CN114169232A - Full-time-period three-dimensional atmospheric pollutant reconstruction method and device, computer equipment and storage medium - Google Patents
Full-time-period three-dimensional atmospheric pollutant reconstruction method and device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN114169232A CN114169232A CN202111434720.5A CN202111434720A CN114169232A CN 114169232 A CN114169232 A CN 114169232A CN 202111434720 A CN202111434720 A CN 202111434720A CN 114169232 A CN114169232 A CN 114169232A
- Authority
- CN
- China
- Prior art keywords
- data
- information
- reconstruction
- atmospheric
- modal
- 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.)
- Granted
Links
- 239000003344 environmental pollutant Substances 0.000 title claims abstract description 137
- 231100000719 pollutant Toxicity 0.000 title claims abstract description 137
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000003860 storage Methods 0.000 title claims abstract description 11
- 238000009826 distribution Methods 0.000 claims abstract description 60
- 238000000605 extraction Methods 0.000 claims abstract description 43
- 239000013598 vector Substances 0.000 claims description 20
- 230000008878 coupling Effects 0.000 claims description 17
- 238000010168 coupling process Methods 0.000 claims description 17
- 238000005859 coupling reaction Methods 0.000 claims description 17
- 230000015654 memory Effects 0.000 claims description 12
- 238000004590 computer program Methods 0.000 claims description 11
- 238000005070 sampling Methods 0.000 claims description 11
- 238000005520 cutting process Methods 0.000 claims description 10
- 238000011176 pooling Methods 0.000 claims description 8
- 239000000284 extract Substances 0.000 claims description 7
- 238000004140 cleaning Methods 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 5
- 230000006798 recombination Effects 0.000 claims description 5
- 238000005215 recombination Methods 0.000 claims description 5
- 238000009792 diffusion process Methods 0.000 claims description 4
- 239000010410 layer Substances 0.000 description 35
- 238000012549 training Methods 0.000 description 32
- 238000013527 convolutional neural network Methods 0.000 description 15
- 238000013528 artificial neural network Methods 0.000 description 10
- 238000001514 detection method Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 4
- 238000012795 verification Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000012512 characterization method Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 239000002356 single layer Substances 0.000 description 3
- 239000003570 air Substances 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 210000004027 cell Anatomy 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000002093 peripheral effect Effects 0.000 description 2
- 238000010521 absorption reaction Methods 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 239000000443 aerosol Substances 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 239000012080 ambient air Substances 0.000 description 1
- 230000002238 attenuated effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000003337 fertilizer Substances 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000008521 reorganization Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
- G16C20/70—Machine learning, data mining or chemometrics
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Chemical & Material Sciences (AREA)
- Crystallography & Structural Chemistry (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a full-time atmospheric pollutant reconstruction method, a full-time atmospheric pollutant reconstruction device, computer equipment and a storage medium, wherein the full-time atmospheric pollutant reconstruction method comprises the steps of obtaining atmospheric pollutant concentration distribution obtained by inversion according to hyper-spectral remote sensing data of an observation area, and obtaining multi-source related information of atmospheric pollutant concentration in the observation area, wherein the multi-source related information comprises meteorological information, traffic information and topographic information, the multi-source related information is preprocessed to serve as a sample, and the atmospheric pollutant concentration distribution serves as a sample label; constructing a multi-modal reconstruction model comprising a feature extraction unit and a reconstruction unit; carrying out supervised learning on the multi-modal reconstruction model by using the samples and the sample labels so as to optimize parameters of the multi-modal reconstruction model; and reconstructing the distribution of the three-dimensional atmospheric pollutants in the whole time period by using the multi-modal reconstruction model with optimized parameters. The method comprehensively predicts the three-dimensional atmospheric pollutant concentration in the whole time period by considering multi-modal information so as to improve the prediction accuracy.
Description
Technical Field
The invention belongs to the field of atmospheric pollutant concentration monitoring, and particularly relates to a full-time atmospheric pollutant reconstruction method and device, computer equipment and a storage medium.
Background
For atmospheric pollution monitoring, the generation, tracing and diffusion prediction of pollutants are all very important. The single point type instrument or the vertical profile observation instrument can only monitor and analyze the pollutant emission of the block where the single point type instrument or the vertical profile observation instrument is located, trace gas from other blocks cannot be traced, and the pollution source cannot be found out. Therefore, with the goal of performing full-time three-dimensional atmospheric pollutant prediction and reconstruction on the whole area, it is imperative to construct a joint defense joint control list for regional emission to monitor, prevent and trace the 24-hour pollutant situation in all the blocks.
In the field of atmospheric pollutant detection, the existing mainstream detection equipment is mainly divided into three types, the first type is mainly a point type instrument taking CNEMC (national ambient air quality monitoring station) as an example, and the detection equipment cannot detect pollutants above the ground and in high altitudes. The second is radar, which can detect atmospheric pollutants at different height levels, but the radar instrument has a blind area. The third is a hyper-spectrum remote sensing station which can absorb and scatter sunlight which is an active light source, carry out spectral measurement by using molecular absorption of Ultraviolet (UV) and visible light (VIS) wave bands of the active light source, and carry out inversion observation on spatial distribution data, which means that the observation is interrupted at night and in rainy days. Meanwhile, a large amount of site distribution is needed for the prediction reconstruction of the atmospheric trace gas. However, due to the high construction and maintenance costs of the hyperspectral remote sensing sites, it is not feasible to construct a large number of hyperspectral remote sensing sites.
In the prior art, machine learning is also adopted to predict the concentration of the atmospheric pollutants, for example, patent document CN108491970A discloses an atmospheric pollutant concentration prediction method based on an RBF neural network, the atmospheric pollutant concentration prediction method is based on weather data and adopts the RBF neural network for prediction, and the prediction method only considers the influence factors such as weather data, which may cause relatively inaccurate prediction.
For another example, patent document CN109492830A discloses a method for predicting the emission concentration of a mobile pollution source based on time-space deep learning, which mainly uses a convolution long-term and short-term memory neural network with an attention mechanism to predict according to the regional time-space distribution characteristics of pollutants of the mobile pollution source. The prediction method directly considers the characteristics of the spatial-temporal distribution of the pollution source pollutant in the region, namely, weather information, pollutant information and timestamp information are considered for prediction, so that the prediction is relatively inaccurate.
