air quality prediction model to obtain an air quality prediction value; the air quality prediction model of the space-time diagram neural network comprises a first gate-controlled cyclic neural network, a graph convolution neural network and a second gate-controlled cyclic neural network which are connected in sequence.
Optionally, the fine-grained air quality prediction system further comprises:
and the preprocessing module is used for respectively carrying out abnormal value elimination, missing value filling and normalization processing on the aerosol data with the geographic information, the ground monitoring point data, the meteorological data and the interest point data in sequence.
Optionally, the construction process of the graph convolution neural network specifically includes:
constructing a graph data structure according to the ground monitoring point data to obtain a spatial relationship topological graph and an adjacent weight matrix among the ground monitoring points;
determining a degree matrix of the ground monitoring point according to the spatial relationship topological graph and the adjacency weight matrix;
and constructing a graph convolution neural network according to the degree matrix and the node characteristic matrix.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention utilizes the satellite to obtain the multi-source data of the atmosphere; the multi-source data comprises aerosol data, ground monitoring point data, meteorological data and interest point data; carrying out numerical value conversion and projection operation on the aerosol data to obtain aerosol data with geographic information; performing space-time matching on aerosol data with geographic information, ground monitoring point data, meteorological data and interest point data to obtain a node characteristic matrix; and inputting the node characteristic matrix into an air-space neural network air quality prediction model to obtain an air quality prediction value. By carrying out space-time matching on multi-source data, the problems that in the prior art, the extraction capability of multi-source space-time big data in two dimensions of time and space is insufficient, and air quality information with high space-time resolution of a city is lost are solved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of a fine-grained air quality prediction method provided by the present invention;
FIG. 2 is a flow chart of aerosol data preprocessing provided by the present invention;
FIG. 3 is a flow chart of a model for predicting air quality of a space-time neural network according to the present invention;
FIG. 4 is a schematic diagram of an air quality prediction model of a space-time neural network provided by the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a fine-grained air quality prediction method and a fine-grained air quality prediction system, which are used for solving the problems that in the prior art, the extraction capability of multi-source space-time big data in two dimensions of time and space is insufficient, and the air quality information with high space-time resolution of a city is lost.
In order to make the aforementioned objects, features and advantages of the present invention more comprehensible, the present invention is described in detail with reference to the accompanying drawings and the detailed description thereof.
As shown in fig. 1, the fine-grained air quality prediction method provided by the present invention includes:
step 101: acquiring multi-source data of the atmosphere by using a satellite; the multi-source data includes aerosol data, ground monitoring point data, meteorological data, and interest point data. The satellite utilized in the present invention is identified as sunflower number eight (Himapari-8).
Step 102: and carrying out numerical value conversion and projection operation on the aerosol data to obtain the aerosol data with geographic information. And the endowment projection operation is to endow acquired himwari-8 aerosol data which cannot be directly extracted with geographic information. Aerosol Data in a nc format is converted into tif format image Data which is universal in the remote sensing field by using a GDAL (geographic Data extraction library) open source grid spatial Data conversion library, the longitude and latitude corresponding to each pixel of the image Data are calculated, a WSG 84 geographic coordinate system is selected as tif, geographic projection information is given to the tif, and accordingly aerosol values of corresponding points can be extracted according to the longitude and latitude of a ground monitoring station.
Step 103: and performing space-time matching on the aerosol data with the geographic information, the ground monitoring point data, the meteorological data and the interest point data to obtain a node characteristic matrix.
Step 104: inputting the node characteristic matrix into an air-map neural network air quality prediction model to obtain an air quality prediction value; as shown in fig. 4, the air quality prediction model of the time-space diagram neural network is a first gated recurrent neural network, a convolutional neural network and a second gated recurrent neural network which are connected in sequence. The predicted value of the air quality is specifically a PM2.5 predicted value.
As an optional implementation manner, before step 103, further comprising:
and respectively carrying out abnormal value elimination, missing value filling and normalization processing on the aerosol data with geographic information, the ground monitoring point data, the meteorological data and the interest point data in sequence.
As an optional implementation manner, the construction process of the graph convolution neural network specifically includes:
constructing a graph data structure according to the ground monitoring point data to obtain a spatial relationship topological graph and an adjacent weight matrix among the ground monitoring points; determining a degree matrix of the ground monitoring point according to the spatial relationship topological graph and the adjacency weight matrix; and constructing a graph convolution neural network according to the degree matrix and the node characteristic matrix.
As an optional implementation, the expression of the adjacency weight matrix is:
Figure BDA0003555784750000061
wherein, wijAs weights in the adjacency weight matrix, dijIs the Euclidean distance, σ, between the station numbered i and the station j2The method comprises the steps of setting a first threshold for controlling the distribution of an initial weight matrix, setting epsilon as a second threshold for controlling the sparsity of the initial weight matrix, setting i as an ith ground monitoring point and setting j as a jth ground monitoring point.
As an optional implementation, the loss function expression of the air quality prediction model of the space-time diagram neural network is as follows:
Figure BDA0003555784750000062
wherein the content of the first and second substances,
Figure BDA0003555784750000063
for the loss function, n is the total number of samples, yiThe value is monitored for the current site,
Figure BDA0003555784750000064
and i is the ith ground monitoring point, which is the predicted value of the current station.
As an alternative embodiment, the expression of the graph convolution neural network is:
Figure BDA0003555784750000065
wherein H(l+1)For the l +1 th layer data, sigma is a nonlinear model Relu activation function,
Figure BDA0003555784750000066
is a contiguous matrix with self-connection,
Figure BDA0003555784750000067
is a degree matrix, theta is a training parameter matrix, HlIs the l-th layer data.
Figure BDA0003555784750000068
Is W + InNamely, it is
Figure BDA0003555784750000069
Adding a self-connected adjacency matrix, W being an adjacency matrix, InIs a matrix of the units,
Figure BDA00035557847500000610
is composed of
Figure BDA00035557847500000611
Degree matrix formed by each node in the graph according to the adjacent weight matrix wijCalculating the site i, j, i belongs to n, j belongs to n, wij＝0,
Figure BDA00035557847500000612
wij≠0,
Figure BDA00035557847500000613
And (4) calculating.
The invention also provides a fine-grained air quality prediction system, which comprises:
the acquisition module is used for acquiring multi-source data of the atmosphere by utilizing a satellite; the multi-source data includes aerosol data, ground monitoring point data, meteorological data, and interest point data.
And the numerical value conversion and projection giving operation module is used for carrying out numerical value conversion and projection giving operation on the aerosol data to obtain the aerosol data with geographic information.
And the space-time matching module is used for performing space-time matching on the aerosol data with the geographic information, the ground monitoring point data, the meteorological data and the interest point data to obtain a node characteristic matrix.
The prediction module is used for inputting the node characteristic matrix into an air-map neural network air quality prediction model to obtain an air quality prediction value; the air quality prediction model of the space-time diagram neural network comprises a first gate-controlled cyclic neural network, a graph convolution neural network and a second gate-controlled cyclic neural network which are connected in sequence.
As an optional implementation, the fine-grained air quality prediction system further comprises:
and the preprocessing module is used for respectively carrying out abnormal value elimination, missing value filling and normalization processing on the aerosol data with the geographic information, the ground monitoring point data, the meteorological data and the interest point data in sequence.
As an optional implementation manner, the construction process of the graph convolution neural network specifically includes:
constructing a graph data structure according to the ground monitoring point data to obtain a spatial relationship topological graph and an adjacent weight matrix among the ground monitoring points; determining a degree matrix of the ground monitoring point according to the spatial relationship topological graph and the adjacency weight matrix; and constructing a graph convolution neural network according to the degree matrix and the node characteristic matrix.
As shown in fig. 3, the present invention further provides a specific processing flow of the fine-grained air quality prediction method in practical application, and the steps are as follows
Step 1, as shown in fig. 2, obtaining high spatio-temporal resolution sunflower number eight (himwari-8) aerosol aod (aerosol Optical depth) data with spatio-temporal resolution of 5km by 5km per hour, and performing numerical conversion and projection operation to obtain aerosol data with geographic information.
Step 2, preprocessing the data of the ground monitoring station, the meteorological data, the POI data and the aerosol data endowed with geographic information in the step 1, wherein the preprocessing operation mainly comprises the operations of abnormal value elimination, missing value filling and normalization processing, the normalization processing method is shown in a formula (1) to ensure that each type of data is in the same dimension, the spatial matching is carried out on the multi-source data through the longitude and latitude information of the ground monitoring station, the multi-source data is uniformly processed into hourly interval data to carry out time matching, the time-space matching of the multi-source data is realized, and a node characteristic matrix X is obtained;
Figure BDA0003555784750000081
in the formula, mean represents a variable mean value, std represents a variable standard deviation, and x is a node characteristic in a node characteristic matrix.
Step 3, according to the distance relationship between the ground monitoring points, using a formula (2) to construct a graph data structure of the ground monitoring stations to obtain a spatial relationship topological graph and an adjacent weight matrix W between the monitoring stations, wherein the construction method is shown in the formula (2);
Figure BDA0003555784750000082
in the formula dijIs the Euclidean distance, σ, between the station numbered i and the station j2And ε is the threshold that controls the initial weight matrix distribution and sparsity, with initial values of 10 and 0.5, respectively.
Step 4, building a space-time diagram neural network model, and calculating an adjacency matrix added with self-connection through an adjacency matrix W of a ground monitoring station topological graph
Figure BDA0003555784750000083
W is an adjacency matrix, InCalculating a degree matrix of a topological graph of a monitored site for an identity matrix
Figure BDA0003555784750000084
Is composed of
Figure BDA0003555784750000085
Degree matrix formed by each node in the graph according to the adjacent weight matrix wijCalculation, w between sites i, jij＝0，
Figure BDA0003555784750000086
wij≠0，
Figure BDA0003555784750000087
Calculating to obtain a graph convolution neural network (GCN) by combining with a node characteristic matrix X, wherein the graph convolution neural network construction principle is shown in a formula (3), and the influence of neighbor nodes on the current site is captured in a random walk mode, namely the graph convolution neural network (GCN) is adopted to learn and extract the spatial correlation of the topological structure diagram of the monitoring site; the method comprises the steps of adopting a gated recurrent neural network GRU to construct a time block for extracting time characteristics of historical data, inputting a time sequence with space characteristics into the GRU to extract time correlation of current station data, constructing an alternating space-time data extractor with a space extractor GCN in the middle and a time extractor GRN at two ends in a mode of combining a graph convolution neural network and the gated recurrent neural network, and continuously capturing station dataThe most basic space-time characteristics among the point data and the network model composition are shown in a figure (3), and the characteristics of the site data in two dimensions of time and space are extracted;
Figure BDA0003555784750000088
in the formula
Figure BDA0003555784750000089
In order to add a self-connected adjacency matrix,
Figure BDA00035557847500000810
is that
Figure BDA00035557847500000811
The degree matrix of (a) is obtained,
Figure BDA00035557847500000812
Inis an identity matrix, HlIf the data is the layer I data, if the data is the first layer, the characteristic matrix X is input, and theta is a training parameter matrix.
Step 5.1, dividing the feature matrix X obtained after the multi-source data in the step 2 are subjected to space-time matching into a training set, a verification set and a test set, inputting the training set into the space-time diagram neural network GCN-GRU model established in the step 4 for training to obtain a primary PM2.5 concentration prediction value, calculating loss through a loss function by combining a PM2.5 monitoring value of an air quality detection station, wherein the loss function adopts a Least Square Error (LSE), and the calculation formula is shown in (4), so that parameter adjustment based on the space-time diagram neural network air quality prediction model is realized;
Figure BDA0003555784750000091
in the formula: n denotes the total number of samples, yiIndicates the current station-monitored value of the current station,
Figure BDA0003555784750000092
predictive value representing current site。
Step 5.2, precision evaluation of the fine-grained air quality prediction method is carried out, wherein the precision evaluation is carried out on the fine-grained air quality prediction method based on the space-time diagram neural network by mainly using a root mean square error (RSME), a calculation formula shown in (5), an average absolute error (MAE), a calculation formula shown in (6) and an average absolute percentage error (MAPE), and a calculation formula shown in (7) and comparing with other prediction methods to verify the precision of the fine-grained air quality prediction method based on the space-time diagram neural network;
Figure BDA0003555784750000093
Figure BDA0003555784750000094
Figure BDA0003555784750000095
where n denotes the total number of samples, yiIndicates the current site monitor value and,
Figure BDA0003555784750000096
indicating the predicted value of the current site. And (5) after the model training is finished, predicting to obtain a PM2.5 predicted value with high space-time resolution, and finishing the model.
The method and the system provided by the invention integrate ground air quality monitoring station data, remote observation remote sensing satellite aerosol AOD data with high space-time resolution, meteorological data and other multi-source city data, so as to solve the problems that the extraction capability of two dimensional features of the multi-source space-time big data in time and space is insufficient, the air quality information with high space-time resolution of the city is lost and the like in the prior art.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principle and the embodiment of the present invention are explained by applying specific examples, and the above description of the embodiments is only used to help understanding the method and the core idea of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.

Claims (9)

1. A fine-grained air quality prediction method is characterized by comprising the following steps:
acquiring multi-source data of the atmosphere by using a satellite; the multi-source data comprises aerosol data, ground monitoring point data, meteorological data and interest point data;
carrying out numerical value conversion and projection operation on the aerosol data to obtain aerosol data with geographic information;
performing space-time matching on the aerosol data with the geographic information, the ground monitoring point data, the meteorological data and the interest point data to obtain a node characteristic matrix;
inputting the node characteristic matrix into an air-map neural network air quality prediction model to obtain an air quality prediction value; the air quality prediction model of the space-time diagram neural network comprises a first gate-controlled cyclic neural network, a graph convolution neural network and a second gate-controlled cyclic neural network which are connected in sequence.
2. The fine-grained air quality prediction method according to claim 1, wherein before the performing the space-time matching on the aerosol data with geographic information, the ground monitoring point data, the meteorological data and the interest point data to obtain a node feature matrix, the method further comprises:
and respectively carrying out abnormal value elimination, missing value filling and normalization processing on the aerosol data with geographic information, the ground monitoring point data, the meteorological data and the interest point data in sequence.
3. The fine-grained air quality prediction method according to claim 1, wherein the construction process of the graph convolution neural network specifically comprises:
constructing a graph data structure according to the ground monitoring point data to obtain a spatial relationship topological graph and an adjacent weight matrix among the ground monitoring points;
determining a degree matrix of the ground monitoring point according to the spatial relationship topological graph and the adjacency weight matrix;
and constructing a graph convolution neural network according to the degree matrix and the node characteristic matrix.
4. The fine-grained air quality prediction method according to claim 3, wherein the expression of the adjacency weight matrix is:
Figure FDA0003555784740000011
wherein wijAs weights in the adjacency weight matrix, dijIs the Euclidean distance, σ, between the station numbered i and the station j2The method comprises the steps of setting a first threshold for controlling the distribution of an initial weight matrix, setting epsilon as a second threshold for controlling the sparsity of the initial weight matrix, setting i as an ith ground monitoring point and setting j as a jth ground monitoring point.
5. The fine-grained air quality prediction method according to claim 1, wherein the loss function expression of the spatio-temporal neural network air quality prediction model is as follows:
Figure FDA0003555784740000021
wherein the content of the first and second substances,
Figure FDA0003555784740000022
for the loss function, n is the total number of samples, yiThe value is monitored for the current site,
Figure FDA0003555784740000026
and i is the predicted value of the current station, and the ith ground monitoring point.
6. The fine-grained air quality prediction method according to claim 1, wherein the expression of the graph convolutional neural network is:
Figure FDA0003555784740000023
wherein H(l+1)For the l +1 th layer data, sigma is a nonlinear model Relu activation function,
Figure FDA0003555784740000024
is a contiguous matrix with self-connection,
Figure FDA0003555784740000025
is a degree matrix, theta is a training parameter matrix, HlIs the l-th layer data.
7. A fine-grained air quality prediction system, comprising:
the acquisition module is used for acquiring multi-source data of the atmosphere by utilizing a satellite; the multi-source data comprises aerosol data, ground monitoring point data, meteorological data and interest point data;
the numerical value conversion and projection giving operation module is used for carrying out numerical value conversion and projection giving operation on the aerosol data to obtain aerosol data with geographic information;
the space-time matching module is used for performing space-time matching on the aerosol data with the geographic information, the ground monitoring point data, the meteorological data and the interest point data to obtain a node characteristic matrix;
the prediction module is used for inputting the node characteristic matrix into an air-map neural network air quality prediction model to obtain an air quality prediction value; the air quality prediction model of the space-time diagram neural network comprises a first gate-controlled cyclic neural network, a graph convolution neural network and a second gate-controlled cyclic neural network which are connected in sequence.
8. The fine-grained air quality prediction system of claim 7 further comprising:
and the preprocessing module is used for respectively carrying out abnormal value elimination, missing value filling and normalization processing on the aerosol data with the geographic information, the ground monitoring point data, the meteorological data and the interest point data in sequence.
9. The fine-grained air quality prediction system according to claim 7, wherein the construction process of the graph convolutional neural network specifically comprises:
constructing a graph data structure according to the ground monitoring point data to obtain a spatial relationship topological graph and an adjacent weight matrix among the ground monitoring points;
determining a degree matrix of the ground monitoring point according to the spatial relationship topological graph and the adjacency weight matrix;
and constructing a graph convolution neural network according to the degree matrix and the node characteristic matrix.
CN202210276051.1A 2022-03-21 2022-03-21 Fine-grained air quality prediction method and system Pending CN114638417A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210276051.1A CN114638417A (en) 2022-03-21 2022-03-21 Fine-grained air quality prediction method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210276051.1A CN114638417A (en) 2022-03-21 2022-03-21 Fine-grained air quality prediction method and system

Publications (1)

Publication Number Publication Date
CN114638417A true CN114638417A (en) 2022-06-17

Family

ID=81949651

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210276051.1A Pending CN114638417A (en) 2022-03-21 2022-03-21 Fine-grained air quality prediction method and system

Country Status (1)

Country Link
CN (1) CN114638417A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117592005A (en) * 2024-01-19 2024-02-23 中国科学院空天信息创新研究院 PM2.5 concentration satellite remote sensing estimation method, device, equipment and medium

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117592005A (en) * 2024-01-19 2024-02-23 中国科学院空天信息创新研究院 PM2.5 concentration satellite remote sensing estimation method, device, equipment and medium
CN117592005B (en) * 2024-01-19 2024-04-26 中国科学院空天信息创新研究院 PM2.5 concentration satellite remote sensing estimation method, device, equipment and medium

Similar Documents

Publication Publication Date Title
CN112465243B (en) Air quality forecasting method and system
CN105243435B (en) A kind of soil moisture content prediction technique based on deep learning cellular Automation Model
CN109063908B (en) Urban AQI prediction and spatial fine-grained AQI grade estimation method based on deep multi-task learning
CN110346517B (en) Smart city industrial atmosphere pollution visual early warning method and system
WO2020043027A1 (en) Method for optimizing boundaries of region when predicting atmospheric pollution
CN114240000A (en) Air quality prediction method based on space-time graph convolution network
CN110346518B (en) Traffic emission pollution visualization early warning method and system thereof
CN113505521B (en) Urban waterlogging rapid forecasting method based on neural network-numerical simulation
CN111369057A (en) Air quality prediction optimization method and system based on deep learning
CN112149887A (en) PM2.5 concentration prediction method based on data space-time characteristics
CN115730684A (en) Air quality detection system based on LSTM-CNN model
Zhang et al. Prediction of urban pm 2.5 concentration based on wavelet neural network
CN113516304A (en) Space-time joint prediction method and device for regional pollutants based on space-time graph network
CN112183625A (en) PM based on deep learning2.5High-precision time-space prediction method
CN112288156A (en) Air quality prediction method based on meta-learning and graph attention space-time neural network
CN115759488A (en) Carbon emission monitoring and early warning analysis system and method based on edge calculation
CN114186723A (en) Distributed photovoltaic power grid virtual prediction system based on space-time correlation
CN113987912A (en) Pollutant on-line monitoring system based on geographic information
CN115629160A (en) Air pollutant concentration prediction method and system based on space-time diagram
CN115237896A (en) Data preprocessing method and system for forecasting air quality based on deep learning
CN114638417A (en) Fine-grained air quality prediction method and system
CN116227362A (en) Municipal drainage pipe network state prediction method based on graph and deep learning
CN115544889A (en) Numerical mode precipitation deviation correction method based on deep learning
Lai et al. Sub-region division based short-term regional distributed PV power forecasting method considering spatio-temporal correlations
CN114882373A (en) Multi-feature fusion sandstorm prediction method based on deep neural network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination