CN114861882B - CO (carbon monoxide) 2 Space-time distribution reconstruction method and system - Google Patents
CO (carbon monoxide) 2 Space-time distribution reconstruction method and system Download PDFInfo
- Publication number
- CN114861882B CN114861882B CN202210490951.6A CN202210490951A CN114861882B CN 114861882 B CN114861882 B CN 114861882B CN 202210490951 A CN202210490951 A CN 202210490951A CN 114861882 B CN114861882 B CN 114861882B
- Authority
- CN
- China
- Prior art keywords
- neural network
- data
- network model
- deep neural
- environment
- 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.)
- Active
Links
Images
Classifications
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/21—Design, administration or maintenance of databases
- G06F16/211—Schema design and management
- G06F16/212—Schema design and management with details for data modelling support
-
- 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
-
- 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)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Testing Or Calibration Of Command Recording Devices (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a CO 2 The space-time distribution reconstruction method and system comprise the following steps: s1: establishing an environment database and a multi-output deep neural network model; s2: by NO 2 The satellite remote sensing data and the environment data perform initial training on the multi-output deep neural network model; s3: by CO 2 The satellite remote sensing data and the environment data perform secondary training on the multi-output deep neural network model after initial training; s4: CO using environment data and trained multi-output deep neural network model 2 Predicting the space-time distribution to obtain CO 2 And reconstructing a result of the space-time distribution. The invention reconstructs NO with higher accuracy 2 Satellite data high space-time resolution data set and NO 2 The information of fossil fuel combustion represented by satellite data is given to a model to realize CO 2 Reconstruction of high spatial-temporal resolution spatial-temporal distribution.
Description
Technical Field
The invention relates to the technical field of environmental monitoring, in particular to a CO 2 A space-time distribution reconstruction method and a system.
Background
At present, CO is reconstituted 2 The machine learning method of the space-time distribution mainly comprises A Neural Network (ANN), a limit gradient propulsion machine (XGBOOST), an optical gradient propulsion machine (Light-GBM) and the like. Reconstruction of CO using machine learning 2 In the technical modeling of space-time distribution, the dependent variable is CO 2 Satellite remote sensing data, the independent variables mainly comprise: land use type, normalized vegetation index, meteorological conditions, population, altitude, road information, and CO 2 Discharge list, etc. The prior art is to CO 2 The lack of support for relevant data information when reconstructing the spatiotemporal distribution results in CO 2 Larger deviation occurs in the reconstruction process; at the same time due to CO 2 The satellite is easy to be interfered by external conditions when acquiring data, and the acquired satellite remote sensing data has small sample size and sampling deviation, so that the satellite remote sensing data is relative to CO 2 Certain undervalue overestimation problems exist in areas with scarce satellite data, while for high vegetation coverage, high altitude industrial areas CO 2 There is a certain high value underestimation problem with concentration.
In view of the above problems, there are some prior art approaches to NO 2 Satellite remote sensing data (TROPOMI-NO) 2 ) Sampling and time-space interpolation are carried out, and the sampling and time-space interpolation is used as independent variable to be added into a model for CO 2 Is a spatial-temporal distribution reconstruction of (c). However, TROPOMI-NO 2 Coverage and space-time resolution of raw data are difficult to meet CO 2 The need for high resolution spatiotemporal distribution reconstruction; and TROPOMI-NO reconstructed by space-time kriging interpolation 2 The result is larger deviation, and the CO is reduced 2 Accuracy of the time-space reconstruction result.
In view of this, the present application is specifically proposed.
Disclosure of Invention
The technical problems to be solved by the invention are as follows: CO using the prior art 2 The accuracy of the result obtained by the reconstruction of the space-time distribution is low, and the aim is to provide a CO 2 Space-time distribution reconstruction method and system, satellite data of TROPOMI-NO2 is used as related information of fossil fuel combustion, and base is usedTransfer learning method for shared parameters to realize CO 2 Reconstruction of the spatial-temporal distribution of the concentration global domain.
The invention is realized by the following technical scheme:
in one aspect, the present invention provides a CO 2 The space-time distribution reconstruction method comprises the following steps:
s1: establishing an environment database, wherein the environment database comprises: TROPOMI-NO 2 Satellite remote sensing data, CO 2 Satellite remote sensing data and environmental related base data;
s2: establishing a multi-output deep neural network model;
s3: using said TROPOMI-NO 2 The satellite remote sensing data and the basic data related to the environment perform initial training on the multi-output deep neural network model;
s4: by using the CO 2 The satellite remote sensing data and the basic data related to the environment perform secondary training on the multi-output deep neural network model after the initial training;
s5: CO is performed by utilizing the basic data related to the environment and the multi-output deep neural network model after the secondary training 2 Predicting the space-time distribution to obtain CO 2 And reconstructing a result of the space-time distribution.
As a further description of the present invention,
the context-dependent underlying data includes: CO 2 Emission data, population density data, altitude data, land utilization data, normalized vegetation index, and meteorological data; the meteorological data includes: surface temperature, surface air pressure, wind speed, wind direction, relative humidity, and planet boundary layer height.
As a further description of the present invention, the step S2 includes:
s11: carrying out 1km gridding treatment and standardization treatment on all data in the environment database;
s12: treating the treated sample with S11 to obtain a sample containing TROPOMI-NO 2 The grid of the satellite remote sensing data is divided into an initial training set and an initial testing set, and the initial training set and the initial testing set are processedThe treated S11 contains CO 2 The grid of satellite remote sensing data is divided into a secondary training set and a secondary testing set.
As a further description of the present invention, the step S2 includes the steps of:
establishing a first deep neural network model and a second deep neural network model, wherein the first deep neural network model and the second deep neural network model have the same model structure and comprise the following steps: a Dense layer, a Batch Normalization layer, an activation function and a Dropout layer;
establishing a data transmission chain corresponding to each layer of the first deep neural network model and the second deep neural network model, wherein the direction of the data transmission chain is from the first deep neural network model to the second deep neural network model;
and connecting the input end of the first depth neural network model and the input end of the second depth neural network model with the output end of the self-encoder.
As a further description of the present invention, the step S3 includes the steps of:
s31: inputting the basic data related to the environment into the multi-output deep neural network model by taking the initial training set as an independent variable;
s32: deep learning the environment-related base data and the initial training set by using the first deep neural network model to fit the environment-related base data and the TROPOMI-NO 2 The initial nonlinear relation between the satellite remote sensing data is achieved, and the fitting result of each layer of the first depth neural network is sent to the corresponding layer of the second depth neural network;
s33: deep learning the environment-related base data, the initial training set and the data from the first deep neural network model by using the second deep neural network model to fit the environment-related base data and the TROPOMI-NO 2 A final nonlinear relationship between satellite remote sensing data.
As a further description of the present invention,
the step S32 includes the following steps:
performing dimension reduction processing on the basic data related to the environment and the initial training set by using the self-encoder;
performing cross validation on the initial training set;
the step S33 includes the following steps:
and testing the multi-output deep neural network model after the initial training by using the initial test set.
As a further description of the present invention, the S4 includes:
s41: inputting the basic data related to the environment into the multi-output deep neural network model by taking the secondary training set as an independent variable;
s42: deep learning the environment-related basic data and the secondary training set by using only the second deep neural network model trained by the S33, and fitting the environment-related basic data and the CO 2 Nonlinear relationships between satellite remote sensing data.
As a further description of the present invention,
before S42, the method includes the following steps: performing dimension reduction processing on the basic data related to the environment and the secondary training set by using the self-encoder;
after S42, the method includes the following steps: and testing the multi-output deep neural network model after the secondary training by using the secondary test set.
In another aspect, the present invention provides a CO 2 A spatio-temporal distribution reconstruction system comprising:
a database creation module for creating a database containing TROPOMI-NO 2 Satellite remote sensing data, CO 2 An environmental database of satellite remote sensing data and environmental related underlying data;
the model creation module is used for creating a multi-output deep neural network model;
initial trainingA module for utilizing the TROPOMI-NO 2 The satellite remote sensing data and the basic data related to the environment perform initial training on the multi-output deep neural network model;
a secondary training module for utilizing the CO 2 The satellite remote sensing data and the basic data related to the environment perform secondary training on the multi-output deep neural network model after the initial training;
model prediction module for using the basic data related to the environment and the multi-output deep neural network model after the secondary training to perform CO 2 Predicting the space-time distribution to obtain CO 2 And reconstructing a result of the space-time distribution.
As a further description of the invention, the system further comprises:
the data processing module is used for carrying out 1km gridding processing and standardization processing on all data in the environment database and containing CO after processing 2 The grid of satellite remote sensing data is divided into a secondary training set and a secondary testing set.
The model creation module includes:
a first model creation unit for creating a first deep neural network model including a Dense layer, a Batch Normalization layer, an activation function and a Dropout layer, and connecting an input end of the first deep neural network model with an output end of a self-encoder;
a second model creation unit for creating a second deep neural network model having the same model structure as the first deep neural network model, and connecting an input terminal of the second deep neural network model with an output terminal of the self-encoder;
the data transmission link creation unit is used for creating a data transmission link corresponding to each layer of the first deep neural network model and the second deep neural network model, and the direction of the data transmission link is from the first deep neural network model to the second deep neural network model.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1. one aspect of the invention utilizes TROPOMI-NO 2 The satellite remote sensing data has the advantages of large quantity and can also be used as related information of fossil fuel combustion, and the satellite remote sensing data is introduced into CO 2 In the space-time distribution reconstruction, the side surface reflects human activity, which can solve the problem of CO in the prior art 2 Support of relevant data information is absent and CO is absent when reconstructing spatiotemporal distribution 2 The satellite remote sensing data acquired by the satellite has a small sample size, so that the problem of larger deviation of a reconstruction result is caused; on the other hand, by establishing a multiple-output deep neural network model and sequentially performing initial training and secondary training on the model, TROPOMI-NO is utilized 2 Satellite data reflects fossil fuel combustion information and NO 2 With CO 2 Emission homology characteristics, TROPOMI-NO 2 Information of representative fossil fuel combustion is given to a model to realize CO 2 Reconstruction of high spatial-temporal resolution spatial-temporal distribution.
2. According to the invention, the dimension reduction processing is carried out on the initial independent variable dimension through the self-encoder part, so that the data learning property is improved;
3. the invention utilizes the transfer learning method to effectively combine the related information of TROPOMI-NO2 and CO2 on the space-time distribution, and the TROPOMI-NO 2 Data is used as auxiliary data to solve the problem of CO 2 Data sparseness problem and increase CO 2 Accuracy of concentration prediction results.
Drawings
In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the drawings that are needed in the examples will be briefly described below, it being understood that the following drawings only illustrate some examples of the present invention and therefore should not be considered as limiting the scope, and that other related drawings may be obtained from these drawings without inventive effort for a person skilled in the art.
FIG. 1 shows a CO according to an embodiment of the present invention 2 A flow diagram of a space-time distribution reconstruction method;
FIG. 2 is a schematic diagram of a multiple-output deep neural network model according to an embodiment of the present invention;
FIG. 3 is a schematic illustration of a CO process according to an embodiment of the present invention 2 And a space-time distribution reconstruction system structure schematic diagram.
Detailed Description
For the purpose of making apparent the objects, technical solutions and advantages of the present invention, the present invention will be further described in detail with reference to the following examples and the accompanying drawings, wherein the exemplary embodiments of the present invention and the descriptions thereof are for illustrating the present invention only and are not to be construed as limiting the present invention.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, it will be apparent to one of ordinary skill in the art that: no such specific details are necessary to practice the invention. In other instances, well-known structures, circuits, materials, or methods have not been described in detail in order not to obscure the invention.
Throughout the specification, references to "one embodiment," "an embodiment," "one example," or "an example" mean: a particular feature, structure, or characteristic described in connection with the embodiment or example is included within at least one embodiment of the invention. Thus, the appearances of the phrases "in one embodiment," "in an example," or "in an example" in various places throughout this specification are not necessarily all referring to the same embodiment or example. Furthermore, the particular features, structures, or characteristics may be combined in any suitable combination and/or sub-combination in one or more embodiments or examples. Moreover, those of ordinary skill in the art will appreciate that the illustrations provided herein are for illustrative purposes and that the illustrations are not necessarily drawn to scale. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.
In the description of the present invention, the terms "front", "rear", "left", "right", "upper", "lower", "vertical", "horizontal", "high", "low", "inner", "outer", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, merely to facilitate description of the present invention and simplify description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the scope of the present invention.
Example 1
Due to the prior art on the CO 2 Lack of support for relevant data information when reconstructing spatiotemporal distribution, and CO 2 Satellites are subject to interference from external conditions when acquiring data, resulting in CO 2 The problem of large deviation in the reconstruction process is solved, and the embodiment provides a CO 2 A method for reconstructing space-time distribution is shown in figure 1, and uses TROPOMI-NO 2 The satellite data of (2) is used as related information of fossil fuel combustion, and the CO is realized based on a transfer learning method of shared parameters 2 Reconstruction of the spatial-temporal distribution of the concentration global domain. The main modeling process is to use TROPOMI-NO 2 Pre-training a deep learning model as a dependent variable to fit the model to the independent variable and TROPOMI-NO 2 Nonlinear relation between them by CO 2 As a dependent variable, the model after the pre-training is finally trained. By the method, the model reconstructs TROPOMI-NO with higher accuracy 2 At the same time of high space-time resolution data set, TROPOMI-NO 2 Information of representative fossil fuel combustion is given to a model to realize CO 2 Reconstruction of high spatial-temporal resolution spatial-temporal distribution. The implementation steps are as follows:
s1: establishing an environment database, wherein the environment database comprises: TROPOMI-NO 2 Satellite remote sensing data, CO 2 Satellite remote sensing data and environmental related basis data. Wherein the context-dependent underlying data includes: CO 2 Emission data, population density data, altitude data, land utilization data, normalized vegetation index, and meteorological data; the meteorological data includes: surface temperature, surface air pressure, wind speed, wind direction, relative humidity, and planet boundary layer height.
By means of TROPOMI-NO 2 The satellite remote sensing data has the advantages of large quantity and can also be used as related information of fossil fuel combustion, and the satellite remote sensing data is introduced into CO 2 In the space-time distribution reconstruction, the side surface reflects human activity, which can solve the problem of CO in the prior art 2 Support of relevant data information is absent and CO is absent when reconstructing spatiotemporal distribution 2 The problem of larger deviation of a reconstruction result is caused by small sample size of satellite remote sensing data acquired by a satellite, and CO is accurately and fully covered 2 The concentration prediction result can provide basis for carbon emission statistics accounting and data support for the establishment of carbon reduction policies.
S2: and carrying out 1km gridding processing and standardization processing on all data in the environment database. The method specifically comprises the following steps:
s21: adopting an area weighted average method for population density data, altitude data, land utilization data and normalized vegetation index variable to convert the spatial resolution into a 1km grid;
s22: the CO2 emission list and meteorological data were resampled to a 1km grid using high Cheng Xieke Rich interpolation.
S23: after the data are processed to the same spatial scale, the data are respectively standardized, so that each independent variable is in the same data scale, and training of the deep learning model is facilitated.
S24: for TROPOMI-NO 2 Satellite remote sensing data and CO 2 The satellite remote sensing data is subjected to 1km gridding treatment, and TROPOMI-NO is contained after the treatment of the steps (1) to (3) 2 The grid of the satellite remote sensing data is divided into an initial training set and an initial testing set; CO-containing after the treatment of the steps (1) to (3) 2 The grid of satellite remote sensing data is divided into a secondary training set and a secondary testing set.
S3: and establishing a multi-output deep neural network model. The method comprises the following steps:
s31: establishing a first deep neural network model and a second deep neural network model, wherein the first deep neural network model and the second deep neural network model have the same model structure and comprise the following steps: a Dense layer, a Batch Normalization layer, an activation function and a Dropout layer;
s32: establishing a data transmission chain corresponding to each layer of the first deep neural network model and the second deep neural network model, wherein the direction of the data transmission chain is from the first deep neural network model to the second deep neural network model;
s33: and connecting the input end of the first depth neural network model and the input end of the second depth neural network model with the output end of the self-encoder.
The deep learning model used is a multi-output depth neural network based on a self-encoder, and as shown in fig. 2, the model is composed of three modules, including the self-encoder and two sub-deep learning modules.
The model utilizes a self-encoder to reduce the dimension of the self-variable quantity, and the embodiment utilizes a first deep neural network model and a second deep neural network model which are respectively constructed by the dimension-reduced variables. The fully connected layers in the first deep neural network model and the second deep neural network model contain the most basic Dense layers of the neural network and use Batch Normalization, an activation function and Dropout. And unidirectional connection exists between the two deep neural network models, namely, each layer of output of the first deep neural network model is added with each layer of output of the second deep neural network model to serve as input data of the next layer of the second deep neural network model.
S34: the invention uses the joint loss function to optimize the model, the loss function corresponding to each output is based on root mean square error (Mean Squared Error, MSE), and the specific function expression is as follows:
output one:
Loss min ＝min(Loss 2 ，Loss 3 )
and output two:
and output three:
and (3) overall output: loss (Low Density) all ＝λ min Loss 1 +(1-λ)Loss 2 +λLoss 3
λ min ＝min(λ，1-λ)
Wherein:
model predictive values of output one, output two and output three respectively; y is 1 、y 2 、y 3 The real values corresponding to the output I, the output II and the output III are respectively; lambda, (1-lambda) and lambda min For the weight applied to each loss function, the value range is (0, 1), and ω is the weight value of the output of the last layer neural network node. Setting the emphasis point of model training by adjusting the value of lambda, when lambda < 0.5, loss in the model 2 The training process occupies larger weight; when lambda > 0.5, loss 3 And takes up more weight in the training process. By lambda min For Loss of 1 Regulated lambda min Is a smaller value between lambda and 1-lambda and is at Loss 1 Middle binding Loss 2 And Loss of 3 A smaller value of Loss in between min . When Loss min When the value is large, the Loss is large 1 The self-encoder module has higher weight in the gradient descent process, so that the self-encoder module can quickly trend to the global optimal value; when Loss min When the value becomes smaller gradually, loss 1 The weight value is correspondingly reduced, so that the variation of the internal parameters of the self-encoder formed by input data and output one is smaller, a more stable independent variable dimension reduction result is provided for the sub-deep learning module (1) and the sub-deep learning module (2), and meanwhile, the L2 regularization is applied to the result of the output one by adding the two normals of omega, so that the self-encoder structure in the model has stronger generalization capability. The invention utilizes the integral Loss function Loss all And optimizing the model to update the internal parameters of the model.
S4: using said TROPOMI-NO 2 The satellite remote sensing data and the basic data related to the environment perform initial training on the multi-output deep neural network model. The method comprises the following steps:
s41: inputting the basic data related to the environment into the multi-output deep neural network model by taking the initial training set as an independent variable;
s42: performing dimension reduction processing on the basic data related to the environment and the initial training set by using the self-encoder;
s43: deep learning the environment-related base data and the initial training set by using the first deep neural network model to fit the environment-related base data and the TROPOMI-NO 2 The initial nonlinear relation between the satellite remote sensing data is achieved, and the fitting result of each layer of the first depth neural network is sent to the corresponding layer of the second depth neural network;
s44: deep learning the environment-related base data, the initial training set and the data from the first deep neural network model by using the second deep neural network model to fit the environment-related base data and the TROPOMI-NO 2 A final nonlinear relationship between satellite remote sensing data;
s45: and testing the multi-output deep neural network model after the initial training by using the initial test set.
For better generalization, the embodiment contains TROPOMI-NO 2 The grid of the satellite remote sensing data is divided into an initial training set and an initial testing set, and cross verification is adopted for the initial training set to determine the optimal parameters of the model; the initial test set is only used for evaluating the generalization capability of the model and does not participate in the pre-training process of the model.
When initial training is performed, the CO2 emission list, population density data, altitude data, land utilization data, normalized vegetation index and meteorological data are used as input data of a model. The output is input data, and the outputs of the first deep neural network model and the second deep neural network model are TROPOMI-NO 2 Satellite remote sensing data, thereby utilizing TROPOMI-NO 2 And the initial optimization of parameters in the model is completed due to the advantage of more data.
By initial training of the model, two different environment-related integrated data (hereinafter referred to as independent variables) and TROPOMI-NO are constructed by a first deep neural network model and a second deep neural network model respectively 2 Nonlinear relationships between satellite remote sensing data. The decoder portion and the first deep neural network model portion of the model reflect the independent variable and the TROPOMI-NO 2 The nonlinear relation between the satellite remote sensing data and the second deep neural network model is combined with TROPOMI-NO in the first deep neural network model 2 Construction of independent variables and TROPOMI-NO from related information of satellite remote sensing data 2 Nonlinear relationship of satellite remote sensing data.
In addition, the initial argument dimension is reduced by the self-encoder part, so that the learning property of the data is improved.
Independent variable and TROPOMI-NO 2 The relation between the satellite remote sensing data is as follows: y is Y TROPOMI-NO2 ＝f(x 1 ，x 2 …，x n )。
S5: by using the CO 2 And carrying out secondary training on the multi-output deep neural network model after the initial training by using the satellite remote sensing data and the basic data related to the environment. The method comprises the following steps:
s51: inputting the basic data related to the environment into the multi-output deep neural network model by taking the secondary training set as an independent variable;
s52: performing dimension reduction processing on the basic data related to the environment and the secondary training set by using the self-encoder;
s53: deep learning the environment-related basic data and the secondary training set by using only the second deep neural network model trained by the S33, and fitting the environment-related basic data and the CO 2 Nonlinear relations between satellite remote sensing data;
s54: and testing the multi-output deep neural network model after the secondary training by using the secondary test set.
This example will contain CO 2 SatelliteThe grid of the remote sensing data is divided into a secondary training set and a secondary testing set, wherein the secondary training set is used for determining the optimal parameters of the model, and the secondary testing set is only used for evaluating the generalization capability of the model.
And when the model is secondarily trained, freezing parameters of the self-encoder and the first depth neural network model in the model, so that the parameters of the self-encoder and the first depth neural network model are not changed in the secondary training process. The variable type of the input data of the model is the same as that of the initial training, and only the second deep neural network model participates in the training to construct independent variables and CO 2 Nonlinear relationship between the two.
Since the first deep neural network model is capable of providing TROPOMI-NO during the initial training process 2 The second deep neural network model can combine with TROPOMI-NO 2 Implementation of satellite remote sensing data on CO 2 Is an efficient prediction of (a). The second depth is the independent variable and CO constructed by the neural network model 2 Relationship between satellite remote sensing data: y is Y CO2 ＝f(x 1 ，x 2 …，x n ，Y TROPOMI-NO2 )。
S6: CO is performed by utilizing the basic data related to the environment and the multi-output deep neural network model after the secondary training 2 Predicting the space-time distribution to obtain CO 2 And reconstructing a result of the space-time distribution.
After the initial training and the secondary training of the model are completed, inputting independent variables in the area range into a trained multi-output deep neural network model, and outputting TROPOMI-NO with the spatial resolution of 1km and the time resolution of days through a first deep neural network model 2 The reconstruction result of the space-time distribution is used for outputting CO with the spatial resolution of 1km and the time resolution of day through a second deep neural network model 2 Reconstruction of the spatio-temporal distribution results, realizing TROPOMI-NO in the region 2 And CO 2 High resolution spatiotemporal distribution reconstruction.
To sum up, the embodiment provides a CO 2 Method for reconstructing spatiotemporal distribution in absence of humanIn the case of class activity direct correlation data, NO is utilized 2 Satellite remote sensing data TROPOMI-NO 2 Reflecting human activities laterally, reconstructing CO 2 High resolution spatiotemporal distribution, accurate full coverage of CO 2 The concentration prediction result can provide basis for carbon emission statistics accounting and data support for the establishment of carbon reduction policies. In addition, the method reduces the dimension of the initial independent variable dimension through the self-encoder part, thereby improving the data learning property. TROPOMI-NO using transfer learning method 2 And CO 2 Information on spatio-temporal distribution is effectively combined, TROPOMI-NO 2 Data is used as auxiliary data to improve CO 2 Accuracy of concentration prediction results.
Example 2
This example provides a CO as described in example 1 2 The corresponding system of the space-time distribution reconstruction method, as shown in fig. 3, comprises:
a database creation module for creating a database containing TROPOMI-NO 2 Satellite remote sensing data, CO 2 An environmental database of satellite remote sensing data and environmental related underlying data;
the model creation module is used for creating a multi-output deep neural network model;
an initial training module for utilizing the TROPOMI-NO 2 The satellite remote sensing data and the basic data related to the environment perform initial training on the multi-output deep neural network model;
a secondary training module for utilizing the CO 2 The satellite remote sensing data and the basic data related to the environment perform secondary training on the multi-output deep neural network model after the initial training;
model prediction module for using the basic data related to the environment and the multi-output deep neural network model after the secondary training to perform CO 2 Predicting the space-time distribution to obtain CO 2 And reconstructing a result of the space-time distribution.
The data processing module is used for carrying out 1km gridding processing and standardization processing on all data in the environment database, and carrying out processingContains CO 2 The grid of satellite remote sensing data is divided into a secondary training set and a secondary testing set.
Wherein, the model creation module includes:
a first model creation unit for creating a first deep neural network model including a Dense layer, a Batch Normalization layer, an activation function and a Dropout layer, and connecting an input end of the first deep neural network model with an output end of a self-encoder;
a second model creation unit for creating a second deep neural network model having the same model structure as the first deep neural network model, and connecting an input terminal of the second deep neural network model with an output terminal of the self-encoder;
a data transmission link creation unit for creating a data transmission link corresponding to each layer of the first deep neural network model and the second deep neural network model, wherein the direction of the data transmission link is from the first deep neural network model to the second deep neural network model
The foregoing description of the embodiments has been provided for the purpose of illustrating the general principles of the invention, and is not meant to limit the scope of the invention, but to limit the invention to the particular embodiments, and any modifications, equivalents, improvements, etc. that fall within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (7)
1. CO (carbon monoxide) 2 The space-time distribution reconstruction method is characterized by comprising the following steps of:
s1: establishing an environment database, wherein the environment database comprises: TROPOMI-NO 2 Satellite remote sensing data, CO 2 Satellite remote sensing data and environmental related base data;
s2: establishing a multi-output deep neural network model;
s3: using said TROPOMI-NO 2 Satellite remote sensing data and said environment-related basis data are deep for said multiple outputsInitial training is carried out on the neural network model;
s4: by using the CO 2 The satellite remote sensing data and the basic data related to the environment perform secondary training on the multi-output deep neural network model after the initial training;
s5: CO is performed by utilizing the basic data related to the environment and the multi-output deep neural network model after the secondary training 2 Predicting the space-time distribution to obtain CO 2 Reconstructing a result of space-time distribution;
the step S3 comprises the following steps:
s31: inputting the basic data related to the environment into the multi-output deep neural network model by taking the initial training set as an independent variable;
s32: deep learning the environment-related base data and the initial training set by using a first deep neural network model to fit the environment-related base data and the TROPOMI-NO 2 The initial nonlinear relation between the satellite remote sensing data is achieved, and the fitting result of each layer of the first depth neural network is sent to the corresponding layer of the second depth neural network;
s33: deep learning the environment-related base data, the initial training set and the data from the first deep neural network model by using the second deep neural network model to fit the environment-related base data and the TROPOMI-NO 2 A final nonlinear relationship between satellite remote sensing data.
2. A CO according to claim 1 2 A method for reconstructing space-time distribution is characterized in that,
the context-dependent underlying data includes: CO 2 Emission data, population density data, altitude data, land utilization data, normalized vegetation index, and meteorological data; the meteorological data includes: surface temperature, surface air pressure, wind speed, wind direction, relative humidity, and planet boundary layer height.
3. A CO according to claim 2 2 The space-time distribution reconstruction method is characterized in that the method comprises the following steps:
s11: carrying out 1km gridding treatment and standardization treatment on all data in the environment database;
s12: treating the treated sample with S11 to obtain a sample containing TROPOMI-NO 2 The grid of the satellite remote sensing data is divided into an initial training set and an initial testing set, and the grid of the satellite remote sensing data is processed by the S11 and contains CO 2 The grid of satellite remote sensing data is divided into a secondary training set and a secondary testing set.
4. A CO according to claim 3 2 The space-time distribution reconstruction method is characterized in that the S2 comprises the following steps:
establishing a first deep neural network model and a second deep neural network model, wherein the first deep neural network model and the second deep neural network model have the same model structure and comprise the following steps: a Dense layer, a Batch Normalization layer, an activation function and a Dropout layer;
establishing a data transmission chain corresponding to each layer of the first deep neural network model and the second deep neural network model, wherein the direction of the data transmission chain is from the first deep neural network model to the second deep neural network model;
and connecting the input end of the first depth neural network model and the input end of the second depth neural network model with the output end of the self-encoder.
5. A CO according to claim 4 2 The space-time distribution reconstruction method is characterized by comprising the following steps before the step S32:
performing dimension reduction processing on the basic data related to the environment and the initial training set by using the self-encoder;
performing cross validation on the initial training set;
the step S33 includes the following steps:
and testing the multi-output deep neural network model after the initial training by using the initial test set.
6. A CO according to claim 4 or 5 2 The space-time distribution reconstruction method is characterized in that the S4 comprises the following steps:
s41: inputting the basic data related to the environment into the multi-output deep neural network model by taking the secondary training set as an independent variable;
s42: deep learning the environment-related basic data and the secondary training set by using only the second deep neural network model trained by the S33, and fitting the environment-related basic data and the CO 2 Nonlinear relationships between satellite remote sensing data.
7. A CO according to claim 6 2 A method for reconstructing space-time distribution is characterized in that,
before S42, the method includes the following steps: performing dimension reduction processing on the basic data related to the environment and the secondary training set by using the self-encoder;
after S42, the method includes the following steps: and testing the multi-output deep neural network model after the secondary training by using the secondary test set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210490951.6A CN114861882B (en) | 2022-05-07 | 2022-05-07 | CO (carbon monoxide) 2 Space-time distribution reconstruction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210490951.6A CN114861882B (en) | 2022-05-07 | 2022-05-07 | CO (carbon monoxide) 2 Space-time distribution reconstruction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114861882A CN114861882A (en) | 2022-08-05 |
| CN114861882B true CN114861882B (en) | 2023-05-09 |
Family
ID=82636115
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210490951.6A Active CN114861882B (en) | 2022-05-07 | 2022-05-07 | CO (carbon monoxide) 2 Space-time distribution reconstruction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114861882B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115310550B (en) * | 2022-08-16 | 2023-07-14 | 国网四川省电力公司电力科学研究院 | Atmospheric carbon dioxide dry air column concentration calculation method and system |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529081A (en) * | 2016-12-03 | 2017-03-22 | 安徽新华学院 | PM2.5 real-time level prediction method and system based on neural net |
| US20210209424A1 (en) * | 2020-01-06 | 2021-07-08 | Quantela Inc | Computer-based method and system for predicting and generating land use land cover (lulc) classification |
| CN113128780A (en) * | 2021-04-29 | 2021-07-16 | 平安普惠企业管理有限公司 | PM2.5 concentration prediction method and device, computer equipment and storage medium |
| CN113297528B (en) * | 2021-06-10 | 2022-07-01 | 四川大学 | NO based on multi-source big data2High-resolution space-time distribution calculation method |
| CN114266200B (en) * | 2022-02-24 | 2022-07-05 | 山东大学 | Nitrogen dioxide concentration prediction method and system |
-
2022
- 2022-05-07 CN CN202210490951.6A patent/CN114861882B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN114861882A (en) | 2022-08-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Charles et al. | A spatiotemporal model for downscaling precipitation occurrence and amounts | |
| CN111199270B (en) | Regional wave height forecasting method and terminal based on deep learning | |
| CN114239422B (en) | Method for improving marine chlorophyll a concentration prediction accuracy based on machine learning | |
| CN112287294B (en) | Space-time bidirectional soil water content interpolation method based on deep learning | |
| CN112651665A (en) | Surface water quality index prediction method and device based on graph neural network | |
| CN114861882B (en) | CO (carbon monoxide) 2 Space-time distribution reconstruction method and system | |
| CN112052627A (en) | Method, device, medium and equipment for estimating near-surface ozone space distribution | |
| CN113516304B (en) | Regional pollutant space-time joint prediction method and device based on space-time diagram network | |
| CN114611608A (en) | Sea surface height numerical value prediction deviation correction method based on deep learning model | |
| Li et al. | Stochastic deep gaussian processes over graphs | |
| Pauthenet et al. | Four-dimensional temperature, salinity and mixed-layer depth in the Gulf Stream, reconstructed from remote-sensing and in situ observations with neural networks | |
| Xu et al. | Integrating an option-oriented attitude analysis into investigating the degree of stabilities in conflict resolution | |
| CN116105697A (en) | Deep learning-based satellite-borne GNSS-R global sea surface effective wave height inversion method | |
| CN116029419A (en) | Deep learning-based long-term new energy daily average generation power prediction method and system | |
| Zhang et al. | SolarGAN: Synthetic annual solar irradiance time series on urban building facades via Deep Generative Networks | |
| CN117543537A (en) | Agent electricity purchasing user electric quantity prediction method, device and storage medium | |
| CN112989557A (en) | Method for improving water reserve change prediction reliability based on neural network selectable model | |
| CN117114168A (en) | Long-time-scale sea surface temperature intelligent forecasting method | |
| CN111811465A (en) | Method for predicting sea wave effective wave height based on multi-sine function decomposition neural network | |
| CN117198540A (en) | Space-time simulation method for zoonosis based on geographic space big data | |
| CN115759291A (en) | Space nonlinear regression method and system based on ensemble learning | |
| Delventhal | The globe as a network: Geography and the origins of the world income distribution | |
| CN116401939A (en) | North sea ice short-term forecasting method based on gradient constraint neural network | |
| CN115330082A (en) | PM2.5 concentration prediction method of LSTM network based on attention mechanism | |
| CN115540832A (en) | Satellite altimetry submarine topography correction method and system based on VGGNet |
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 |