Disclosure of Invention
In view of the above, an object of the present invention is to provide a full-time three-dimensional atmospheric pollutant reconstruction method, apparatus, computer device and storage medium, which comprehensively predicts the full-time three-dimensional atmospheric pollutant concentration by considering multi-modal information to improve the prediction accuracy.
In order to achieve the purpose, the invention provides the following technical scheme:
in a first aspect, an embodiment provides a full-time three-dimensional reconstruction method for atmospheric pollutants, which includes the following steps:
acquiring atmospheric pollutant concentration distribution obtained by inversion according to hyper-spectral remote sensing data of an observation area, and acquiring multi-source related information of the atmospheric pollutant concentration in the observation area, wherein the multi-source related information comprises meteorological information, traffic information and topographic information, the multi-source related information is preprocessed to be used as a sample, and the atmospheric pollutant concentration distribution is used as a sample label;
constructing a multi-modal reconstruction model comprising a feature extraction unit and a reconstruction unit, wherein the feature extraction unit comprises a plurality of feature extraction modules, each feature extraction module correspondingly extracts a feature vector of each type of relevant information in a sample, and the reconstruction unit performs fusion operation on the feature vectors of all relevant information to reconstruct the concentration distribution of the atmospheric pollutants;
carrying out supervised learning on the multi-modal reconstruction model by using the samples and the sample labels so as to optimize parameters of the multi-modal reconstruction model;
and reconstructing the distribution of the three-dimensional atmospheric pollutants in the whole time period by using the multi-modal reconstruction model with optimized parameters.
In one embodiment, the meteorological information includes latitude and longitude wind speed, vertical wind speed, temperature, humidity, pressure, atmospheric boundary layer thickness; the traffic information includes traffic flow data; the terrain information comprises terrain data and earth surface elevation data.
In one embodiment, the multi-source related information further includes population density information and/or a chemical-meteorological coupling model.
In one embodiment, the meteorological information is from a prediction output of a chemo-meteorological coupling model; the chemical-meteorological coupling model information includes an atmospheric pollutant concentration of an output thereof.
In one embodiment, the acquired multi-source related information is subjected to data cleaning, data filling, one-hot coding, standardization and cutting recombination to form a sample and input into the multi-modal reconstruction model.
During cutting and restructuring, determining the data size N according to the service life length of the atmospheric pollutant, cutting each type of relevant information corresponding to each sampling height according to the data size to obtain each type of relevant information corresponding to each geographic grid, splicing the relevant information corresponding to all the sampling heights M corresponding to each geographic grid according to each type of relevant information to form a data block of N M, and inputting the data block into the multi-mode restructuring model to perform feature extraction by using 1 feature extraction model.
In one embodiment, the determining the data size N × N according to the lifetime length of the atmospheric pollutant comprises:
according to the area which can be covered by the atmospheric pollutants along with the airflow diffusion for several hours under the normal latitudinal and longitudinal wind speed, the ratio of the minimum edge length value of the square envelope curve of the covered area formed when the concentration of the atmospheric pollutants reaches the pollutant standard to the unit area grid size value observed by the equipment is taken as N.
In one embodiment, the feature extraction module comprises an alternating structure of a plurality of convolutional layers and pooling layers, the convolutional kernel sizes of the convolutional layers are sequentially reduced according to the data flow sequence, and the number of channels is also sequentially smaller, so that feature extraction and dissimilarity features are realized, and feature vectors are obtained;
in one embodiment, the reconstruction unit performs full-connection operation on the input feature vectors by using a plurality of full-connection layers to reconstruct and generate a reconstructed atmospheric pollutant concentration distribution.
In a second aspect, an embodiment provides a full-time three-dimensional reconstruction apparatus for atmospheric pollutants, including:
the acquisition module is used for acquiring atmospheric pollutant concentration distribution obtained by inversion according to hyper-spectral remote sensing data of an observation area and acquiring multi-source related information of the atmospheric pollutant concentration in the observation area, wherein the multi-source related information comprises meteorological information, traffic information and topographic information, the multi-source related information is preprocessed to serve as a sample, and the atmospheric pollutant concentration distribution is used as a sample label;
the model building module is used for building a multi-modal reconstruction model comprising a feature extraction unit and a reconstruction unit, wherein the feature extraction unit comprises a plurality of feature extraction modules, each feature extraction module correspondingly extracts a feature vector of each type of relevant information in a sample, and the reconstruction unit performs fusion operation on the feature vectors of all relevant information to reconstruct the concentration distribution of the atmospheric pollutants;
the supervised learning module is used for carrying out supervised learning on the multi-modal reconstruction model by using the samples and the sample labels so as to optimize parameters of the multi-modal reconstruction model;
and the reconstruction application module is used for reconstructing the distribution of the three-dimensional atmospheric pollutants in the whole time period by utilizing the multi-modal reconstruction model optimized by the parameters.
In a third aspect, embodiments provide a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the memory stores a multi-modal reconstruction model constructed by the full-time three-dimensional atmospheric pollutants reconstruction method of the first aspect; the processor, when executing the computer program, implements the steps of:
acquiring multi-source related information of atmospheric pollutant concentration in the whole period, and preprocessing the multi-source related information to form a prediction sample;
and performing reconstruction calculation on the prediction sample by using the multi-modal reconstruction model to obtain the three-dimensional atmospheric pollutant concentration distribution.
In a fourth aspect, embodiments provide a computer-readable storage medium, on which a computer program is stored, which when executed by a processor, implements the steps of the method for reconstructing a full-time three-dimensional atmospheric pollutant according to the first aspect.
