Atmospheric NO2 spatiotemporal prediction algorithm combining hyperspectral satellite and artificial intelligence
By combining hyperspectral satellites with artificial intelligence, and utilizing meteorological and remote sensing data and neural network models, the problem of low spatial resolution of nitrogen dioxide concentration was solved, achieving higher-precision nitrogen dioxide concentration predictions.
Patent Information
- Application Number
- CN202211452402.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-11-17
- Publication Date
- 2025-09-16
- Estimated Expiration
- 2042-11-17
AI Technical Summary
The spatial resolution of nitrogen dioxide concentration in existing technologies is low, resulting in inaccurate monitoring.
Combining hyperspectral satellites with artificial intelligence, by acquiring meteorological monitoring data, hyperspectral satellite monitoring data and geographic information remote sensing data, atmospheric physical and chemical models and neural network models are used to predict nitrogen dioxide concentrations, and the nitrogen dioxide stratified concentration and column concentration data are integrated to improve spatial resolution.
It achieves higher spatial resolution and more accurate predictions of nitrogen dioxide concentrations, improving monitoring accuracy.
Smart Images
Figure CN115730718B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of pollutant monitoring, and in particular to an atmospheric NO2 spatiotemporal prediction algorithm, device, equipment and storage medium that combines hyperspectral satellites with artificial intelligence. Background Art
[0002] According to the National Ambient Air Quality Standard, nitrogen dioxide concentration is one of the six basic ambient air pollutant indicators, so it is necessary to monitor nitrogen dioxide in the ambient air. Existing calculations of nitrogen dioxide concentration are generally based on iterative calculations of atmospheric physical and chemical models.
[0003] However, the spatial resolution of nitrogen dioxide concentrations calculated by atmospheric physical and chemical models is low. Summary of the Invention
[0004] The main purpose of this application is to provide an atmospheric NO2 spatiotemporal prediction algorithm, device, equipment and medium that combines hyperspectral satellite and artificial intelligence, aiming to solve the technical problem of low spatial resolution of nitrogen dioxide concentration calculated in existing methods.
[0005] To achieve the above objectives, the present application provides an atmospheric NO2 spatiotemporal prediction algorithm that combines hyperspectral satellites with artificial intelligence. The method comprises:
[0006] Acquiring meteorological monitoring data, hyperspectral satellite monitoring data, and geographic information remote sensing data of the target area, wherein the geographic information remote sensing data includes surface building data, vegetation cover data, population data, and elevation data;
[0007] Obtaining nitrogen dioxide stratified concentration prediction data for the target area based on the meteorological monitoring data and a preset atmospheric physical and chemical model;
[0008] Inverting the nitrogen dioxide column concentration data for the target area based on the hyperspectral satellite monitoring data; wherein the spatial resolution of the nitrogen dioxide column concentration data is higher than the spatial resolution of the nitrogen dioxide layered concentration prediction data;
[0009] According to the geographic information remote sensing data, the nitrogen dioxide concentration stratification prediction data and the nitrogen dioxide column concentration data are fused to obtain the nitrogen dioxide concentration prediction data of the target area in the target area at the target time.
[0010] In a possible embodiment of the present application, obtaining the nitrogen dioxide concentration prediction data for a target area in the target region according to the nitrogen dioxide concentration stratification prediction data, the nitrogen dioxide column concentration data, and the geographic information remote sensing data includes:
[0011] Inputting the nitrogen dioxide concentration stratification prediction data, the nitrogen dioxide column concentration data, and the geographic information remote sensing data into a trained nitrogen dioxide concentration prediction neural network model to obtain four-dimensional nitrogen dioxide concentration prediction data of the target area output by the nitrogen dioxide concentration prediction neural network model; the four-dimensional nitrogen dioxide concentration prediction data includes four-dimensional space-time coordinate information and a nitrogen dioxide concentration prediction value corresponding to the four-dimensional space-time coordinate, and the four-dimensional space-time coordinate information includes time information, longitude information, latitude information, and elevation information;
[0012] The nitrogen dioxide concentration prediction data is extracted from the four-dimensional nitrogen dioxide concentration prediction data.
[0013] In a possible embodiment of the present application, the nitrogen dioxide concentration prediction neural network model includes:
[0014] A first feature extraction module is used to extract features from the nitrogen dioxide concentration stratification prediction data to obtain the spatiotemporal distribution characteristics of the nitrogen dioxide concentration stratification;
[0015] A second feature extraction module is used to perform feature extraction and feature fusion on the nitrogen dioxide column concentration data to obtain spatiotemporal distribution characteristics of the nitrogen dioxide column concentration;
[0016] A third feature extraction module is used to perform spatial feature extraction and dimensionality transformation on the geographic information remote sensing data to obtain spatiotemporal distribution features of the geographic information;
[0017] The fully connected layer is used to fuse the spatiotemporal distribution characteristics of the nitrogen dioxide concentration layer, the spatiotemporal distribution characteristics of the nitrogen dioxide column concentration, and the spatiotemporal distribution characteristics of the geographic information to obtain the four-dimensional nitrogen dioxide concentration prediction data.
[0018] In a possible embodiment of the present application, the second feature extraction module is further used to complete the nitrogen dioxide concentration stratified prediction data if the regional spatial coverage of the nitrogen dioxide column concentration data is greater than a first preset percentage and less than 1, to obtain the nitrogen dioxide column concentration data after data completion, and to perform feature extraction on the nitrogen dioxide column concentration data after data completion to obtain the spatiotemporal distribution characteristics of the nitrogen dioxide concentration stratification.
[0019] In a possible embodiment of the present application, before obtaining meteorological monitoring data, hyperspectral satellite monitoring data, and geographic information remote sensing data of the target area, wherein the geographic information remote sensing data includes surface building data, vegetation cover data, population data, and elevation data, the algorithm further includes:
[0020] Acquire training sample data, the training sample data including meteorological monitoring data, hyperspectral satellite monitoring data, and geographic information remote sensing data of the target area, as well as first hyperspectral satellite monitoring data of the target area at a preset time and real-time national control station monitoring data within a preset day;
[0021] Determining a training loss function for training the nitrogen dioxide concentration prediction neural network model;
[0022] The nitrogen dioxide concentration prediction neural network model is trained based on the training loss function and the training sample data until the value of the training loss function meets a preset condition, thereby obtaining a trained nitrogen dioxide concentration prediction neural network model.
[0023] In a possible embodiment of the present application, the training of the nitrogen dioxide concentration prediction neural network model based on the training loss function and the training sample data until the value of the training loss function meets a preset condition, thereby obtaining a trained nitrogen dioxide concentration prediction neural network model, includes:
[0024] Training is performed based on the meteorological monitoring data, the hyperspectral satellite monitoring data, and the geographic information remote sensing data to obtain first training result data for a first preset area within the target area at a preset time and second training result data for a second preset area within the target area at a preset day; wherein the altitude of the first preset area is higher than the altitude of the second preset area;
[0025] Verifying the first training result data according to the first hyperspectral satellite data to obtain spatial correlation data between the first hyperspectral satellite data and the first training result data;
[0026] Verifying the second training result data according to the real-time national control site monitoring data to obtain time correlation data and absolute value difference data between the second training result data and the real-time national control site monitoring data;
[0027] Using the spatial correlation data as a first loss function value, the temporal correlation data as a second loss function value, and the absolute value difference data as a third loss function value, and obtaining a final loss function value based on the first loss function value, the second loss function value, and the third loss function value;
[0028] Determine whether the final loss function value meets the preset conditions;
[0029] If not, the nitrogen dioxide concentration prediction neural network model is updated, and the training is returned to be performed based on the meteorological monitoring data, the hyperspectral satellite monitoring data and the geographic information remote sensing data to obtain the first training result data of the first preset area in the target area at the preset time and the second training result data of the second preset area in the target area on the preset day, until the value of the training loss function meets the preset conditions, thereby obtaining a trained nitrogen dioxide concentration prediction neural network model.
[0030] In a possible embodiment of the present application, obtaining nitrogen dioxide column concentration data of the target area based on the hyperspectral satellite monitoring data includes:
[0031] Filtering valid monitoring data with a cloud cover ratio less than or equal to a preset threshold from the hyperspectral satellite monitoring data;
[0032] Based on the effective monitoring data, the nitrogen dioxide column concentration data of the target area is inverted.
[0033] In a second aspect, the present application also provides a spatiotemporal prediction device for atmospheric NO2 that combines a hyperspectral satellite with artificial intelligence, comprising:
[0034] A data acquisition module is used to acquire meteorological monitoring data, hyperspectral satellite monitoring data, and geographic information remote sensing data of the target area, wherein the geographic information remote sensing data includes surface building data, vegetation cover data, population data, and elevation data;
[0035] a stratified concentration prediction module, configured to obtain stratified nitrogen dioxide concentration prediction data for the target area based on the meteorological monitoring data and a preset atmospheric physical and chemical model;
[0036] a column concentration prediction module, configured to invert nitrogen dioxide column concentration data of the target area based on the hyperspectral satellite monitoring data; wherein the spatial resolution of the nitrogen dioxide column concentration data is higher than the spatial resolution of the nitrogen dioxide layered concentration prediction data;
[0037] A data fusion module is used to fuse the nitrogen dioxide concentration stratification prediction data and the nitrogen dioxide column concentration data based on the geographic information remote sensing data to obtain the nitrogen dioxide concentration prediction data of the target area in the target area at the target time.
[0038] In a third aspect, the present application also provides an atmospheric NO2 spatiotemporal prediction device combining hyperspectral satellites and artificial intelligence, comprising a processor, a memory, and an atmospheric NO2 spatiotemporal prediction program combining hyperspectral satellites and artificial intelligence stored in the memory. When the atmospheric NO2 spatiotemporal prediction program combining hyperspectral satellites and artificial intelligence is run by the processor, the steps of the atmospheric NO2 spatiotemporal prediction algorithm combining hyperspectral satellites and artificial intelligence as described above are implemented.
[0039] In a fourth aspect, the present application also provides a computer-readable storage medium, on which is stored a spatiotemporal prediction program for atmospheric NO2 that combines hyperspectral satellites and artificial intelligence. When the spatiotemporal prediction program for atmospheric NO2 that combines hyperspectral satellites and artificial intelligence is executed by a processor, the spatiotemporal prediction algorithm for atmospheric NO2 that combines hyperspectral satellites and artificial intelligence as described above is implemented.
[0040] The embodiments of the present application propose an atmospheric NO2 spatiotemporal prediction algorithm, device, equipment and storage medium that combine hyperspectral satellite and artificial intelligence. The method includes obtaining meteorological monitoring data, hyperspectral satellite monitoring data and geographic information remote sensing data of a target area; obtaining nitrogen dioxide stratified concentration prediction data of the target area based on the meteorological monitoring data and a preset atmospheric physical and chemical model; inverting nitrogen dioxide column concentration data of the target area based on the hyperspectral satellite monitoring data; and fusing the nitrogen dioxide concentration stratification prediction data and the nitrogen dioxide column concentration data based on the geographic information remote sensing data to obtain nitrogen dioxide concentration prediction data of a target area in the target area at a target time.
[0041] Therefore, compared with the nitrogen dioxide concentration data predicted by the existing atmospheric physical and chemical models, when predicting nitrogen dioxide concentration, the present application uses geographic information remote sensing data to integrate the nitrogen dioxide stratified concentration prediction data obtained by the atmospheric physical and chemical model and the nitrogen dioxide column concentration data obtained by inverting the hyperspectral satellite monitoring data, thereby effectively utilizing the hyperspectral satellite monitoring data with higher spatial resolution to obtain higher spatial resolution in the target area, that is, more accurate nitrogen dioxide concentration prediction data. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 This is a schematic diagram of the structure of an atmospheric NO2 spatiotemporal prediction device that combines a hyperspectral satellite and artificial intelligence in the hardware operating environment involved in the embodiment of the present application;
[0043] Figure 2 This is a flow chart of the first embodiment of the atmospheric NO2 spatiotemporal prediction algorithm combining hyperspectral satellite and artificial intelligence in this application;
[0044] Figure 3This is a flow chart of the second embodiment of the atmospheric NO2 spatiotemporal prediction algorithm combining hyperspectral satellite and artificial intelligence in this application;
[0045] Figure 4 This is a schematic diagram of the modules of the neural network model for predicting nitrogen dioxide concentration in this application;
[0046] Figure 5 This is a schematic diagram of the first feature extraction module in the nitrogen dioxide concentration prediction neural network model of this application;
[0047] Figure 6 This is a schematic diagram of the second feature extraction module in the nitrogen dioxide concentration prediction neural network model of this application;
[0048] Figure 7 This is a schematic diagram of the third feature extraction module in the nitrogen dioxide concentration prediction neural network model of this application;
[0049] Figure 8 This is a flowchart of the third embodiment of the atmospheric NO2 spatiotemporal prediction algorithm combining hyperspectral satellite and artificial intelligence in this application;
[0050] Figure 9 This is a module diagram of the first embodiment of the atmospheric NO2 spatiotemporal prediction device combining hyperspectral satellite and artificial intelligence in this application.
[0051] The realization of the objectives, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0052] It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.
[0053] In the related prior art, nitrogen dioxide concentration prediction data is still obtained through iterative calculation based on atmospheric physical and chemical models. However, the spatial resolution of nitrogen dioxide concentration calculated by atmospheric physical and chemical models is low.
[0054] To this end, the present application provides an atmospheric NO2 spatiotemporal prediction algorithm that combines hyperspectral satellites with artificial intelligence, which assimilates and fuses the nitrogen dioxide stratified concentration prediction data obtained by iterative calculation based on atmospheric physical and chemical models and the nitrogen dioxide column concentration data obtained by inverting hyperspectral satellite monitoring data to obtain nitrogen dioxide concentration prediction data with higher spatial resolution and higher accuracy.
[0055] The inventive concept of the present application is further described below with reference to some specific embodiments.
[0056] Reference Figure 1 , Figure 1This is a schematic diagram of the structure of an atmospheric NO2 spatiotemporal prediction device that combines a hyperspectral satellite with artificial intelligence in the hardware operating environment involved in the embodiment of the present application.
[0057] like Figure 1 As shown, the atmospheric NO2 spatiotemporal prediction device that combines hyperspectral satellite and artificial intelligence may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard). Optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a wireless fidelity (WI-FI) interface). The memory 1005 may be a high-speed random access memory (RAM) memory or a stable non-volatile memory (NVM), such as a disk storage. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.
[0058] Those skilled in the art will understand that Figure 1 The structure shown in the figure does not constitute a limitation on the atmospheric NO2 spatiotemporal prediction device that combines hyperspectral satellites with artificial intelligence, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.
[0059] like Figure 1 As shown, the memory 1005 as a storage medium may include an operating system, a data storage module, a network communication module, a user interface module, and an atmospheric NO2 spatiotemporal prediction program that combines hyperspectral satellites with artificial intelligence.
[0060] exist Figure 1In the atmospheric NO2 spatiotemporal prediction device combining hyperspectral satellite and artificial intelligence shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the atmospheric NO2 spatiotemporal prediction device combining hyperspectral satellite and artificial intelligence in this application can be set in the atmospheric NO2 spatiotemporal prediction device combining hyperspectral satellite and artificial intelligence. The atmospheric NO2 spatiotemporal prediction device combining hyperspectral satellite and artificial intelligence calls the atmospheric NO2 spatiotemporal prediction program combining hyperspectral satellite and artificial intelligence stored in the memory 1005 through the processor 1001, and executes the atmospheric NO2 spatiotemporal prediction algorithm combining hyperspectral satellite and artificial intelligence provided in the embodiment of this application.
[0061] Based on the above hardware structure but not limited to the above hardware structure, this application provides a first embodiment of an atmospheric NO2 spatiotemporal prediction algorithm combining hyperspectral satellite and artificial intelligence. Figure 2 , Figure 2 The figure shows a flow chart of the first embodiment of the atmospheric NO2 spatiotemporal prediction algorithm combining hyperspectral satellite and artificial intelligence.
[0062] It should be noted that although a logical order is shown in the flowcharts, in some cases, the steps shown or described may be performed in an order different from that shown or described here.
[0063] In this embodiment, the atmospheric NO2 spatiotemporal prediction algorithm combining hyperspectral satellite and artificial intelligence includes:
[0064] Step S100: Acquire meteorological monitoring data, hyperspectral satellite monitoring data, and geographic information remote sensing data of the target area. The geographic information remote sensing data includes surface building data, vegetation cover data, population data, and elevation data.
[0065] In this embodiment, the executor of the hyperspectral satellite and artificial intelligence atmospheric NO2 spatiotemporal prediction algorithm is a hyperspectral satellite and artificial intelligence atmospheric NO2 spatiotemporal prediction device. It is understood that the hyperspectral satellite and artificial intelligence atmospheric NO2 spatiotemporal prediction device can be a terminal device such as a computer, which can obtain meteorological monitoring data, hyperspectral satellite monitoring data, and geographic information remote sensing data via wired or wireless means.
[0066] Among them, meteorological monitoring data is data provided by the global meteorological monitoring platform, including but not limited to temperature, humidity, wind speed and pressure.
[0067] Hyperspectral satellite monitoring data is provided by hyperspectral satellites, which can be domestically produced and combine sub-meter spatial and temporal resolution. Specifically, hyperspectral satellite monitoring data includes, but is not limited to, relative humidity (Rh), temperature (Temperature), pressure (Pressure), boundary layer height (Hpbl), wind speed-longitude (U) and wind speed-latitude (V) data.
[0068] Geographic information remote sensing data includes, but is not limited to, surface building data, vegetation cover data, population data, and elevation data. Specifically, remote sensing data is obtained by remote sensing satellites detecting the reflection of electromagnetic waves from objects on the Earth's surface and the electromagnetic waves they emit, thereby extracting information about the objects, enabling remote object identification, and converting these electromagnetic waves into visual images. There is a wide variety of remote sensing data, but each type of data has different levels, and different levels of data are processed differently. In this embodiment, MODIS MOD17 surface data is used as an example for detailed explanation. MODIS MOD17 is a standard land data product of MODIS data at Level 4. Specifically, in this embodiment, surface building data includes, but is not limited to, point of interest (POI) data and traffic network data. The spatial resolution of surface building data can be 5 km x 5 km. Vegetation cover data includes, but is not limited to, the enhanced vegetation index (EVI), the normalized vegetation index (NDVI), and land cover type data. The spatial resolution of vegetation cover data can be 1 km x 1 km. Population data includes, but is not limited to, GDP data, total population data, and annual population density data. Elevation data includes, but is not limited to, DEM elevation data and altitude data. The spatial resolution of population and elevation data can be 1 km × 1 km.
[0069] Step S200: Obtaining nitrogen dioxide stratified concentration prediction data for the target area based on the meteorological monitoring data and a preset atmospheric physical and chemical model.
[0070] Specifically, after acquiring meteorological monitoring data, the atmospheric NO2 spatiotemporal prediction device, which combines hyperspectral satellites with artificial intelligence, can generate low-spatial-resolution stratified NO2 concentration forecasts within one hour using a preset atmospheric physicochemical model, historical pollution inventories, and meteorological monitoring data from the target area. The spatial resolution of the stratified NO2 concentration forecasts can be as low as 20 km x 20 km.
[0071] Understandably, the pollution inventory is based on surveys of target areas and is therefore historical in nature and may not represent current realities. Therefore, this embodiment will subsequently utilize hyperspectral satellite monitoring data for combined processing to obtain more accurate nitrogen dioxide concentration prediction data.
[0072] It is worth mentioning that, in this embodiment, the nitrogen dioxide stratified concentration prediction data is stratified according to the atmospheric pressure value in the direction away from the ground, for example, it can be divided into 44 layers.
[0073] In addition, in this embodiment, the preset atmospheric physical and chemical model can be the WRF-CHEM atmospheric physical and chemical model. In the WRF-CHEM model, chemical and meteorological processes use the same horizontal and vertical coordinate systems and the same physical parameterization scheme, without temporal interpolation, and can consider the feedback effects of chemical processes on meteorological processes.
[0074] Step S300: inverting the nitrogen dioxide column concentration data of the target area based on the hyperspectral satellite monitoring data; wherein the spatial resolution of the nitrogen dioxide column concentration data is higher than the spatial resolution of the nitrogen dioxide stratified concentration prediction data.
[0075] Specifically, in this step, the atmospheric NO2 spatiotemporal prediction device combining hyperspectral satellite and artificial intelligence uses hyperspectral satellite monitoring data to invert and obtain the nitrogen dioxide column concentration. Among them, the spatial resolution of the nitrogen dioxide column concentration data can be 5km×5km, so that the spatial resolution of the nitrogen dioxide column concentration data is higher than the spatial resolution of the nitrogen dioxide layered concentration prediction data. It can be understood that those skilled in the art know how to implement the specific steps of inverting hyperspectral satellite monitoring data to obtain the nitrogen dioxide column concentration data of the target area, which will not be repeated here. It is worth mentioning that the spatial resolution of the nitrogen dioxide column concentration data is 5km×5km, that is, during the processing, one pixel represents an area of 5km×5km on the ground.
[0076] As an embodiment, step S300 specifically includes:
[0077] Step S301: Filter out valid monitoring data with a cloud cover ratio less than or equal to a preset threshold from the hyperspectral satellite monitoring data.
[0078] Step S302: invert and obtain nitrogen dioxide column concentration data of the target area based on the effective monitoring data.
[0079] Specifically, due to the presence of clouds, and sometimes the clouds will completely cover the area represented by a pixel. At this time, the hyperspectral satellite cannot accurately monitor the relative humidity Rh data, temperature data, pressure data, boundary layer height Hpbl data, U wind speed-longitude data and V wind speed-latitude data of the area corresponding to the pixel. Therefore, for the hyperspectral satellite monitoring data, it is also necessary to first eliminate the invalid monitoring data whose cloud cover ratio is greater than the preset threshold, so as to filter out the valid monitoring data whose cloud cover ratio is less than or equal to the preset threshold, and invert the nitrogen dioxide column concentration data of the target area based on the valid monitoring data. Among them, the preset threshold can be 50%. Of course, the preset threshold can also be adaptively changed according to the accuracy requirements.
[0080] In this embodiment, filtering out valid monitoring data with a cloud cover ratio less than or equal to a preset threshold can make the final nitrogen dioxide concentration prediction data more accurate.
[0081] Step S400: Based on the geographic information remote sensing data, the nitrogen dioxide concentration stratification prediction data and the nitrogen dioxide column concentration data are fused to obtain the nitrogen dioxide concentration prediction data of the target area in the target region at the target time.
[0082] Specifically, after calculating the nitrogen dioxide concentration stratification prediction data and the nitrogen dioxide column concentration data, the atmospheric NO2 spatiotemporal prediction device that combines hyperspectral satellites with artificial intelligence can assimilate and fuse the nitrogen dioxide column concentration data with higher spatial resolution and better timeliness with the nitrogen dioxide concentration stratification prediction data based on the spatial characteristics provided by the geographic information remote sensing data, thereby obtaining the nitrogen dioxide concentration prediction data for each area in the target area at each time within 24 hours of the day, that is, obtaining the nitrogen dioxide concentration prediction data for the target area in the target area at the target time. It is worth mentioning that in this embodiment, the target area is a spatial area defined by longitude, latitude and altitude.
[0083] Compared with the nitrogen dioxide concentration data predicted by the existing atmospheric physical and chemical models, this application, when predicting nitrogen dioxide concentration, integrates the nitrogen dioxide stratified concentration prediction data obtained by the atmospheric physical and chemical models and the nitrogen dioxide column concentration data obtained by inverting the hyperspectral satellite monitoring data based on geographic information remote sensing data, thereby effectively utilizing the hyperspectral satellite monitoring data with higher spatial resolution to obtain higher spatial resolution in the target area, that is, more accurate nitrogen dioxide concentration prediction data.
[0084] Based on the above embodiments, the second embodiment of the atmospheric NO2 spatiotemporal prediction algorithm combining hyperspectral satellite and artificial intelligence is proposed. Figure 3 , Figure 3 This is a flow chart of the second embodiment of the atmospheric NO2 spatiotemporal prediction algorithm combining hyperspectral satellite and artificial intelligence in this application.
[0085] In this embodiment, step S400 specifically includes:
[0086] Step S401: Input the nitrogen dioxide concentration stratified prediction data, the nitrogen dioxide column concentration data, and the geographic information remote sensing data into a trained nitrogen dioxide concentration prediction neural network model to obtain four-dimensional nitrogen dioxide concentration prediction data of the target area output by the nitrogen dioxide concentration prediction neural network model; the four-dimensional nitrogen dioxide concentration prediction data includes four-dimensional space-time coordinate information and a nitrogen dioxide concentration prediction value corresponding to the four-dimensional space-time coordinate, and the four-dimensional space-time coordinate information includes time information, longitude information, latitude information, and elevation information.
[0087] Step S402: extracting the nitrogen dioxide concentration prediction data from the four-dimensional nitrogen dioxide concentration prediction data.
[0088] In this embodiment, a convolutional neural network model is constructed to predict the nitrogen dioxide concentration, thereby improving the calculation speed.
[0089] Specifically, the four-dimensional nitrogen dioxide concentration prediction data can be expressed as [N, T, E / W, N / S, H]. N is the predicted nitrogen dioxide concentration, T represents the target time, E / W represents the longitude of the target area, N / S represents the latitude of the target area, and H represents the altitude of the target area. The target area can thus be determined based on longitude, latitude, and altitude.
[0090] For details, see Figure 4 , the nitrogen dioxide concentration prediction neural network model includes:
[0091] A first feature extraction module is used to extract features from the nitrogen dioxide concentration stratification prediction data to obtain the spatiotemporal distribution characteristics of the nitrogen dioxide concentration stratification;
[0092] A second feature extraction module is used to perform feature extraction and feature fusion on the nitrogen dioxide column concentration data to obtain spatiotemporal distribution characteristics of the nitrogen dioxide column concentration;
[0093] A third feature extraction module is used to perform spatial feature extraction and dimensionality transformation on the geographic information remote sensing data to obtain spatiotemporal distribution features of the geographic information;
[0094] The fully connected layer is used to fuse the spatiotemporal distribution characteristics of the nitrogen dioxide concentration layer, the spatiotemporal distribution characteristics of the nitrogen dioxide column concentration, and the spatiotemporal distribution characteristics of the geographic information to obtain the four-dimensional nitrogen dioxide concentration prediction data.
[0095] It is understandable that the nitrogen dioxide concentration prediction neural network model further includes an input layer. The first feature extraction module, the second feature extraction module, and the third feature extraction module are all connected to the input layer.
[0096] See Figure 5 The first feature extraction module includes an UpSampling3D upsampling layer, a Conv3D convolution layer, and an Inception Block, which are connected in sequence. The UpSampling3D upsampling layer samples the stratified prediction data of nitrogen dioxide concentration with lower spatial resolution into upsampled data with higher spatial resolution. The Conv3D convolution layer is used to extract features from the upsampled data to obtain the prediction data features. The Inception Block mainly uses multiple convolution operations with different convolution kernel sizes to capture prediction data features at more scales, especially time and spatial scales, to obtain the spatiotemporal distribution characteristics of nitrogen dioxide concentration stratification.
[0097] See Figure 6 , the second feature extraction module includes a first Inception Block, a first full-leveling layer, and a second Inception Block connected in sequence. Among them, the first Inception Block can include multiple blocks, so as to process the nitrogen dioxide column concentration data at multiple times. It can be understood that, corresponding to the target area, the hyperspectral satellite has a transit time period relative to the target area, and the hyperspectral satellite monitoring data can only be obtained during the transit time period, that is, the nitrogen dioxide column concentration data is related to the transit time period, and thus has multiple. The first Inception Block is used to capture the prediction data features of the nitrogen dioxide column concentration data at more scales, especially the time scale and spatial scale, so as to obtain the spatiotemporal distribution characteristics of the nitrogen dioxide column concentration.
[0098] As an embodiment, the first Inception Block and the second Inception Block each include an int Layer input layer, a branch one consisting of a 1×1 2D convolution layer, a branch two consisting of a 1×1 2D convolution layer and a 3×3 2D convolution layer, and a branch three consisting of a 3×3 2D convolution layer. Branch one and branch two are both connected to a 3×3 2D convolution layer, wherein the processing results of branch one and branch two are superimposed and input to the 3×3 2D convolution layer. The 3×3 2D convolution layer and branch three are connected to an out Layer output layer. wherein the processing results of branch three and the 3×3 2D convolution layer are superimposed and input to the out Layer output layer.
[0099] In addition, see Figure 7The third feature extraction module includes a building geographic information feature extraction submodule, a vegetation geographic information feature extraction submodule, a second fully connected layer and a matrix dimension transformation layer. The building geographic information feature extraction submodule, the vegetation geographic information feature extraction submodule and the matrix dimension transformation layer are all connected to the second fully connected layer.
[0100] Among them, the building geographic information feature extraction submodule includes a first Conv2D convolution layer, a third fully connected layer, a first GI Block geographic information feature extraction block and a mean pooling layer connected in sequence. The input layer is used to input POI point of interest data, Traffic Network road network data and GDP gross product data, Total population data and annual population density data, etc. into the building geographic information feature extraction submodule, so that the building geographic information feature extraction submodule performs feature extraction, feature fusion, geographic information feature extraction and mean pooling on the above data to obtain building-type geographic distribution features. It is worth mentioning that the first Conv2D convolution layer can include multiple layers to perform feature extraction on each type of data in the POI point of interest data, Traffic Network road network data and GDP gross product data, Total population data and annual population density data, etc.
[0101] The vegetation geographic information feature extraction submodule includes a second Conv2D convolution layer, a fourth fully connected layer and a second geographic information feature extraction block connected in sequence. The input layer inputs the EVI enhanced vegetation index, NDVI normalized vegetation index and LandCover Type surface cover type data into the vegetation geographic information feature extraction submodule. The vegetation geographic information feature extraction submodule is used to perform feature extraction, feature fusion and geographic information feature extraction on the EVI enhanced vegetation index, NDVI normalized vegetation index and Land Cover Type surface cover type data in sequence, thereby obtaining the vegetation class geographic distribution characteristics. It is worth mentioning that the first Conv2D convolution layer can include multiple layers to perform feature extraction on each type of data such as the EVI enhanced vegetation index, NDVI normalized vegetation index and Land Cover Type surface cover type data.
[0102] Then, the second fully connected layer fuses the geographic distribution features of buildings and vegetation to obtain the spatial distribution features of geographic information. The Reshape matrix dimension transformation layer transforms the spatial distribution features of geographic information to obtain the spatiotemporal distribution features of geographic information.
[0103] As an embodiment, the first geographic information feature extraction block and the second geographic information feature extraction block are both constructed as follows: each includes an int Layer input layer, a branch 1 consisting of a 1×1×1 3D convolution layer, a branch 2 consisting of two 3×3×3 3D convolution layers connected together, and then a 3×3×3 3D convolution layer. The processing results of branches 1 and 2 are superimposed and input to the 3×3×3 3D convolution layer. The 3×3×3 3D convolution layer is connected to an out Layer output layer.
[0104] Finally, the fully connected layer fuses the spatiotemporal distribution characteristics of nitrogen dioxide concentration layers, the spatiotemporal distribution characteristics of nitrogen dioxide column concentrations, and the spatiotemporal distribution characteristics of geographic information to obtain the four-dimensional nitrogen dioxide concentration prediction data.
[0105] As an embodiment, the fully connected layer may also be connected to an Inception Block to capture more scales through multiple convolution operations with different convolution kernel sizes, thereby obtaining more accurate four-dimensional nitrogen dioxide concentration prediction data.
[0106] In one embodiment, the second feature extraction module is further used to complete the nitrogen dioxide concentration stratified prediction data if the regional spatial coverage of the nitrogen dioxide column concentration data is greater than or equal to a first preset percentage and less than 1, to obtain the nitrogen dioxide column concentration data after data completion, and to perform feature extraction on the nitrogen dioxide column concentration data after data completion to obtain the spatiotemporal distribution characteristics of the nitrogen dioxide concentration stratification.
[0107] Specifically, on the image, for a certain target area A in any target area, it is composed of multiple pixels. As mentioned above, when calculating the nitrogen dioxide column concentration, pixels with a cloud cover ratio greater than or equal to the preset threshold will be eliminated, that is, such pixels will not be inverted to obtain nitrogen dioxide column concentration data. Therefore, for target area A, when the nitrogen dioxide column concentration is needed, some of the pixels therein may not have corresponding data. That is, the data does not cover these pixels, and only the pixels for which the nitrogen dioxide column concentration is inverted are covered. Therefore, at this time, the regional spatial coverage rate of target area A is not 1, but less than 1.
[0108] Thus, when the regional spatial coverage of the nitrogen dioxide column concentration data is greater than the first preset percentage and less than 1, the second feature extraction module performs data completion on the nitrogen dioxide concentration stratified prediction data through bilinear interpolation to obtain the data-completed nitrogen dioxide column concentration data. Of course, it is understandable that for data with regional spatial coverage less than or equal to the first preset percentage, we can eliminate data to improve the accuracy of data processing. Yes, the first preset percentage can be 70%. Of course, the specific value of the first preset percentage can also be adjusted according to actual needs.
[0109] Based on the above embodiments, the third embodiment of the atmospheric NO2 spatiotemporal prediction algorithm combining hyperspectral satellite and artificial intelligence is proposed. Figure 8 , Figure 8 This is a flow chart of the third embodiment of the atmospheric NO2 spatiotemporal prediction algorithm combining hyperspectral satellite and artificial intelligence in this application.
[0110] In this embodiment, the nitrogen dioxide concentration prediction neural network model is trained in the following manner:
[0111] Step S10: Acquire training sample data, wherein the training sample data includes meteorological monitoring data, hyperspectral satellite monitoring data, and geographic information remote sensing data of the target area, as well as first hyperspectral satellite monitoring data of the target area at a preset time and real-time national control station monitoring data within a preset day.
[0112] Specifically, in this embodiment, the training sample data includes the first hyperspectral satellite monitoring data of the target area at a preset time and the real-time national control station monitoring data for 24 hours within a preset day. That is, when the nitrogen dioxide concentration prediction neural network model processes the meteorological monitoring data, hyperspectral satellite monitoring data and geographic information remote sensing data of the target area and outputs the four-dimensional nitrogen dioxide concentration prediction data of the target area, it is verified by the first hyperspectral satellite monitoring data at the preset time and the real-time national control station monitoring data for 24 hours within the preset day, so that the result output by the trained nitrogen dioxide concentration prediction neural network model is close to the data obtained by hyperspectral satellite monitoring and the monitoring data of the national control station, thereby improving the training effect of the nitrogen dioxide concentration prediction neural network model.
[0113] In one example, the preset time may be noon. It is understandable that, generally speaking, the cloud cover at this time is relatively small, so the monitoring data is more accurate.
[0114] Step S20: Determine a training loss function for training the nitrogen dioxide concentration prediction neural network model.
[0115] Step S30: training the nitrogen dioxide concentration prediction neural network model based on the training loss function and the training sample data until the value of the training loss function meets a preset condition, thereby obtaining a trained nitrogen dioxide concentration prediction neural network model.
[0116] After the training samples and loss function are determined, the nitrogen dioxide concentration prediction neural network model can be trained. In this embodiment, step S30 specifically includes:
[0117] Step S31: Training is performed based on the meteorological monitoring data, the hyperspectral satellite monitoring data, and the geographic information remote sensing data to obtain first training result data for a first preset area within the target area at the preset time and second training result data for a second preset area within the target area on the preset day; wherein the altitude of the first preset area is higher than the altitude of the second preset area.
[0118] Step S31: Verify the first training result data according to the first hyperspectral satellite data to obtain spatial correlation data between the first hyperspectral satellite data and the first training result data.
[0119] Step S32: verify the second training result data according to the real-time national control site monitoring data to obtain time correlation data and absolute value difference data between the second training result data and the real-time national control site monitoring data.
[0120] Step S33: using the spatial correlation data as a first loss function value, the temporal correlation data as a second loss function value, and the absolute value difference data as a third loss function value, and obtaining a final loss function value based on the first loss function value, the second loss function value, and the third loss function value;
[0121] Step S34: Determine whether the final loss function value meets the preset conditions.
[0122] In this example, the loss function is described by Formula 1, which is:
[0123] Loss=α·Loss1+β·Loss2+λ·Loss3;
[0124] Loss1 is the first loss function value, which can be determined by Formula 2:
[0125]
[0126] Loss2 is the second loss function value, which can be determined by Formula 3:
[0127]
[0128] Loss3 is the third loss function value, which can be determined by Formula 3:
[0129]
[0130] Here, α, β, and λ are all constants. In one example, α=0.5, β=0.5, and λ=1.
[0131] For ease of understanding, an example is shown below. The first preset area is the middle layer of the target area, and the preset time is noon, i.e., 12:00 p.m. The second preset area can be the ground layer area near the center of the target area. Therefore, the first training result data is the prediction data corresponding to the middle layer area of the target area at 12:00 p.m. extracted from the four-dimensional nitrogen dioxide concentration prediction data output by the nitrogen dioxide concentration prediction neural network model for the target area. The second training result data is the prediction data corresponding to the ground layer area near the center of the target area on the preset day output by the nitrogen dioxide concentration prediction neural network model for the target area.
[0132] It is understandable that the altitude of the first preset area is different from the altitude of the second preset area to make the final training result more accurate. Of course, the first preset area and the second preset area can also be adaptively selected based on the terrain, building distribution or population distribution, which is not limited here.
[0133] For the first training result data, the verification process is as follows: the nitrogen dioxide column concentration data obtained by inverting the first hyperspectral satellite data can be used to constrain its spatial distribution, and the first training result data of the four-dimensional nitrogen dioxide concentration prediction data structure is constructed as a vector X Psatellite-1 , and construct the nitrogen dioxide column concentration data into a vector X Tsatellite-1 , calculate the spatial angle between the two vectors, that is, substitute it into formula 2, and obtain the spatial correlation data of the first hyperspectral satellite data and the first training result data, which is also the value of the first loss function.
[0134] For the second training result data, the verification process is as follows: the real-time national control station monitoring data, that is, the 24-hour measured data, can be used to constrain the time distribution and spatial distribution of the data, and the second training result data of the four-dimensional nitrogen dioxide concentration prediction data structure can be constructed as a vector X Psate-2 The real-time national control site monitoring data at the corresponding time is constructed as vector X Tsate-2 , calculate the spatial angle between the two vectors, that is, substitute it into formula 3, and obtain the time correlation data of the second training result data and the real-time national control site monitoring data, that is, the value of the second loss function. And also construct the second training result data at any time j as vector X Psate-j,2The real-time national control site monitoring data corresponding to time j is constructed as vector X Tsate-j，2 , substituting into formula 4, we can obtain the absolute value difference data between the second training result data and the real-time national control site monitoring data, which is also the value of the third loss function.
[0135] After obtaining the values of the first, second, and third loss functions, these three values are substituted into Formula 1 for weighted superposition to obtain the value of the final loss function. The final loss function value is then compared with the pre-set conditions set before training. If they meet, training ends. If not, the following step S35 is executed.
[0136] Step S35: If not, update the nitrogen dioxide concentration prediction neural network model, and return to execute the training based on the meteorological monitoring data, the hyperspectral satellite monitoring data and the geographic information remote sensing data to obtain first training result data of a first preset area in the target area and second training result data of a second preset area in the target area, until the value of the training loss function meets the preset condition, and obtain a trained nitrogen dioxide concentration prediction neural network model.
[0137] Specifically, during the training process, the SGD optimizer can be used to perform iterative model training until the final result is obtained, which is the trained nitrogen dioxide concentration prediction neural network model.
[0138] In this embodiment, the loss function is correlated with the spatial correlation, temporal correlation and absolute value correlation between the training data and the verification data, so that the four-dimensional nitrogen dioxide concentration prediction data output by the trained nitrogen dioxide concentration prediction neural network model has a high accuracy in both spatial and temporal dimensions.
[0139] Based on the same inventive concept, see Figure 9 , a first embodiment of the atmospheric NO2 spatiotemporal prediction device combining hyperspectral satellite and artificial intelligence is proposed in this application, comprising:
[0140] A data acquisition module is used to acquire meteorological monitoring data, hyperspectral satellite monitoring data, and geographic information remote sensing data of the target area, wherein the geographic information remote sensing data includes surface building data, vegetation cover data, population data, and elevation data;
[0141] a stratified concentration prediction module, configured to obtain stratified nitrogen dioxide concentration prediction data for the target area based on the meteorological monitoring data and a preset atmospheric physical and chemical model;
[0142] a column concentration prediction module, configured to invert the nitrogen dioxide column concentration data of the target area based on the hyperspectral satellite monitoring data; wherein the spatial resolution of the nitrogen dioxide column concentration data is higher than the spatial resolution of the nitrogen dioxide layered concentration prediction data;
[0143] A data fusion module is used to fuse the nitrogen dioxide concentration stratification prediction data and the nitrogen dioxide column concentration data based on the geographic information remote sensing data to obtain the nitrogen dioxide concentration prediction data of the target area in the target area at the target time.
[0144] It should be noted that the various implementations of the atmospheric NO2 spatiotemporal prediction device combining hyperspectral satellites and artificial intelligence in this embodiment and the technical effects achieved can refer to the various implementations of the atmospheric NO2 spatiotemporal prediction algorithm combining hyperspectral satellites and artificial intelligence in the aforementioned embodiments, and will not be repeated here.
[0145] In addition, an embodiment of the present application further proposes a computer storage medium, on which is stored a spatiotemporal prediction program for atmospheric NO2 that combines hyperspectral satellites and artificial intelligence. When the spatiotemporal prediction program for atmospheric NO2 that combines hyperspectral satellites and artificial intelligence is executed by a processor, the steps of the spatiotemporal prediction algorithm for atmospheric NO2 that combines hyperspectral satellites and artificial intelligence as described above are implemented. Therefore, no further details will be given here. In addition, the description of the beneficial effects of adopting the same method will not be repeated. For technical details not disclosed in the computer-readable storage medium embodiment involved in this application, please refer to the description of the method embodiment of this application. As an example, program instructions can be deployed to be executed on one computing device, or on multiple computing devices located at one location, or on multiple computing devices distributed at multiple locations and interconnected by a communication network.
[0146] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The above-described program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes in the above-described method embodiments. The above-described storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).
[0147] It should also be noted that the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the device embodiments provided in this application, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art can understand and implement the present invention without inventive work.
[0148] Through the description of the above embodiments, it is clear to those skilled in the art that the present application can be implemented by means of software plus necessary general hardware, and of course it can also be implemented by dedicated hardware including application-specific integrated circuits, dedicated CPUs, dedicated memories, dedicated components, etc. In general, all functions performed by computer programs can be easily implemented with corresponding hardware, and the specific hardware structures used to implement the same function can also be various, such as analog circuits, digital circuits, or dedicated circuits, etc. However, for the present application, software program implementation is a better implementation method in most cases. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer's floppy disk, USB flash drive, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the method of each embodiment of the present application.
[0149] The above are only preferred embodiments of the present application and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.
Claims
1. A spatiotemporal prediction method for atmospheric NO2 combining hyperspectral satellite and artificial intelligence, characterized in that: The method comprises: Acquiring meteorological monitoring data, hyperspectral satellite monitoring data, and geographic information remote sensing data of the target area, wherein the geographic information remote sensing data includes surface building data, vegetation cover data, population data, and elevation data; Obtaining, based on the meteorological monitoring data and a preset atmospheric physicochemical model, predicted nitrogen dioxide concentration data for the target area, wherein the predicted nitrogen dioxide concentration data is obtained by using a pollution inventory obtained from historical statistics of the target area and the meteorological monitoring data using the preset atmospheric physicochemical model, and the predicted nitrogen dioxide concentration data is stratified according to atmospheric pressure values in a direction away from the ground; Inverting the nitrogen dioxide column concentration data for the target area based on the hyperspectral satellite monitoring data; wherein the spatial resolution of the nitrogen dioxide column concentration data is higher than the spatial resolution of the nitrogen dioxide layered concentration prediction data; Inputting the nitrogen dioxide layered concentration prediction data, the nitrogen dioxide column concentration data, and the geographic information remote sensing data into a trained nitrogen dioxide concentration prediction neural network model to obtain four-dimensional nitrogen dioxide concentration prediction data of the target area output by the nitrogen dioxide concentration prediction neural network model; the four-dimensional nitrogen dioxide concentration prediction data includes four-dimensional space-time coordinate information and a nitrogen dioxide concentration prediction value corresponding to the four-dimensional space-time coordinate, and the four-dimensional space-time coordinate information includes time information, longitude information, latitude information, and elevation information; Extracting nitrogen dioxide concentration prediction data of a target area in the target region at a target time from the four-dimensional nitrogen dioxide concentration prediction data; Before obtaining meteorological monitoring data, hyperspectral satellite monitoring data, and geographic information remote sensing data of the target area, wherein the geographic information remote sensing data includes surface building data, vegetation cover data, population data, and elevation data, the method further includes: Acquire training sample data, the training sample data including meteorological monitoring data, hyperspectral satellite monitoring data, and geographic information remote sensing data of the target area, as well as first hyperspectral satellite monitoring data of the target area at a preset time and real-time national control station monitoring data within a preset day, where the preset time is noon; Determining a training loss function for training the nitrogen dioxide concentration prediction neural network model; Training is performed based on the meteorological monitoring data, the hyperspectral satellite monitoring data, and the geographic information remote sensing data to obtain first training result data for a first preset area within the target area at a preset time and second training result data for a second preset area within the target area on a preset day; wherein the altitude of the first preset area is higher than the altitude of the second preset area, the first preset area is a middle layer area of the target area, and the second preset area is a near-ground layer area near the center point of the target area; Verifying the first training result data according to the first hyperspectral satellite data to obtain spatial correlation data between the first hyperspectral satellite data and the first training result data; Verifying the second training result data according to the real-time national control site monitoring data to obtain time correlation data and absolute value difference data between the second training result data and the real-time national control site monitoring data; Using the spatial correlation data as a first loss function value, the temporal correlation data as a second loss function value, and the absolute value difference data as a third loss function value; Performing weighted superposition on the first loss function value, the second loss function value, and the third loss function value to obtain a final loss function value; Determine whether the final loss function value meets the preset conditions; If not, the nitrogen dioxide concentration prediction neural network model is updated, and the training is returned to be performed based on the meteorological monitoring data, the hyperspectral satellite monitoring data and the geographic information remote sensing data to obtain the first training result data of the first preset area in the target area at the preset time and the second training result data of the second preset area in the target area on the preset day, until the value of the training loss function meets the preset conditions, thereby obtaining a trained nitrogen dioxide concentration prediction neural network model.
2. The atmospheric NO2 spatiotemporal prediction method combining hyperspectral satellite and artificial intelligence according to claim 1 is characterized in that: The nitrogen dioxide concentration prediction neural network model includes: A first feature extraction module is used to extract features from the nitrogen dioxide stratified concentration prediction data to obtain spatiotemporal distribution features of the nitrogen dioxide concentration stratification; A second feature extraction module is used to perform feature extraction and feature fusion on the nitrogen dioxide column concentration data to obtain spatiotemporal distribution characteristics of the nitrogen dioxide column concentration; A third feature extraction module is used to perform spatial feature extraction and dimensionality transformation on the geographic information remote sensing data to obtain spatiotemporal distribution features of the geographic information; The fully connected layer is used to perform feature fusion on the stratified spatiotemporal distribution characteristics of the nitrogen dioxide concentration, the spatiotemporal distribution characteristics of the nitrogen dioxide column concentration, and the spatiotemporal distribution characteristics of the geographic information to obtain the four-dimensional nitrogen dioxide concentration prediction data.
3. The atmospheric NO2 spatiotemporal prediction method combining hyperspectral satellite and artificial intelligence according to claim 2 is characterized in that: The second feature extraction module is also used to complete the nitrogen dioxide stratified concentration prediction data if the regional spatial coverage of the nitrogen dioxide column concentration data is greater than a first preset percentage and less than 1, to obtain the nitrogen dioxide column concentration data after data completion, and to perform feature extraction on the nitrogen dioxide column concentration data after data completion to obtain the spatiotemporal distribution characteristics of the nitrogen dioxide concentration stratification.
4. The atmospheric NO2 spatiotemporal prediction method combining hyperspectral satellite and artificial intelligence according to claim 1, characterized in that: The step of obtaining nitrogen dioxide column concentration data of the target area based on the hyperspectral satellite monitoring data includes: Filtering valid monitoring data with a cloud cover ratio less than or equal to a preset threshold from the hyperspectral satellite monitoring data; Based on the effective monitoring data, the nitrogen dioxide column concentration data of the target area is inverted.
5. A spatiotemporal prediction device for atmospheric NO2 that combines hyperspectral satellites with artificial intelligence, characterized in that: include: A data acquisition module is used to acquire meteorological monitoring data, hyperspectral satellite monitoring data, and geographic information remote sensing data of the target area, wherein the geographic information remote sensing data includes surface building data, vegetation cover data, population data, and elevation data; a stratified concentration prediction module, configured to obtain stratified nitrogen dioxide concentration prediction data for the target area based on the meteorological monitoring data and a preset atmospheric physical and chemical model, wherein the stratified nitrogen dioxide concentration prediction data is obtained by using a pollution inventory obtained from historical statistics of the target area and the meteorological monitoring data using the preset atmospheric physical and chemical model, and wherein the stratified nitrogen dioxide concentration prediction data is stratified according to the atmospheric pressure value in a direction away from the ground; a column concentration prediction module, configured to invert nitrogen dioxide column concentration data of the target area based on the hyperspectral satellite monitoring data; wherein the spatial resolution of the nitrogen dioxide column concentration data is higher than the spatial resolution of the nitrogen dioxide layered concentration prediction data; a data fusion module, configured to input the nitrogen dioxide layered concentration prediction data, the nitrogen dioxide column concentration data, and the geographic information remote sensing data into a trained nitrogen dioxide concentration prediction neural network model, to obtain four-dimensional nitrogen dioxide concentration prediction data for a target area output by the nitrogen dioxide concentration prediction neural network model; the four-dimensional nitrogen dioxide concentration prediction data comprising four-dimensional spatiotemporal coordinate information and a nitrogen dioxide concentration prediction value corresponding to the four-dimensional spatiotemporal coordinates, the four-dimensional spatiotemporal coordinate information comprising time information, longitude information, latitude information, and elevation information; and extract nitrogen dioxide concentration prediction data for a target area in the target area at a target time from the four-dimensional nitrogen dioxide concentration prediction data; The data acquisition module is also used to obtain training sample data, which includes meteorological monitoring data, hyperspectral satellite monitoring data and geographic information remote sensing data of the target area, as well as first hyperspectral satellite monitoring data of the target area at a preset time and real-time national control station monitoring data within a preset day, where the preset time is noon; determine a training loss function for training the nitrogen dioxide concentration prediction neural network model; train according to the meteorological monitoring data, the hyperspectral satellite monitoring data and the geographic information remote sensing data to obtain first training result data of a first preset area in the target area at the preset time and second training result data of a second preset area in the target area on the preset day; wherein the altitude of the first preset area is higher than the altitude of the second preset area, the first preset area is the middle layer area of the target area, and the second preset area is the near-ground layer area near the center point of the target area; verify the first training result data according to the first hyperspectral satellite data to obtain the first hyperspectral satellite data and the first training result data. spatial correlation data of the data; verifying the second training result data according to the real-time national control site monitoring data to obtain the time correlation data and absolute value difference data of the second training result data and the real-time national control site monitoring data; using the spatial correlation data as the first loss function value, the time correlation data as the second loss function value, and the absolute value difference data as the third loss function value; weighted superposition of the first loss function value, the second loss function value and the third loss function value to obtain the final loss function value; judging whether the final loss function value meets the preset conditions; if not, updating the nitrogen dioxide concentration prediction neural network model, and returning to execute the training according to the meteorological monitoring data, the hyperspectral satellite monitoring data and the geographic information remote sensing data, to obtain the first training result data of the first preset area in the target area at the preset time and the second training result data of the second preset area in the target area on the preset day, until the value of the training loss function meets the preset conditions, and a trained nitrogen dioxide concentration prediction neural network model is obtained.
6. A spatiotemporal prediction device for atmospheric NO2 that combines hyperspectral satellites with artificial intelligence, characterized in that: A processor, a memory, and a spatiotemporal prediction program for atmospheric NO2 that combines hyperspectral satellites and artificial intelligence stored in the memory. When the spatiotemporal prediction program for atmospheric NO2 that combines hyperspectral satellites and artificial intelligence is executed by the processor, the steps of the spatiotemporal prediction method for atmospheric NO2 that combines hyperspectral satellites and artificial intelligence as described in any one of claims 1 to 4 are implemented.
7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a spatiotemporal prediction program for atmospheric NO2 that combines hyperspectral satellites and artificial intelligence. When the spatiotemporal prediction program for atmospheric NO2 that combines hyperspectral satellites and artificial intelligence is executed by a processor, the spatiotemporal prediction method for atmospheric NO2 that combines hyperspectral satellites and artificial intelligence as described in any one of claims 1 to 4 is implemented.