A PM with particle diameter restriction 2.5 Deep Learning Remote Sensing Estimation Method

Abstract

The present invention discloses a PM deep learning remote sensing estimation method with particle diameter constraint, which constructs a deep neural network model based on satellite remote sensing aerosol optical depth (AOD), reanalysis data, ground station observation data, etc., and utilizes the prior relationship of the particle diameter sizes of PM1, PM 2.5 , and PM 2.5 (i.e., PM1 concentration ≤ PM 10 concentration ≤ PM 2.5 concentration) to establish a constraint condition, and constructs a constraint term of the deep neural network loss function based on this, and proposes a PM deep learning remote sensing estimation method with particle diameter constraint. The present invention can alleviate the phenomena of high-value underestimation and low-value overestimation occurring in the deep learning PM 10 estimation, improve the accuracy of PM 2.5 remote sensing estimation, and can be widely applied to the technical field of PM 2.5 concentration remote sensing estimation. 2.5 The present invention can alleviate the phenomena of high-value underestimation and low-value overestimation occurring in the deep learning PM 2.5 estimation, improve the accuracy of PM remote sensing estimation, and can be widely applied to the technical field of PM concentration remote sensing estimation.

Description

Technical Field

[0001] The present invention relates to PM 2.5 The field of remote sensing concentration estimation technology, especially a PM2.5 2.5 Deep learning methods for remote sensing estimation. Background Art

[0002] Fine particulate matter (PM 2.5 ) refers to particles with a dynamic diameter of less than or equal to 2.5 microns in the ambient air. 2.5 In such an environment, the respiratory system, cardiovascular system, nervous system, etc. of the human body will be greatly harmed. 2.5 Dynamic monitoring research is of great significance. At present, my country's ambient air automatic monitoring stations are sparsely distributed and cannot accurately and effectively monitor PM in a large area. 2.5 Therefore, satellite remote sensing data are used to estimate large-scale PM 2.5 Concentration has important academic significance and application value.

[0003] PM based on physical and chemical models 2.5 Remote sensing estimation methods have always attracted widespread attention. This type of method has a clear mechanism and is highly interpretable. However, this type of method requires the acquisition of a large number of physical and chemical parameters, and its estimation accuracy is often limited. In addition, statistical models and machine learning methods only need to mine relationship models from a large number of ground station observation data and related variables to obtain higher estimation accuracy. With the rapid development of ground station monitoring networks, statistical models and machine learning methods are favored by more researchers. Among these methods, deep learning (deep neural networks) has very powerful nonlinear fitting capabilities and high prediction accuracy, so it is very popular in atmospheric PM2.5. 2.5 It has been widely used in remote sensing estimation.

[0004] Although deep learning models have achieved high estimation accuracy, they are prone to underestimation of high values ​​and overestimation of low values, and even lead to the estimation of PM 2.5 Concentrations higher than PM 10 Or less than PM1 and other issues that are inconsistent with prior knowledge. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a PM with particle diameter restriction. 2.5 Deep learning remote sensing estimation method can reduce deep learning estimation PM 2.5 Abnormal values ​​appear, alleviating PM 2.5 The phenomenon of overestimation of low values ​​and underestimation of high values ​​in the deep learning estimation process is solved to improve the robustness of the deep learning model.

[0006] A first aspect of an embodiment of the present invention provides a PM with particle diameter restriction. 2.5 Deep learning remote sensing estimation methods, including:

[0007] Obtain site observation data, reanalysis data, satellite remote sensing data and auxiliary data. The site observation data includes PM 2.5 Site Data and PM 10 Site data;

[0008] According to the reanalysis data, combined with the PM 2.5 Station data, obtain PM1 equivalent observation data;

[0009] According to the PM1 equivalent observation data, the PM 2.5 Site data and the PM 10 The site data is combined with the prior relationship of particle diameter size to determine the neural network loss function of particle diameter constraint;

[0010] Based on the deep neural network, combined with the neural network loss function, PM is constructed 2.5 Estimation model;

[0011] According to the sample library, the PM 2.5 The estimation model is trained and optimized to determine the PM 2.5 Estimate the optimal parameters of the model;

[0012] The reanalysis data, the satellite remote sensing data and the auxiliary data are input into the PM 2.5 Estimation model, get PM 2.5 Estimation results.

[0013] Optionally, the obtaining of reanalysis data includes at least one of the following:

[0014] Obtain weather condition data;

[0015] Get air quality data.

[0016] Optionally, the obtaining of air quality data includes at least one of the following:

[0017] Get PM1 concentration;

[0018] Get PM 2.5 concentration.

[0019] Optionally, the reanalysis data is combined with the PM 2.5 Site data, obtain PM1 equivalent observation data, including:

[0020] According to the PM1 concentration and PM 2.5The ratio of the concentration of PM 2.5 The station data is used to calculate the PM1 equivalent observation value to obtain the PM1 equivalent observation data. The expression of the PM1 equivalent observation is:

[0021]

[0022] Among them, PM1 represents PM1 equivalent observation data, PM1 reanalysis data represents PM1 concentration obtained from reanalysis data, PM 2.5 Reanalysis data refers to PM obtained from reanalysis data. 2.5 Concentration, PM 2.5 Indicates the PM observed at the site 2.5 data.

[0023] Optionally, the PM1 equivalent observation data, the PM 2.5 Site data and the PM 10 The site data is combined with the prior relationship of particle diameter size to determine the neural network loss function of particle diameter constraint, including:

[0024] According to the PM1 equivalent observation data, the PM 2.5 Site data and the PM 10 Site data, combined with PM1, PM 2.5 and PM 10 The a priori relationship between the concentration values ​​of 2.5 Constraints on remote sensing estimates;

[0025] According to the constraint conditions, the neural network loss function of the particle diameter constraint is determined, and the expression of the neural network loss function is:

[0026]

[0027] Where N is the number of samples, i represents the i-th sample, PM 2.5 Represents the PM obtained from site observation data 2.5 data, Indicates PM 2.5 PM estimated by deep learning model 2.5 data, λ1 and λ2 represent regularization parameters, ψ(PM1) represents the PM1 constraint term, and ψ(PM 10 ) means PM 10 Constraint term, ReLU() represents the linear rectification function, PM1 represents PM1 equivalent observation data, PM 10 Represents the PM obtained from site observations 10 data.

[0028] Optionally, the deep neural network is combined with the neural network loss function to construct PM 2.5 Estimation models, including:

[0029] A deep neural network is constructed according to the satellite remote sensing data and the auxiliary data. The expression of the deep neural network is:

[0030] PM 2.5 =f(AOD,x1,x2,x3,...),

[0031] Among them, PM 2.5 PM represents the PM estimated by the deep neural network 2.5 data; f() represents a deep neural network, AOD represents satellite remote sensing aerosol optical thickness data, and x1, x2, x3,... represent auxiliary data.

[0032] Optionally, the PM 2.5 The estimation model is trained and optimized to determine the PM 2.5 Estimate the optimal parameters of the model, including:

[0033] Dividing the sample library into data sets;

[0034] The divided sample library includes a training set, a validation set and a test set.

[0035] Optionally, the PM 2.5 The estimation model is trained and optimized to determine the PM 2.5 Estimate the optimal parameters of the model, including:

[0036] According to the training set and the validation set, the PM is determined by cross validation. 2.5 Estimate the optimal parameters of the model;

[0037] Based on the optimal parameters, the PM is evaluated in combination with the test set. 2.5 Estimate the performance of the model.

[0038] Optionally, the reanalysis data, the satellite remote sensing data and the auxiliary data are input into the PM after determining the optimal parameters. 2.5 Estimation model, get PM 2.5 Estimated results include:

[0039] The input reanalysis data, the satellite remote sensing data and the auxiliary data are used to calculate the PM 2.5 Estimating the model does a forward pass.

[0040] A second aspect of the present invention provides a PM with particle diameter restriction. 2.5Deep learning remote sensing estimation system, including:

[0041] The first module is used to obtain site observation data, reanalysis data, satellite remote sensing data and auxiliary data. The site observation data includes PM 2.5 Site Data and PM 10 Site data;

[0042] The second module is used to combine the PM 2.5 Station data, obtain PM1 equivalent observation data;

[0043] The third module is used to calculate the PM1 equivalent observation data and the PM 2.5 Site data and the PM 10 The site data is combined with the prior relationship of particle diameter size to determine the neural network loss function of particle diameter constraint;

[0044] The fourth module is used to construct PM based on deep neural network and combined with the neural network loss function. 2.5 Estimation model;

[0045] The fifth module is used to analyze the PM according to the sample library. 2.5 The estimation model is trained and optimized to determine the PM 2.5 Estimate the optimal parameters of the model;

[0046] The sixth module is used to input the reanalysis data, the satellite remote sensing data and the auxiliary data into the PM 2.5 Estimation model, get PM 2.5 Estimation results.

[0047] The embodiment of the present invention also discloses a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device can read the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the above method.

[0048] The embodiment of the present invention is based on site observation data, reanalysis data, satellite remote sensing data and auxiliary data, and combines PM1, PM 2.5 and PM 10 The prior relationship between the diameters of the three particles is used to construct the constraint term of the deep neural network loss function, and finally the optimal PM is obtained through sample library training. 2.5 Estimation model, realizing PM with particle diameter constraints 2.5 Deep learning remote sensing estimation method. The present invention can reduce deep learning estimation PM 2.5Outliers that appear further alleviate the deep learning PM 2.5 The high-value underestimation and low-value overestimation phenomena in the estimation improve the robustness of the deep learning model and improve PM 2.5 Accuracy of remote sensing estimates. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0050] Figure 1 PM with particle diameter restriction provided by the embodiment of the present invention 2.5 Schematic diagram of the process of deep learning remote sensing estimation method;

[0051] Figure 2 A schematic diagram of a process for constructing a deep learning model for particle diameter constraints provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0053] In order to make the content and technical solution of this application clearer, the relevant terms and their meanings are explained:

[0054] PM1: A general term for solid particles or liquid droplets with an aerodynamic equivalent diameter less than or equal to 1 micron in the ambient air, also known as respirable particulate matter.

[0055] PM 2.5 : Generally refers to fine particulate matter, which means particles in the ambient air with an aerodynamic equivalent diameter of less than or equal to 2.5 microns.

[0056] PM 10 : Usually refers to particles in the ambient air with an aerodynamic equivalent diameter less than or equal to 10 microns, also known as inhalable particles or floating dust.

[0057] The technical solution of the present invention is used to alleviate PM 2.5 The phenomenon of underestimation of high values ​​and overestimation of low values ​​in remote sensing estimation using deep learning.

[0058] The implementation principle of the method of the present invention is described in detail below in conjunction with the accompanying drawings:

[0059] The embodiment of the present invention provides a PM with particle diameter restriction. 2.5 Deep learning remote sensing estimation methods, Figure 1 The PM with particle diameter restriction provided by the embodiment of the present invention is shown as follows: 2.5 Schematic diagram of the process of deep learning remote sensing estimation method, including:

[0060] Obtain site observation data, reanalysis data, satellite remote sensing data and auxiliary data. Site observation data include PM 2.5 Site Data and PM 10 Site data;

[0061] It should be noted that satellite remote sensing data is the aerosol optical thickness (AOD) obtained based on satellite remote sensing. Auxiliary data include population data and terrain data, etc. Site observations are obtained by collecting ground site observation data, among which PM 2.5 The data are PM observed at ground stations. 2.5 Concentration, PM 10 The data are PM observed at ground stations. 10 concentration.

[0062] According to the reanalysis data, combined with PM 2.5 Station data, obtain PM1 equivalent observation data;

[0063] According to PM1 equivalent observation data, PM 2.5 Site Data and PM 10 The site data is combined with the prior relationship of particle diameter size to determine the neural network loss function of particle diameter constraint;

[0064] Based on deep neural network, combined with neural network loss function, PM is constructed 2.5 Estimation model;

[0065] According to the sample library, PM 2.5 Estimation model training and optimization to determine PM 2.5 Estimate the optimal parameters of the model;

[0066] Reanalysis data, satellite remote sensing data and auxiliary data were input into the PM after the optimal parameters were determined. 2.5 Estimation model, get PM 2.5 Estimation results.

[0067] Specifically, the method of the embodiment of the present invention is based on satellite remote sensing aerosol optical thickness (AOD), reanalysis data, ground station observation data and auxiliary data, and based on PM1, PM 2.5 and PM 10 The prior relationship between the diameters of the three particles is used to construct the constraint term of the deep neural network loss function, and finally the optimal PM is obtained through sample library training.2.5 Estimation model, realizing PM with particle diameter constraints 2.5 Deep learning methods for remote sensing estimation.

[0068] In some embodiments, obtaining reanalysis data includes at least one of the following:

[0069] Obtain weather condition data;

[0070] Get air quality data.

[0071] It should be noted that meteorological conditions data include meteorological-related data such as wind speed, temperature, humidity and pressure.

[0072] In some embodiments, obtaining air quality data includes at least one of the following:

[0073] Get PM1 concentration;

[0074] Get PM 2.5 concentration.

[0075] In some embodiments, based on the reanalysis data, combined with PM 2.5 Site data, obtain PM1 equivalent observation data, including:

[0076] According to the PM1 concentration and PM 2.5 The proportional relationship of concentration, combined with PM 2.5 The station data is used to calculate the PM1 equivalent observation value and obtain the PM1 equivalent observation data. The expression of PM1 equivalent observation is:

[0077]

[0078] Among them, PM1 represents PM1 equivalent observation data, PM1 reanalysis data represents PM1 concentration obtained from reanalysis data, PM 2.5 Reanalysis data refers to PM obtained from reanalysis data. 2.5 Concentration, PM 2.5 Indicates the PM observed at the site 2.5 data.

[0079] In some embodiments, based on PM1 equivalent observation data, PM 2.5 Site Data and PM 10 The site data is combined with the prior relationship of particle diameter size to determine the neural network loss function of particle diameter constraint, including:

[0080] According to PM1 equivalent observation data, PM 2.5 Site Data and PM 10 Site data, combined with PM1, PM 2.5 and PM 10The a priori relationship between the concentration values ​​of 2.5 Constraints on remote sensing estimates;

[0081] According to the constraint conditions, the neural network loss function of the particle diameter constraint is determined. The expression of the neural network loss function is:

[0082]

[0083] Where N is the number of samples, i represents the i-th sample, PM 2.5 Represents the PM obtained from site observation data 2.5 data, Indicates PM 2.5 PM estimated by the estimation model 2.5 data, λ1 and λ2 represent regularization parameters, ψ(PM1) represents the PM1 constraint term, and ψ(PM 10 ) means PM 10 Constraint term, ReLU() represents the linear rectification function, PM1 represents the PM1 data obtained by PM1 equivalent observation, PM 10 Represents the PM obtained from site observation data 10 data.

[0084] It should be noted that PM1 concentration ≤ PM 2.5 Concentration ≤PM 10 The numerical relationship of concentration is known, and when the deep learning model estimates PM 2.5 Greater than PM1 and less than PM 10 When , it indicates that the estimated result is normal and the constraint item does not work (PM1 and PM 10 The constraint terms are all 0). When the PM estimated by the model 2.5 When it is less than PM1, the value of ψ(PM1) is positive, which constrains the model, that is, limiting PM 2.5 The lower bound of the estimated result (PM 2.5 ≥PM1), can effectively reduce PM 2.5 The phenomenon of underestimation. When the model estimates PM 2.5 Greater than PM 10 When ψ(PM 10 ) value is positive, which constrains the model, that is, limiting PM 2.5 The estimated upper bound (PM 2.5 ≤PM 10 ), which can effectively reduce PM 2.5 Overestimation phenomenon.

[0085] In some embodiments, based on a deep neural network and in combination with a neural network loss function, a PM is constructed. 2.5 Estimation models, including:

[0086] A deep neural network is constructed according to the satellite remote sensing data and the auxiliary data. The expression of the deep neural network is:

[0087] PM 2.5 =f(AOD,x1,x2,x3,...),

[0088] Among them, PM 2.5 PM represents the PM estimated by the deep neural network 2.5 data; f() represents a deep neural network, AOD represents satellite remote sensing data, and x1, x2, x3, ... represent auxiliary data.

[0089] It should be noted that the present invention has no obvious restrictions on deep neural networks. It mainly lies in how to construct a loss function constrained by particle diameter, and use the loss function as the optimization target, and use optimization algorithms such as stochastic gradient descent and Adam to perform model iterative training until the convergence condition or termination condition is reached (that is, building a deep neural network model with constraints).

[0090] In some embodiments, based on the sample library, PM 2.5 Estimation model training and optimization to determine PM 2.5 Estimate the optimal parameters of the model, including

[0091] Dividing the sample library into data sets;

[0092] The divided sample library includes a training set, a validation set and a test set.

[0093] In some embodiments, based on the sample library, PM 2.5 Estimation model training and optimization to determine PM 2.5 Estimate the optimal parameters of the model, including:

[0094] According to the training set and the validation set, the PM is determined by cross validation. 2.5 Estimate the optimal parameters of the model;

[0095] According to the best parameters, PM is evaluated on the test set. 2.5 Estimate the performance of the model.

[0096] In some embodiments, the reanalysis data, the satellite remote sensing data and the auxiliary data are input into the PM after determining the optimal parameters. 2.5 Estimation model, get PM 2.5 Estimated results include:

[0097] The input reanalysis data, the satellite remote sensing data and the auxiliary data are used to calculate the PM 2.5Estimating the model does a forward pass.

[0098] It should be noted that the satellite remote sensing data, reanalysis data and auxiliary data of the area to be estimated are input into the PM 2.5 In the estimation model, the input data is forwarded and the output result is calculated, namely: PM 2.5 concentration.

[0099] Specifically, in some embodiments, referring to Figure 2 The technical solution of the method embodiment of the present invention can be implemented by the following steps:

[0100] Step 1: Data collection, including ground station data, reanalysis data, remote sensing data and auxiliary data.

[0101] Remote sensing data refers to aerosol optical depth data (AOD); reanalysis data (such as ERA5) include temperature, wind speed, humidity, atmospheric boundary layer height, downwelling solar radiation, precipitation, PM1 and PM 2.5 etc.; Ground station data include PM 2.5 Data and PM 10 Data; auxiliary data include population density and terrain, etc. The collected data are preprocessed to remove outliers, and reprojected and resampled to keep the temporal and spatial resolutions consistent.

[0102] Step 2: PM1 and PM2.5 based on reanalysis data 2.5 The proportional relationship is based on the PM 2.5 Get PM1 concentration value.

[0103] At present, PM1 concentration data cannot be obtained from air quality monitoring stations, so the PM1 and PM 2.5 The proportional relationship is similar to the PM 2.5 Solve for the PM1 equivalent observation value, the formula is as follows:

[0104]

[0105] In the formula, PM1 represents the PM1 equivalent observation value, that is, the PM1 concentration, PM1 reanalysis data represents the PM1 concentration obtained from the reanalysis data, PM 2.5 Reanalysis data refers to PM obtained from reanalysis data. 2.5 Concentration, PM 2.5 Indicates the PM observed at the site 2.5 data.

[0106] Step 3: Based on PM1 and PM 2.5 With PM 10The a priori relationship between particle diameter size, that is, PM1 concentration ≤ PM 2.5 Concentration ≤PM 10 concentration, establish PM 2.5 The constraints in the remote sensing estimation process, the specific expression of the loss function (Loss) is:

[0107]

[0108] Where N is the number of samples, i represents the i-th sample, PM 2.5 Indicates the PM measured at the ground station 2.5 concentration, PM represents the PM estimated by the deep learning model 2.5 concentration, ψ(PM1) is the PM1 constraint term, ψ(PM 10 ) is PM 10 constraint term, λ1 and λ2 represent regularization parameters.

[0109] For the ψ(PM1) constraint, ReLU() is the linear rectification function, PM1 represents the PM1 equivalent observation obtained in step 2, PM represents the PM estimated by the deep learning model 2.5 concentration; for ψ(PM 10 ) Constraints, PM 10 Represents the PM measured at the ground-based site 10 When the deep learning model estimates PM 2.5 Greater than PM1 and less than PM 10 When , it indicates that the estimated result is normal and the constraint item does not work (PM1 and PM 10 The constraint terms are all 0). When the PM estimated by the model 2.5 When it is less than PM1, the value of ψ(PM1) is positive, which constrains the model, that is, limiting PM 2.5 The lower bound of the estimated result (PM 2.5 ≥PM1), can effectively reduce PM 2.5 The phenomenon of underestimation. When the model estimates PM 2.5 Greater than PM 10 When ψ(PM 10 ) value is positive, which constrains the model, that is, limiting PM 2.5 The estimated upper bound (PM 2.5 ≤PM 10 ), which can effectively reduce PM 2.5 Overestimation phenomenon.

[0110] Step 4: Build a deep neural network model and use the loss function designed in step 3 as the optimization target to construct a PM with particle diameter constraints. 2.5 Deep learning remote sensing estimation model.

[0111] First, a fully connected feedforward neural network (deep neural network) with multiple hidden layers is constructed, and its general expression is:

[0112] PM 2.5 =f(AOD,x1,x2,x3,...),

[0113] Where, PM 2.5 PM represents the PM estimated by the deep neural network 2.5 Data, f() refers to a deep neural network. It should be noted that the present invention has no obvious restrictions on deep learning models, but clarifies how to construct a loss function constrained by particle diameter (i.e., a deep neural network model with constraints). AOD is remote sensing aerosol optical thickness data, which refers to other auxiliary data. For the constructed deep neural network, the loss function designed in step 3 is used as the optimization target, and the model is iteratively trained using optimization algorithms such as stochastic gradient descent and Adam until the convergence condition or termination condition is reached.

[0114] Step 5: Use the sample library to train the model, find the best model parameters, and determine PM 2.5 Estimate the model.

[0115] First, the collected sample library is divided into training set, validation set and test set. Then, the optimal model parameters are selected by cross-validation, and the model accuracy is evaluated using the test set.

[0116] Step 6: Input satellite remote sensing data, reanalysis data and auxiliary data into the trained deep learning model to obtain PM 2.5 Estimation results.

[0117] Specifically, the satellite remote sensing data, reanalysis data and auxiliary data of the area to be estimated are input into the model, the input data is forwarded, and the output result is calculated, namely: PM 2.5 concentration.

[0118] The embodiment of the present invention provides a PM with particle diameter restriction. 2.5 Deep learning remote sensing estimation system, including:

[0119] The first module is used to obtain site observation data, reanalysis data, satellite remote sensing data and auxiliary data. Site observation data include PM 2.5 Site Data and PM 10 Site data;

[0120] The second module is used to combine PM 2.5 Station data, obtain PM1 equivalent observation data;

[0121] The third module is used to calculate the PM1 equivalent observation data and PM 2.5 Site Data and PM 10 The site data is combined with the prior relationship of particle diameter size to determine the neural network loss function of particle diameter constraint;

[0122] The fourth module is used to build PM based on deep neural network and combined with neural network loss function. 2.5 Estimation model;

[0123] The fifth module is used to analyze PM 2.5 Estimation model training and optimization to determine PM 2.5 Estimate the optimal parameters of the model;

[0124] The sixth module is used to input reanalysis data, satellite remote sensing data and auxiliary data into the PM 2.5 Estimation model, get PM 2.5 Estimation results.

[0125] The contents of the method embodiments of the present invention are all applicable to the system embodiments. The functions specifically implemented by the system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method.

[0126] The embodiment of the present invention also discloses a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device can read the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the above method.

[0127] In summary, the present invention estimates PM based on deep learning models 2.5 The concentration is prone to underestimate the high value and overestimate the low value, and this phenomenon leads to the estimated PM 2.5 Concentrations higher than PM 10 or less than PM1, which is inconsistent with prior knowledge, a PM2.5 method with particle diameter constraints is proposed. 2.5 Deep learning remote sensing estimation method, based on satellite remote sensing aerosol optical thickness (AOD), reanalysis data, ground station observation data, etc., builds a deep neural network model, using PM1, PM 2.5 and PM 10 The prior relationship between the diameters of the three particles is used to establish constraints, and based on this, the constraints of the deep neural network loss function are constructed, and finally the deep learning model of particle diameter constraints is constructed. This model can reduce the deep learning estimation of PM 2.5 Outliers that appear further alleviate the deep learning PM 2.5The high-value underestimation and low-value overestimation phenomena in the estimation improve the robustness of the deep learning model and improve PM 2.5 The accuracy of remote sensing estimates is 2.5 It has good application prospects in the field of estimation.

[0128] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the present invention is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part for which is described as a larger operation is performed independently.

[0129] In addition, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features described may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in separate physical devices or software modules. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present invention. More specifically, in view of the properties, functions, and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional skills of the engineer. Therefore, those skilled in the art can implement the present invention set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0130] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0131] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0132] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0133] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0134] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0135] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

[0136] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the described embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A PM with particle diameter restriction 2.5 The deep learning remote sensing estimation method is characterized by: include: Obtain site observation data, reanalysis data, satellite remote sensing data and auxiliary data. The site observation data includes PM 2.5 Site Data and PM 10 Site data; Wherein, the obtaining of reanalysis data includes at least one of the following: Obtain weather condition data; Get air quality data; The obtaining of air quality data includes at least one of the following: Get PM1 concentration; Get PM 2.5 concentration; According to the reanalysis data, combined with the PM 2.5 Station data, obtain PM1 equivalent observation data; Wherein, according to the reanalysis data, combined with the PM 2.5 Site data, obtain PM1 equivalent observation data, including: According to the PM1 concentration and PM 2.5 The ratio of the concentration of PM 2.5 The station data is used to perform PM1 equivalent observation calculation to obtain PM1 equivalent observation data. The expression of the PM1 equivalent observation is: Among them, PM1 represents PM1 equivalent observed PM1 data, PM1 reanalysis data represents PM1 concentration obtained from reanalysis data, PM 2.5 Reanalysis data refers to PM obtained from reanalysis data. 2.5 Concentration, PM 2.5 Indicates the PM observed at the site 2.5 data; According to the PM1 equivalent observation data, the PM 2.5 Site data and the PM 10 The site data is combined with the prior relationship of particle diameter size to determine the neural network loss function of particle diameter constraint; Wherein, the PM1 equivalent observation data, the PM 2.5 Site data and the PM 10 The site data is combined with the prior relationship of particle diameter size to determine the neural network loss function of particle diameter constraint, including: According to the PM1 equivalent observation data, the PM 2.5 Site data and the PM 10 Site data, combined with PM1, PM 2.5 and PM 10 The a priori relationship between the concentration values ​​is used to determine the PM 2.5 Constraints on remote sensing estimates; According to the constraint conditions, the neural network loss function of the particle diameter constraint is determined, and the expression of the neural network loss function is: Where N is the number of samples, i represents the i-th sample, PM 2.5 Represents the PM obtained from site observation data 2.5 data, Indicates PM 2.5 PM estimated by the estimation model 2.5 data, λ1 and λ2 represent regularization parameters, ψ(PM1) represents the PM1 constraint term, and ψ(PM 10 ) means PM 10 Constraint term, ReLU() represents the linear rectification function, PM1 represents PM1 equivalent observation data, PM 10 Represents the PM obtained from site observations 10 data; Based on the deep neural network, combined with the neural network loss function, PM is constructed 2.5 Estimation model; Among them, the deep neural network is combined with the neural network loss function to construct PM 2.5 Estimation models, including: A deep neural network is constructed according to the satellite remote sensing data and the auxiliary data. The expression of the deep neural network is: PM 2.5 =f(AOD,x1,x2,x3,...), Among them, PM 2.5 PM represents the PM estimated by the deep neural network 2.5 data; f() represents a deep neural network, AOD represents satellite remote sensing aerosol optical thickness data, x1, x2, x3, ... represent auxiliary data; According to the sample library, the PM 2.5 The estimation model is trained and optimized to determine the PM 2.5 Estimate the optimal parameters of the model; The reanalysis data, the satellite remote sensing data and the auxiliary data are input into the PM after the optimal parameters are determined. 2.5 Estimation model, get PM 2.5 Estimation results.

2. A PM particle diameter-constrained according to claim 1 2.5 The deep learning remote sensing estimation method is characterized by: The PM 2.5 The estimation model is trained and optimized to determine the PM 2.5 Estimate the optimal parameters of the model, including: Dividing the sample library into data sets; The divided sample library includes a training set, a validation set and a test set.

3. A PM particle diameter-constrained according to claim 2 2.5 The deep learning remote sensing estimation method is characterized by: The PM 2.5 The estimation model is trained and optimized to determine the PM 2.5 Estimate the optimal parameters of the model, including: According to the training set and the validation set, the PM is determined by cross validation. 2.5 Estimate the optimal parameters of the model; Based on the optimal parameters, the PM is evaluated in combination with the test set. 2.5 Estimate the performance of the model.

4. A PM particle diameter-constrained according to claim 1 2.5 The deep learning remote sensing estimation method is characterized by: The reanalysis data, the satellite remote sensing data and the auxiliary data are input into the PM after determining the optimal parameters. 2.5 Estimation model, get PM 2.5 Estimated results include: The input reanalysis data, the satellite remote sensing data and the auxiliary data are used to calculate the PM 2.5 Estimating the model does a forward pass.

5. A PM with particle diameter restriction 2.5 The deep learning remote sensing estimation system is characterized by: include: The first module is used to obtain site observation data, reanalysis data, satellite remote sensing data and auxiliary data. The site observation data includes PM 2.5 Site Data and PM 10 Site data; Wherein, the obtaining of reanalysis data includes at least one of the following: Obtain weather condition data; Get air quality data; The obtaining of air quality data includes at least one of the following: Get PM1 concentration; Get PM 2.5 concentration; The second module is used to combine the PM 2.5 Station data, obtain PM1 equivalent observation data; Wherein, according to the reanalysis data, combined with the PM 2.5 Site data, obtain PM1 equivalent observation data, including: According to the PM1 concentration and PM 2.5 The ratio of the concentration of PM 2.5 The station data is used to perform PM1 equivalent observation calculation to obtain PM1 equivalent observation data. The expression of the PM1 equivalent observation is: Among them, PM1 represents PM1 equivalent observed PM1 data, PM1 reanalysis data represents PM1 concentration obtained from reanalysis data, PM 2.5 Reanalysis data refers to PM obtained from reanalysis data. 2.5 Concentration, PM 2.5 Indicates the PM observed at the site 2.5 data; The third module is used to calculate the PM1 equivalent observation data and the PM 2.5 Site data and the PM 10 The site data is combined with the prior relationship of particle diameter size to determine the neural network loss function of particle diameter constraint; Wherein, the PM1 equivalent observation data, the PM 2.5 Site data and the PM 10 The site data is combined with the prior relationship of particle diameter size to determine the neural network loss function of particle diameter constraint, including: According to the PM1 equivalent observation data, the PM 2.5 Site data and the PM 10 Site data, combined with PM1, PM 2.5 and PM 10 The a priori relationship between the concentration values ​​is used to determine the PM 2.5 Constraints on remote sensing estimates; According to the constraint conditions, the neural network loss function of the particle diameter constraint is determined, and the expression of the neural network loss function is: Where N is the number of samples, i represents the i-th sample, PM 2.5 Represents the PM obtained from site observation data 2.5 data, Indicates PM 2.5 PM estimated by the estimation model 2.5 data, λ1 and λ2 represent regularization parameters, ψ(PM1) represents the PM1 constraint term, and ψ(PM 10 ) means PM 10 Constraint term, ReLU() represents the linear rectification function, PM1 represents PM1 equivalent observation data, PM 10 Represents the PM obtained from site observations 10 data; The fourth module is used to construct PM based on deep neural network and combined with the neural network loss function. 2.5 Estimation model; wherein, based on the deep neural network, combined with the neural network loss function, PM is constructed 2.5 Estimation models, including: A deep neural network is constructed according to the satellite remote sensing data and the auxiliary data. The expression of the deep neural network is: PM 2.5 =f(AOD,x1,x2,x3,...), Among them, PM 2.5 PM represents the PM estimated by the deep neural network 2.5 data; f() represents a deep neural network, AOD represents satellite remote sensing aerosol optical thickness data, x1, x2, x3, ... represent auxiliary data; The fifth module is used to analyze the PM according to the sample library. 2.5 The estimation model is trained and optimized to determine the PM 2.5 Estimate the optimal parameters of the model; The sixth module is used to input the reanalysis data, the satellite remote sensing data and the auxiliary data into the PM after determining the optimal parameters. 2.5 Estimation model, get PM 2.5 Estimation results.

Images