Compared with the prior art, the invention has the beneficial effects that at least:
the sample adopts multi-modal data brought by multi-source related information such as meteorological information, traffic information, topographic information, population density information, a chemical-meteorological coupling model and the like, and the sample label adopts accurate atmospheric pollutant concentration distribution obtained by inversion of hyperspectral remote sensing data, so that the multi-modal reconstruction model can accurately learn the mapping relation from the multi-modal data to the atmospheric pollutant concentration distribution in the process of supervised learning, and the accuracy of reconstruction and prediction of the atmospheric pollutant concentration distribution of the multi-modal reconstruction model is improved. Meanwhile, the multi-modal reconstruction model can predict the concentration distribution of the atmospheric pollutants according to multi-modal data, so that the whole-time and whole-region prediction of the concentration distribution of the atmospheric pollutants is realized, and the comprehensiveness of prediction application is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flow chart of a method for reconstructing a three-dimensional atmospheric pollutant over time according to an embodiment;
FIG. 2 is a schematic diagram of a multi-modal reconstruction model structure and a training process thereof according to an embodiment;
FIG. 3 shows NO in Anhui part of area after reconstitution according to an embodiment2A gas concentration profile;
FIG. 4 is a comparison graph of Pearson correlation coefficients for the multi-modal reconstruction model and the chemical-meteorological coupling model (WRF-CHEM) provided by an embodiment;
FIG. 5 is a graph comparing the error of atmospheric pollutant concentrations of the multi-modal reconstruction model and the chemical-meteorological coupling model provided by an embodiment;
fig. 6 is a schematic structural diagram of a full-time three-dimensional reconstruction device for atmospheric pollutants according to an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the detailed description and specific examples, while indicating the scope of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention.
Fig. 1 is a flowchart of a full-time three-dimensional atmospheric pollutant reconstruction method according to an embodiment. As shown in fig. 1, the method for reconstructing a full-time three-dimensional atmospheric pollutant provided by the embodiment includes the following steps:
step 1, obtaining atmospheric pollutant concentration distribution obtained by inversion according to hyper-spectral remote sensing data of an observation area, obtaining multi-source related information of atmospheric pollutant concentration in the observation area, and reconstructing sample data in time and space.
In the embodiment, when training sample data is constructed, the observation region should have a certain number of hyperspectral remote sensing sites, or a certain number of hyperspectral remote sensing sites and a prediction result with a chemical-meteorological coupling model (WRF-CHEM), so as to ensure that the observation data is valid and can be used for training sample data. Meanwhile, the observation area also requires that the original atmospheric pollutants exceed the standard within a period of time in the past, or traceable external pollutants diffuse. The observation area requires sites to be distributed in various functional areas such as rural areas, cities, suburbs, mountainous areas and the like.
Because the hyper-spectral remote sensing data acquired by the hyper-spectral remote sensing site can be inverted to obtain very accurate atmospheric pollutant concentration distribution, the three-dimensional atmospheric pollutant vertical profile data (namely, the atmospheric pollutant concentration distribution) extracted from the hyper-spectral remote sensing site is used as a sample label.
In the embodiment, the time span of a hyper-spectral remote sensing station required by training is taken as a reference, and multi-source related information of atmospheric pollutant concentration in the same observation area with the same span is obtained, wherein the multi-source related information specifically comprises meteorological information, traffic information, terrain information, population density information and chemical-meteorological coupling model information. The meteorological information includes, but is not limited to, latitude and longitude wind speed, vertical wind speed, temperature, humidity, pressure, and atmospheric boundary layer thickness, and may be extracted from the prediction result of the chemical-meteorological coupling model, or may be obtained from other channels without limitation. The traffic information includes but is not limited to traffic flow data, and can be extracted from the network map. The terrain information includes, but is not limited to, terrain data, surface elevation data, which may be extracted from Landsat satellite data. The population density information includes, but is not limited to, population density data, which can be extracted from various local statistical office data. The chemical-meteorological coupling model information includes, but is not limited to, atmospheric pollutant concentration, latitude and longitude wind speed, vertical wind speed and the like output by the model.
The related information of each type is used as modal data, such as longitude and latitude wind speed as one modal data, vertical wind speed as another modal data, traffic flow data as another modal data, and multiple modal data processes are subjected to space-time reconstruction to form sample data for training and predicting the concentration distribution of the atmospheric pollutants.
The multi-modal data is required to be preprocessed before forming training samples and inputting the training samples into the multi-modal reconstruction model, wherein the preprocessing comprises data cleaning, data filling, one-hot coding, standardization and cutting and reorganization.
And (3) data cleaning: for data before training, a certain amount of unreasonable and untrained data, such as 0 value, an overlarge value and the like, appear in the data due to other problems such as storage, instrument error, noise, faults and the like, the error data of the type of the data center with obvious deviation needs to be manually deleted and temporarily assigned with null values, meanwhile, in order to improve the stability of the algorithm, hyperspectral remote sensing observation values with large measurement errors, including values with root mean square value (RMS) larger than 0.1 during fitting or corresponding aerosol optical thickness larger than 3, and the like, are removed.
For data filling: a large number of default values appear after the data are cleaned, and if the data on the same day have too many default values, the data on the same day can be judged to be invalid and deleted. And filling data into the default part of the default value with the low aggregation degree, wherein the filling mode can be that the attributes in the initial data set are divided into numerical attributes and non-numerical attributes to be processed respectively. If the null value is numerical, filling the missing attribute value according to the average value of the values of the attribute in all other objects; if the null value is non-numerical, the missing attribute value is filled up by the value with the most value of the attribute in all other objects (namely the value with the highest frequency of occurrence) according to the mode principle in statistics.
For one-hot encoding: for discrete parameter types such as time, most time objects are in a discontinuous state, the value of the data of the type generally only takes certain fixed values, so the data cannot be directly put into a neural network for training, before that, the data needs to be mapped into another set of continuous coding structure without changing the representation capability of the data, the operation is an independent operation, the value of the discrete feature is expanded to the Euclidean space, and a certain value of the discrete feature corresponds to a certain point of the Euclidean space. Using one-hot encoding for discrete features makes the distance calculation between features more reasonable.
For data normalization: in the multi-modal reconstruction model formed by the convolutional neural network, the type distribution of data is very wide, to ensure that data of different magnitudes have substantially the same characterization capabilities with respect to changes in convolutional neural network weights, if the data is directly input into the convolutional neural network, the data with large magnitude such as height and the like has larger influence on the convolutional neural network relative to the data with concentration, but in practice the characterization capability of a feature on a convolutional neural network should not be related to the magnitude or unit of the value, and if the center of some types of data is at a height that is either well above zero or well below zero, the magnitude change of the data is greatly attenuated in convolutional neural training, if the mean of the data is at zero, the size change characterization capability of the data can be reasonably loaded into the multi-modal reconstruction model. This requires standardization of the data. Normalization is to scale the data to fall within a small specific interval, subject to normal distribution, removing unit and dimension constraints of the data.
Aiming at cutting recombination: the acquired multi-model data can be converted into a multi-layer or single-layer area plane data set, the traffic information and population density information refer to single-layer area plane data, and the meteorological information, the topographic information and the chemical-meteorological coupling model information are multi-layer area plane data. In order to make the acquired multi-modal data convenient for the processing of the convolutional neural network, data segmentation and recombination are required. The cutting and recombining process comprises the following steps: determining data size N according to the life length of the atmospheric pollutants, cutting each type of relevant information corresponding to each sampling height according to the data size to obtain each type of relevant information corresponding to each geographic grid, and splicing the relevant information corresponding to all sampling heights M corresponding to each geographic grid according to each type of relevant information to form a data block of N M, wherein each height corresponds to one sampling layer, and M can also be understood as M sampling layers. And inputting each data block of N by M into the multi-modal reconstruction model, and performing feature extraction by 1 feature extraction model. Each mode corresponds to a data block of N × M, and if there are L modes, the data blocks of L N × M form 1 training sample for predicting the atmospheric pollutant concentration distribution of the corresponding geographic grid region.
When the modal data such as traffic flow data has only one sampling height, the data at one sampling height is copied to construct a data block of N × M when the data is clipped and reorganized.
In an embodiment, the determining the data size N × N according to the lifetime length of the atmospheric pollutant includes: according to the area which can be covered by the atmospheric pollutants along with the airflow diffusion for several hours under the normal latitudinal and longitudinal wind speed, the ratio of the minimum edge length value of the square envelope curve of the covered area formed when the concentration of the atmospheric pollutants reaches the pollutant standard to the unit area grid size value observed by the equipment is taken as N. The chemical-meteorological coupling model is cut into N x N data by using meteorological data (such as temperature, air pressure, latitude, longitude and vertical wind speed) centering on a target point and other relevant parameters such as vehicle flow, land use type and vertical distribution of atmospheric pollutants.
As training samples and sample labels for the convolutional deep learning, in order to make the convolutional neural network have certain effectiveness and prevent overfitting, the training should have enough data, and a plurality of data of hyper-spectral remote sensing sites with time span of at least two years and multi-modal data are used as a training set and a verification set in the neural network.
In the embodiment, the training set validation set configuration mode of the convolutional neural network is as follows: the splitting ratio of the training set and the verification set can be selected and is not limited to 70% and 30%, and the splitting mode can be selected and is not limited to random splitting. And taking all data of other non-long-term hyper-spectral remote sensing sites in the selected observation area as a verification set without splitting. And simultaneously, splitting a training set by using k-fold cross validation, splitting all data into k equal parts according to the data size, extracting one part from each round of training to be used as a test set, performing model training by using the other k-1 parts as the training set, and performing repeated sampling for k times in total iteration to ensure that all data are distributed to the test and training sets so as to reduce the dependence degree of the model on the classified characteristics and reduce the degree of overfitting.
And 2, constructing a multi-modal reconstruction model based on the convolution application network, wherein the multi-modal reconstruction model comprises a feature extraction unit and a reconstruction unit.
In an embodiment, the feature extraction unit includes a plurality of feature extraction modules, and each feature extraction module correspondingly extracts a feature vector of each type of related information in the sample. In a possible implementation manner, the feature extraction module includes an alternating structure of a plurality of convolution layers and pooling layers, and the convolution kernel sizes of the convolution layers are sequentially reduced and the number of channels is sequentially smaller according to the data flow sequence, so that feature extraction and dissimilarity features are realized, and feature vectors are obtained. In another possible implementation, the data block of N × M corresponding to each mode is first subjected to feature extraction for a convolution layer with a large window size and a large output channel, then subjected to a pooling layer, and further subjected to feature differentiation for a convolution layer with a small window size and a small output channel, and finally subjected to feature extraction to obtain feature output.
In an embodiment, the reconstruction unit performs a fusion operation on the feature vectors of all the relevant information to reconstruct the atmospheric pollutant concentration distribution. In one possible embodiment, the reconstruction unit performs a full-connection operation on the input feature vectors using a plurality of full-connection layers to reconstruct and generate a reconstructed atmospheric pollutant concentration distribution. In another possible implementation, the feature vectors output by all the feature extraction modules are subjected to full-connection operation through two full-connection layers to obtain the prediction result of the atmospheric pollutant concentration distribution. It should be noted that the number of nodes in the final full-link layer determines the height layer for outputting the atmospheric pollutant concentration distribution, that is, if there are P outputs corresponding to P nodes, the atmospheric pollutant concentrations corresponding to P heights are indicated. The P heights can be equal to or different from the M inputted heights and are set according to requirements.
It should be noted that the number of convolutional layers, the number of pooling layers, the number of full-link layers, and the number of nodes of the convolutional neural network may be selected according to various parameters such as the data volume, the preprocessing mode, the reconstructed gas life type, and the like.
And 3, performing supervised learning on the multi-modal reconstruction model by using the samples and the sample labels to optimize parameters of the multi-modal reconstruction model.
During supervised learning, a difference value of an atmospheric pollutant concentration distribution reconstructed prediction result output by the multi-modal reconstruction model and an atmospheric pollutant concentration distribution obtained by inversion of hyper-spectral remote sensing data serving as a label is used as a loss function to feed back the convolutional neural network so as to optimize parameters of the convolutional neural network, and finally, a corresponding number of samples are selected according to the size of batch processing parameters to optimize a network node weight structure in the next iteration.
After each batch of training is finished, the prediction effect of the multi-modal reconstruction model of the training is evaluated by using the test set, the performance of the multi-modal reconstruction model is generally judged by using the precision, the accuracy, the recall rate and the like, the continuity of the data comparison follow-up data of the hyper-spectral remote sensing site at night and in rainy days is observed to evaluate the reconstruction capability of the multi-modal reconstruction model, and meanwhile, the difference of the performance of the multi-modal reconstruction model is compared and analyzed with the chemical-meteorological coupling model in the region with the hyper-spectral remote sensing site.
And 4, reconstructing the distribution of the three-dimensional atmospheric pollutants in the whole time period by using the multi-modal reconstruction model with optimized parameters.
When the multi-modal reconstruction model is used for reconstructing and predicting the distribution of the three-dimensional atmospheric pollutants in the whole time period, the obtained multi-source related information, namely multi-modal data is obtained, then the multi-modal data is subjected to data cleaning, data filling, single-hot encoding, standardization, cutting and recombination according to the mode of the step 1 to form the multi-modal reconstruction model with optimized sample input parameters, and the reconstruction of the distribution of the three-dimensional atmospheric pollutants in the whole time period is obtained through calculation.
The method for reconstructing the full-time three-dimensional atmospheric pollutants provided by the embodiment is described in the following specific experimental examples.
As shown in FIG. 3, the experiment example selects the natural NO in Anhui region2And (4) pollution conditions, such as detection of local fire, local chemical pollution gas emission and the like. The method includes the steps that hyper-spectral remote sensing sites arranged in the area are selected as much as possible, a certain number of hyper-spectral remote sensing sites in a ring area around the area are properly selected, meanwhile, various functional areas such as suburban areas, suburban areas and mountainous areas in the urban area are selected, and suburban areas such as scientific island sites and urban areas such as western science are selected. If necessary to study the delimited region NO2Origin and propagation of pollution, e.g. produced by the other, but by the gas streamCarry to local NO2And (4) pollution. The hyper-spectral remote sensing sites which are already arranged in the area and the peripheral ring are selected as much as possible, and meanwhile, a certain number of hyper-spectral remote sensing sites in the peripheral two-ring block of the area are properly selected. Because deep learning needs observation data of the hyper-spectral remote sensing sites as labels to optimize the weight structure of the network, and at least 4000 hyper-spectral vertical profiles are needed, enough observation data of the hyper-spectral remote sensing sites are needed as learning labels to train an effective network, so that at least four long-term hyper-spectral remote sensing sites which have been operated for two years and meet the above conditions are needed as training sets, and at least four short-term hyper-spectral remote sensing sites which have been operated for two months are needed as test sets.
As shown in FIG. 2, the experimental example extracts NO from the chemical-meteorological coupling model2Concentration, longitude and latitude wind speed, vertical wind speed, temperature, humidity, boundary layer thickness and other meteorological information at different heights. And extracting traffic information such as traffic flow information of Anhui provinces from a network map, extracting information such as landform data and surface elevation data of Anhui provinces from Landsat satellite data, extracting various information such as population density information from data of an Anhui province statistical bureau, and randomly splitting the obtained information into a training set and a verification set according to a ratio of 7: 3.
Hyper-spectral remote sensing site data for training set and weather information of WRF-CHEM should have a certain amount of NO within two years of past time span2The occurrence, transmission, etc. of pollutants, e.g. NO, which has been of greater influence in the past in the area2And (4) pollution, wherein the data time is selected to configure the starting point one month before the pollution occurs. The hyper-spectral remote sensing site data of the test set at least comprises NO once2Native high value or propagate high value cases.
After obtaining the multi-modal data, the multi-modal data is subjected to data cleaning, data filling, one-hot encoding and standardization.
When training the convolutional neural network, a batch parameter is set and input]Representing a data block size of 4 x 4 with 44 vertical layers, this parameter selection principleIs NO2The gas has a transmission distance over its lifetime of not more than 80km and a cell size of 20km so a cell size of 4 x 4 is chosen. The batch parameter is the number of samples used to update the weights of the multi-modal convolutional neural network neurons in each iteration, and the number of samples used to update the network weights in each learning iteration. The network inputs are six different types of image inputs, temperature, air pressure, boundary layer thickness, vehicle flow and vertical wind speed, and NO2Each input is a data block of 4 × 4 area size and height-divided into 44 layers of original data.
As shown in fig. 2, each feature extraction module first performs a convolution operation with a window size of 3 × 3 to extract features of the blob, sets the data channel of the first convolution layer to 64, and then performs a pooling to compress the data size, with a pooling span of 2 × 2, via a plurality of identical feature extraction modules. Then into the second convolutional layer, with window size 1 convolutional layer output channel 32, and finally again using the 2x2 span pooling layer further compression feature.
Two full-connection layers with 64 and 6 nodes are added, the final six-dimensional vector is used as an output result, and each dimension corresponds to the plaque in the hyperspectral remote sensing NO2Reconstructing a height corresponding to the detection value, wherein six dimensions correspond to six layers of the label, each dimension corresponds to the plaque under the same longitude and latitude, and NO observed by the hyperspectral remote sensing site of the corresponding block is used2The difference between the six-layer columnar distribution and the output reconstructed value is calculated by taking the six-layer columnar distribution as a label, the difference is fed back to the neurons of the neural network to optimize the weight structure between the neurons, and the loss function is
And (3) obtaining a multi-modal reconstruction model which is trained and can represent the reconstructed hyperspectral remote sensing predicted value after a large amount of WRF-CHEM predicted data are used for carrying out repeated iteration comparison and the difference value of the vertical concentration distribution and optimizing the weight structure of the neuron according to the result.
In order to evaluate the performance of the neural network, a WRF-CHEM model is selected according to the comparison model, and meanwhile, a first selected index is a Pearson correlation coefficient between reconstructed data given by the neural network and real observation data of hyperspectral remote sensing, wherein the Pearson correlation coefficient is divided into a single-layer Pearson correlation coefficient and an overall Pearson correlation coefficient. The data of 18 sites in Anhui region are selected to carry out multi-mode convolutional neural network and WRF-CHEM model training, and the training result is shown in figure 4 after the integral Pearson correlation coefficient is calculated. The results show that the Pearson correlation coefficients of the multi-mode convolutional neural network are all higher than those of the WRF-CHEM model. The multi-mode convolution neural network is similar to a real observed value, and the prediction accuracy of the neural network is higher.
Some other evaluation index combination evaluation should also be used, including Spearman correlation coefficient, MAE and RMSE. The method comprises the steps of extracting a sample, performing multi-modal reconstruction model and WRF-CHEM model training on the sample and data of a scientific and western district site, a combined fertilizer scientific island site, a constant central square site and a southern Xin large site in a nearby area, wherein the training results show that the Spearman correlation coefficients of the multi-modal reconstruction model and the WRF-CHEM model of the four sites are 0.747 and 0.443, 0.778 and 0.736, 0.714 and 0.535, and 0.706 and 0.528 respectively, the Spearman correlation coefficients of the multi-modal reconstruction model are higher than those of the WRF-CHEM model, and the data show that the multi-modal reconstruction model is more strongly connected with a true value. Meanwhile, the performance of the two models is evaluated by comparing the training result with the Mean Absolute Error (MAE) and the Root Mean Square Error (RMSE) of the real observation data, and the result is shown in FIG. 5. Compared with the WRF-CHEM model, the multi-modal reconstruction model in the invention has smaller error and certain advantages in the evaluation of the error index through the graph 5, which shows that the multi-modal reconstruction model in the invention has certain prediction accuracy and practical value.
Fig. 6 is a schematic structural diagram of a full-time three-dimensional reconstruction device for atmospheric pollutants according to an embodiment. As shown in fig. 6, the reconstruction apparatus according to the embodiment includes:
the acquisition module is used for acquiring atmospheric pollutant concentration distribution obtained by inversion according to hyper-spectral remote sensing data of an observation area and acquiring multi-source related information of the atmospheric pollutant concentration in the observation area, wherein the multi-source related information comprises meteorological information, traffic information and topographic information, the multi-source related information is preprocessed to serve as a sample, and the atmospheric pollutant concentration distribution is used as a sample label;
the model building module is used for building a multi-modal reconstruction model comprising a feature extraction unit and a reconstruction unit, wherein the feature extraction unit comprises a plurality of feature extraction modules, each feature extraction module correspondingly extracts a feature vector of each type of relevant information in a sample, and the reconstruction unit performs fusion operation on the feature vectors of all relevant information to reconstruct the concentration distribution of the atmospheric pollutants;
the supervised learning module is used for carrying out supervised learning on the multi-modal reconstruction model by using the samples and the sample labels so as to optimize parameters of the multi-modal reconstruction model;
and the reconstruction application module is used for reconstructing the distribution of the three-dimensional atmospheric pollutants in the whole time period by utilizing the multi-modal reconstruction model optimized by the parameters.
It should be noted that, when the reconstruction apparatus for full-time three-dimensional atmospheric pollutants provided in the foregoing embodiment reconstructs full-time three-dimensional atmospheric pollutants, the division of the functional modules is taken as an example, and the function distribution may be completed by different functional modules according to needs, that is, the internal structure of the terminal or the server is divided into different functional modules, so as to complete all or part of the functions described above. In addition, the device for reconstructing a three-dimensional atmospheric pollutant in a full period of time provided by the above embodiment and the method for reconstructing a three-dimensional atmospheric pollutant in a full period of time provided by the above embodiment belong to the same concept, and the specific implementation process is described in the embodiment of the method for reconstructing a three-dimensional atmospheric pollutant in a full period of time, and is not described herein again.
The embodiment also provides a computer device, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the memory stores a multi-modal reconstruction model constructed by the reconstruction method of the full-time three-dimensional atmospheric pollutants; the processor, when executing the computer program, implements the steps of:
step 1, obtaining multisource related information of atmospheric pollutant concentration in the whole time period, and preprocessing the multisource related information to form a prediction sample;
and 2, performing reconstruction calculation on the prediction sample by using the multi-modal reconstruction model to obtain the full-time three-dimensional atmospheric pollutant concentration distribution.
In practical applications, the memory may be a volatile memory at the near end, such as RAM, a non-volatile memory, such as ROM, FLASH, a floppy disk, a mechanical hard disk, etc., or a remote storage cloud. The processor may be a Central Processing Unit (CPU), a microprocessor unit (MPU), a Digital Signal Processor (DSP), or a Field Programmable Gate Array (FPGA), i.e., the reconstruction step of the full-time three-dimensional atmospheric pollutants may be realized by these processors.
Embodiments also provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program is configured to perform the above method for reconstructing three-dimensional atmospheric pollutants in a full time period when being processed and executed, and comprises the following steps:
step 1, obtaining atmospheric pollutant concentration distribution obtained by inversion according to hyper-spectral remote sensing data of an observation area, obtaining multi-source related information of atmospheric pollutant concentration in the observation area, and reconstructing sample data in time and space;
step 2, constructing a multi-modal reconstruction model based on the convolution application network, wherein the multi-modal reconstruction model comprises a feature extraction unit and a reconstruction unit;
step 3, supervised learning is carried out on the multi-modal reconstruction model by using the samples and the sample labels so as to optimize parameters of the multi-modal reconstruction model;
and 4, reconstructing the distribution of the three-dimensional atmospheric pollutants in the whole time period by using the multi-modal reconstruction model with optimized parameters.
In embodiments, the computer-readable storage medium may be a ROM, a Random Access Memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.
The above-mentioned embodiments are intended to illustrate the technical solutions and advantages of the present invention, and it should be understood that the above-mentioned embodiments are only the most preferred embodiments of the present invention, and are not intended to limit the present invention, and any modifications, additions, equivalents, etc. made within the scope of the principles of the present invention should be included in the scope of the present invention.
Claims (10)
1. A full-time three-dimensional atmospheric pollutant reconstruction method is characterized by comprising the following steps:
acquiring atmospheric pollutant concentration distribution obtained by inversion according to hyper-spectral remote sensing data of an observation area, and acquiring multi-source related information of the atmospheric pollutant concentration in the observation area, wherein the multi-source related information comprises meteorological information, traffic information and topographic information, the multi-source related information is preprocessed to be used as a sample, and the atmospheric pollutant concentration distribution is used as a sample label;
constructing a multi-modal reconstruction model comprising a feature extraction unit and a reconstruction unit, wherein the feature extraction unit comprises a plurality of feature extraction modules, each feature extraction module correspondingly extracts a feature vector of each type of relevant information in a sample, and the reconstruction unit performs fusion operation on the feature vectors of all relevant information to reconstruct the concentration distribution of the atmospheric pollutants;
carrying out supervised learning on the multi-modal reconstruction model by using the samples and the sample labels so as to optimize parameters of the multi-modal reconstruction model;
and reconstructing the distribution of the three-dimensional atmospheric pollutants in the whole time period by using the multi-modal reconstruction model with optimized parameters.
2. The method for reconstructing three-dimensional atmospheric pollutants in all time according to claim 1, wherein the meteorological information comprises longitude and latitude wind speed, vertical wind speed, temperature, humidity, pressure, atmospheric boundary layer thickness; the traffic information includes traffic flow data; the terrain information comprises terrain data and earth surface elevation data.
3. The method for reconstructing a three-dimensional atmospheric pollutant for a full period of time according to claim 1, wherein the multi-source related information further comprises population density information and/or chemical-meteorological coupling model information.
4. The method for reconstructing three-dimensional atmospheric pollutants according to claim 2 or 3, wherein the meteorological information is derived from a prediction output of a chemical-meteorological coupling model;
the chemical-meteorological coupling model information includes an atmospheric pollutant concentration of an output thereof.
5. The full-time three-dimensional atmospheric pollutant reconstruction method according to claim 1, characterized in that the acquired multi-source related information is subjected to data cleaning, data filling, one-hot encoding, standardization and clipping and recombination to form a sample and input the sample into a multi-modal reconstruction model;
during cutting and restructuring, determining the data size N according to the service life length of the atmospheric pollutant, cutting each type of relevant information corresponding to each sampling height according to the data size to obtain each type of relevant information corresponding to each geographic grid, splicing the relevant information corresponding to all the sampling heights M corresponding to each geographic grid according to each type of relevant information to form a data block of N M, and inputting the data block into the multi-mode restructuring model to perform feature extraction by using 1 feature extraction model.
6. The method for reconstructing three-dimensional atmospheric pollutants according to claim 5, wherein the determining the data size N x N according to the life length of the atmospheric pollutants comprises:
according to the area which can be covered by the atmospheric pollutants along with the airflow diffusion for several hours under the normal latitudinal and longitudinal wind speed, the ratio of the minimum edge length value of the square envelope curve of the covered area formed when the concentration of the atmospheric pollutants reaches the pollutant standard to the size value of the unit area grid observed and divided by the equipment is taken as N.
7. The full-time three-dimensional atmospheric pollutant reconstruction method according to claim 1, characterized in that the feature extraction module comprises an alternating structure of a plurality of convolution layers and pooling layers, the convolution kernel sizes of the convolution layers are sequentially reduced according to the data flow sequence, and the number of channels is also sequentially smaller, so as to realize feature extraction and dissimilarity features, and obtain feature vectors;
the reconstruction unit adopts a plurality of full-connection layers to perform full-connection operation on the input characteristic vectors so as to reconstruct and generate reconstructed atmospheric pollutant concentration distribution.
8. A device for reconstructing three-dimensional atmospheric pollutants in a full time period, comprising:
the acquisition module is used for acquiring atmospheric pollutant concentration distribution obtained by inversion according to hyper-spectral remote sensing data of an observation area and acquiring multi-source related information of the atmospheric pollutant concentration in the observation area, wherein the multi-source related information comprises meteorological information, traffic information and topographic information, the multi-source related information is preprocessed to serve as a sample, and the atmospheric pollutant concentration distribution is used as a sample label;
the model building module is used for building a multi-modal reconstruction model comprising a feature extraction unit and a reconstruction unit, wherein the feature extraction unit comprises a plurality of feature extraction modules, each feature extraction module correspondingly extracts a feature vector of each type of relevant information in a sample, and the reconstruction unit performs fusion operation on the feature vectors of all relevant information to reconstruct the concentration distribution of the atmospheric pollutants;
the supervised learning module is used for carrying out supervised learning on the multi-modal reconstruction model by using the samples and the sample labels so as to optimize parameters of the multi-modal reconstruction model;
and the reconstruction application module is used for reconstructing the distribution of the three-dimensional atmospheric pollutants in the whole time period by utilizing the multi-modal reconstruction model optimized by the parameters.
9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the memory stores a multimodal reconstruction model constructed using the full-time three-dimensional atmospheric pollutants reconstruction method of any one of claims 1 to 7; the processor, when executing the computer program, implements the steps of:
acquiring multi-source related information of atmospheric pollutant concentration in the whole time period, and preprocessing the multi-source related information to form a prediction sample;
and performing reconstruction calculation on the prediction sample by using the multi-modal reconstruction model to obtain the full-time three-dimensional atmospheric pollutant concentration distribution.
10. A computer readable storage medium having stored thereon a computer program, wherein the computer program is configured to, when executed, perform the steps of the method for reconstructing three-dimensional full-time atmospheric pollutants as claimed in any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111434720.5A CN114169232B (en) | 2021-11-29 | 2021-11-29 | Reconstruction method and device for full-time three-dimensional atmospheric pollutants, computer equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111434720.5A CN114169232B (en) | 2021-11-29 | 2021-11-29 | Reconstruction method and device for full-time three-dimensional atmospheric pollutants, computer equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114169232A true CN114169232A (en) | 2022-03-11 |
| CN114169232B CN114169232B (en) | 2024-08-09 |
Family
ID=80481472
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111434720.5A Active CN114169232B (en) | 2021-11-29 | 2021-11-29 | Reconstruction method and device for full-time three-dimensional atmospheric pollutants, computer equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114169232B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114792115A (en) * | 2022-05-17 | 2022-07-26 | 哈尔滨工业大学 | Telemetry signal outlier removing method, device and medium based on deconvolution reconstruction network |
| CN114936664A (en) * | 2022-03-18 | 2022-08-23 | 桂林电子科技大学 | Watershed non-point source pollution prediction method based on hybrid deep learning model |
| CN115455814A (en) * | 2022-08-31 | 2022-12-09 | 南京云创大数据科技股份有限公司 | Pollution source searching and pollutant distribution predicting method and system based on deep learning |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070010207A1 (en) * | 2005-07-05 | 2007-01-11 | Dooley John A | System and method for the ultra-precise analysis and characterization of RF propagation dynamics in wireless communication networks |
| CN109213839A (en) * | 2018-09-12 | 2019-01-15 | 北京英视睿达科技有限公司 | Hot spot grid pollutant data capture method based on the study of gridding depths of features |
| CN113297528A (en) * | 2021-06-10 | 2021-08-24 | 四川大学 | NO based on multi-source big data2High-resolution space-time distribution calculation method |
| CN113297527A (en) * | 2021-06-09 | 2021-08-24 | 四川大学 | PM based on multisource city big data2.5Overall domain space-time calculation inference method |
| WO2021208393A1 (en) * | 2020-04-15 | 2021-10-21 | 北京工业大学 | Inversion estimation method for air pollutant emission inventory |
-
2021
- 2021-11-29 CN CN202111434720.5A patent/CN114169232B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070010207A1 (en) * | 2005-07-05 | 2007-01-11 | Dooley John A | System and method for the ultra-precise analysis and characterization of RF propagation dynamics in wireless communication networks |
| CN109213839A (en) * | 2018-09-12 | 2019-01-15 | 北京英视睿达科技有限公司 | Hot spot grid pollutant data capture method based on the study of gridding depths of features |
| WO2021208393A1 (en) * | 2020-04-15 | 2021-10-21 | 北京工业大学 | Inversion estimation method for air pollutant emission inventory |
| CN113297527A (en) * | 2021-06-09 | 2021-08-24 | 四川大学 | PM based on multisource city big data2.5Overall domain space-time calculation inference method |
| CN113297528A (en) * | 2021-06-10 | 2021-08-24 | 四川大学 | NO based on multi-source big data2High-resolution space-time distribution calculation method |
Non-Patent Citations (2)
| Title |
|---|
| 万永权;徐方勤;燕彩蓉;苏厚勤;: "融合气象参数及污染物浓度的空气质量预测方法", 计算机应用与软件, no. 08, 12 August 2018 (2018-08-12) * |
| 曹欢;和栋材;李小飞;: "基于PIE的大气污染遥感监测系统的设计与实现", 地理空间信息, no. 05, 24 May 2018 (2018-05-24) * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114936664A (en) * | 2022-03-18 | 2022-08-23 | 桂林电子科技大学 | Watershed non-point source pollution prediction method based on hybrid deep learning model |
| CN114792115A (en) * | 2022-05-17 | 2022-07-26 | 哈尔滨工业大学 | Telemetry signal outlier removing method, device and medium based on deconvolution reconstruction network |
| CN114792115B (en) * | 2022-05-17 | 2023-04-07 | 哈尔滨工业大学 | Telemetry signal outlier removing method, device and medium based on deconvolution reconstruction network |
| CN115455814A (en) * | 2022-08-31 | 2022-12-09 | 南京云创大数据科技股份有限公司 | Pollution source searching and pollutant distribution predicting method and system based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114169232B (en) | 2024-08-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114169232A (en) | Full-time-period three-dimensional atmospheric pollutant reconstruction method and device, computer equipment and storage medium | |
| CN111666656A (en) | Rainfall estimation method and rainfall monitoring system based on microwave rainfall attenuation | |
| Aryee et al. | Development of high spatial resolution rainfall data for Ghana | |
| CN106951979A (en) | The crop maturity phase Forecasting Methodology that remote sensing, crop modeling are merged with weather forecast | |
| CN111310898A (en) | Landslide hazard susceptibility prediction method based on RNN | |
| CN110134907B (en) | Rainfall missing data filling method and system and electronic equipment | |
| Biard et al. | Automated detection of weather fronts using a deep learning neural network | |
| CN110019595A (en) | A kind of integrated method and system of multi-source meteorological data | |
| CN112163375A (en) | Long-time sequence near-surface ozone inversion method based on neural network | |
| Jiang et al. | A Municipal PM2. 5 Forecasting Method Based on Random Forest and WRF Model. | |
| CN113987912A (en) | Pollutant on-line monitoring system based on geographic information | |
| CN115481558A (en) | Atmospheric pollution emission inversion method, system and equipment based on machine learning | |
| Fan et al. | A comparative study of four merging approaches for regional precipitation estimation | |
| CN117075138A (en) | Remote sensing measurement and calculation method, system and medium for canopy height of 30-meter forest in area | |
| CN115420690A (en) | Near-surface trace gas concentration inversion model and inversion method | |
| CN114898138A (en) | Method for predicting forest fire based on landform and remote sensing data | |
| CN114819264A (en) | Photovoltaic power station irradiance ultra-short term prediction method based on space-time dependence and storage medium | |
| CN117851927A (en) | Cloud prediction method based on random forest and meteorological data | |
| Xiu-Yan et al. | Global horizontal irradiance prediction model considering the effect of aerosol optical depth based on the Informer model | |
| CN117219183A (en) | High coverage near ground NO in cloudy rain areas 2 Concentration estimation method and system | |
| CN117233869A (en) | Site short-term wind speed prediction method based on GRU-BiTCN | |
| CN116705186A (en) | Deep learning-based wind cloud satellite near-surface air temperature inversion method | |
| CN115825338A (en) | Ozone monitoring downscaling method, device, equipment and computer readable storage medium | |
| CN114441456B (en) | Vegetation coverage inversion method combining multi-source remote sensing data with geographical partitioning data | |
| CN114998719A (en) | Forest fire prediction method based on deep learning and multi-source remote sensing data |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |