Calculation method for methane emission in oil and gas field production area

Abstract

The invention belongs to the technical field of atmospheric remote sensing, and relates to an oil and gas field production area methane emission calculation method which comprises the following steps: acquiring simulated spectral intensity data and a column weight function; acquiring actually measured spectral intensity data; inputting the simulated spectral intensity data, the column weight function and the actually measured spectral intensity data into a weighted correction differential spectral absorption algorithm to construct a broadband spectral structure; training a full-connection neural network based on the broadband spectrum structure and the selected features; improving a weighted correction differential spectral absorption algorithm by using the trained full-connection neural network; acquiring the methane total column concentration according to an improved weighted correction differential spectral absorption algorithm; separating the total methane column concentration into a methane column concentration increment; calculating the methane emission of the oil and gas field production area according to the methane column concentration increment. According to the method, the influence of each factor is better and quantitatively expressed by using the full-connection neural network, the uncertainty caused by order selection of a low-order polynomial is reduced, the inversion result is more accurate, and the network structure is clear.

Description

Technical Field

[0001] The invention belongs to the technical field of atmospheric remote sensing, and in particular relates to a method for calculating methane emissions in oil and gas field production areas. Background Art

[0002] The greenhouse effect of CH4 can reach 28 times that of CO2 on a 100-year time scale, and its contribution to anthropogenic radiative forcing of long-lived greenhouse gases is about 18%. Controlling CH4 emissions has a significant effect on mitigating global warming. Satellite monitoring can provide global CH4 concentration data with high temporal and spatial resolution, and improve the data support capabilities for CH4 spatiotemporal dynamic monitoring, CH4 balance accounting, and emission reduction measures assessment. Oil and gas field production areas are important sources of methane emissions, and it is of great significance to apply satellite remote sensing technology to methane emission monitoring in oil and gas field production areas.

[0003] Remote sensing inversion of methane column concentration is the basis for calculating emissions. Among existing satellite remote sensing technologies, TROPOMI (TROPOspheric Monitoring Instrument) carried on the Sentienl-5P satellite has excellent temporal, spatial and spectral resolution. WFM-DOAS (weighted differential absorption spectroscopy) is the mainstream algorithm used for inversion of methane column concentration in the TROPOMI shortwave infrared band. This algorithm introduces a column weight function to fit the "narrow-band" spectral structure caused by gas absorption, taking into account the complex atmospheric radiation transmission mechanism process; however, for the "broadband" spectral structure caused by Mie scattering, Rayleigh scattering and aerosols, it is only fitted by a low-order polynomial, which makes it difficult to quantify the influence of each factor and has a large uncertainty. Summary of the invention

[0004] In view of the above problems, the present invention proposes a method for calculating methane emissions from oil and gas field production areas. The technical solution adopted by the present invention is as follows: a method for calculating methane emissions from oil and gas field production areas, characterized in that the method comprises the following steps: obtaining simulated spectral intensity data and column weight functions; obtaining measured spectral intensity data; inputting the simulated spectral intensity data, column weight functions and measured spectral intensity data into a weighted corrected differential spectral absorption algorithm to construct a broadband spectral structure; training a fully connected neural network based on the broadband spectral structure and selected features; improving the weighted corrected differential spectral absorption algorithm using the trained fully connected neural network; obtaining the total methane column concentration according to the improved weighted corrected differential spectral absorption algorithm; separating the total methane column concentration into methane column concentration increments; and calculating the methane emission value of the oil and gas field production area according to the methane column concentration increments.

[0005] Furthermore, obtaining simulated spectral intensity data and column weight functions includes: inputting the atmospheric profile parameters, surface parameters, solar spectrum, pure absorption parameters and observation parameters of the study area into the radiation transfer model, and obtaining the simulated spectral intensity of the shortwave infrared band and the column weight functions of the target gas and interfering gas, temperature and air pressure after simulation by the forward radiation transfer model.

[0006] Furthermore, the atmospheric profile parameters include CH4, CO, H2O, temperature and pressure profile data; the surface parameters include surface type and surface albedo; the observation parameters include the longitude and latitude of the study area, the solar zenith angle, the satellite zenith angle, the satellite azimuth angle, the aerosol type and the instrument function.

[0007] Furthermore, the simulated spectral intensity data, the column weight function and the measured spectral intensity data are input into a weighted corrected differential spectral absorption algorithm, and constructing a broadband spectral structure includes: inputting the simulated spectral intensity data, the column weight function and the measured spectral intensity data into the weighted corrected differential spectral absorption algorithm, and the weighted corrected differential spectral absorption algorithm is:

[0008]

[0009] in, is the measured spectral intensity data; V t is the real column concentration of the target gas to be inverted; b t Represent other real atmospheric and surface parameters; is the target gas prior column concentration; are other a priori atmospheric and surface parameters; It represents the simulated spectral intensity under the prior atmospheric and surface parameters; is the target gas column concentration to be fitted; is the jth other atmospheric and surface parameter to be fitted; is the jth other atmospheric and surface prior parameters; F is the number of variables of other atmospheric and surface parameters other than the target gas used; i is the corresponding wavelength, P i is a polynomial about wavelength; unknown parameters and polynomial coefficients were determined by linear least squares method;

[0010] The measured spectral intensity data is subtracted from the adjustment polynomial P i The optimal fitting result obtained after the order of is constructed, and the broadband spectrum structure Y is:

[0011]

[0012] Furthermore, training a fully connected neural network based on the broadband spectral structure and feature vector includes: selecting four variables, namely wavelength λ, surface albedo, aerosol optical thickness AOD and aerosol single scattering albedo SSA, as features, recorded as feature vector A=(λ, albedo, AOD, SSA), pairing them with the broadband spectral structure Y, and constructing a feature and label data set; setting a fully connected neural network, including an input layer, a hidden layer and an output layer; configuring network parameters, and training the fully connected neural network.

[0013] Furthermore, the first hidden layer is provided with multiple neurons, which are activated by the Relu function; the second hidden layer is provided with multiple neurons, and the output result is made to conform to the probability distribution by the Softmax function, and the L2 regularization formula is used; wherein, the formula of the Relu function is: Relu(x)=max(x,0), where x represents the input value of the Relu function; that is, when the input value is greater than 0, the input value is taken as the output result, otherwise it is 0; the Softmax function is:

[0014] y i represents the i-th output, n represents the total number of outputs; the L2 regularization formula is: w is the weight of each link in the network structure, To sum up the squares of the weights;

[0015] Using the cross entropy function and Adam optimizer, the proportion of the validation data set is selected to train the fully connected neural network; the cross entropy function formula is:

[0016] H(y_,y)=-∑y - *ln y;y - is the true value, y is the predicted value in probability form, representing the distance between the two probability distributions;

[0017] In the Adam optimizer, the first-order momentum formula is: t =β1·m t-1 +(1-β1)·g t ; g t is the current parameter gradient, t is the current iteration number, and β1 is the hyperparameter;

[0018] The deviation correction formula for the first-order momentum is:

[0019] The second-order momentum formula is: β2 is a hyperparameter;

[0020] The deviation correction formula for the second-order momentum is:

[0021] Then the gradient descent is:

[0022]

[0023] The parameters are updated as follows:

[0024]

[0025] Further, obtaining the total column concentration of methane according to the improved weighted modified differential spectral absorption algorithm includes: replacing the low-order polynomial P in the weighted modified differential spectral absorption algorithm with the trained fully connected neural network i Fitting the broadband spectral structure, the improved weighted corrected differential spectral absorption algorithm is obtained as follows:

[0026] Among them, FCNN is a fully connected neural network, and the total column concentration of methane in the study area is obtained by least squares fitting.

[0027] Furthermore, separating the total methane column concentration into methane column concentration increments includes: expressing the total methane column concentration as the sum of the methane column concentration in the stratosphere, the background value of the methane column concentration in the troposphere and the incremental value of the methane column concentration in the lower troposphere through the relationship between the methane column concentration and the air pressure, and separating it into the methane column concentration increments through least squares fitting.

[0028] Furthermore, the total methane column concentration is expressed as the sum of the methane column concentration in the stratosphere, the background value of the methane column concentration in the troposphere, and the incremental value of the methane column concentration in the lower troposphere through the relationship between the methane column concentration and the air pressure. Through the least squares fitting, the methane column concentration increment is separated into: the total methane column concentration n is expressed as: Where n is the total column concentration of methane, XCH4 is the dry air mixing volume ratio, and p d，sfc is the surface pressure, g is the gravitational acceleration constant, m air is the molar mass of dry air;

[0029] It represents the conversion relationship between XCH4 and the total column concentration of methane n. Due to the characteristics of methane in the atmosphere, the total column concentration of methane n is regarded as consisting of the methane column concentration in the stratosphere, the methane background concentration in the troposphere and the methane incremental concentration in the lower troposphere; the total column concentration of methane n is expressed as:

[0030] n=kXCH4 s p d，tp +kXCH4 t (p d，sfc -p d，tp )+Δn; where k is the conversion coefficient, XCH4 s is the average stratospheric methane column concentration, p d，tp is the pressure at the top of the troposphere, XCH4 tis the background value of tropospheric methane column concentration; p d，sfc is the surface pressure, Δn is the increment of tropospheric methane column concentration; then the increment of tropospheric methane column concentration Δn is:

[0031] Δn=n-(c0+c1p d，sfc ); where c0 = k(XCH4 s -XCH4 t ) d，tp ;

[0032] c1=kXCH4 t ; Among them, the background value of methane column concentration within a certain range and the air pressure at the top of the troposphere are constants. c0 and c1 are obtained by linear fitting. The total methane column concentration n is obtained by combining the improved weighted corrected differential spectral absorption algorithm, and the methane column concentration increment Δn is calculated.

[0033] Further, calculating the methane emission value of the oil and gas field production area according to the methane column concentration increment includes: calculating the methane emission value E of the oil and gas field production area by combining the methane column concentration increment Δn through the horizontal flux divergence method:

[0034]

[0035] Among them, E is the methane emission, F is the methane flux, Δn is the methane column concentration increment, and v is the horizontal wind field, using the average wind field from the surface to a height of 500m. is the horizontal divergence operator;

[0036] Where v = (u, v), u is the wind speed component in the south-north direction, and v is the wind speed component in the east-west direction; x and y are two orthogonal directions, i and j are unit vectors in the south-north and east-west directions respectively; the methane emission E in the oil and gas field production area is calculated by the fourth-order central finite difference algorithm.

[0037] This method uses a fully connected neural network (FCNN) to fit the spectral "broadband" structure with multi-source data as features, making full use of the nonlinear fitting ability of the fully connected neural network. Compared with the traditional WFM-DOAS algorithm that uses low-order polynomials to fit the "broadband" spectral structure, the use of a fully connected neural network can better quantitatively represent the influence of various factors, better express the spectral structure formed by factors such as Mie scattering, Rayleigh scattering and aerosols, reduce the uncertainty caused by the selection of the order of low-order polynomials, make the inversion results more accurate, and the network structure is clear, which is easy to promote and use.

[0038] Other features and advantages of the present invention will be described in the following description, and part of them will become obvious from the description, or be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the description and the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 A flow chart of a method for calculating methane emissions in oil and gas field production areas;

[0041] Figure 2 Schematic diagram of the fully connected neural network structure used for "broadband" spectral structure fitting. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] The embodiment of the present invention provides a method for calculating methane emissions in oil and gas field production areas. The method is based on remote sensing inversion technology of methane emissions in oil and gas field production areas fused with FCNN and WFM-DOAS. Figure 1 As shown, the method includes the following steps: S1, obtaining simulated spectral intensity data and column weight function; S2, obtaining measured spectral intensity data; S3, inputting the simulated spectral intensity data, column weight function and measured spectral intensity data into a weighted corrected differential spectral absorption algorithm to construct a broadband spectral structure; S4, training a fully connected neural network based on the broadband spectral structure and selected features; S5, improving the weighted corrected differential spectral absorption algorithm using the trained fully connected neural network; S6, obtaining the total methane column concentration according to the improved weighted corrected differential spectral absorption algorithm; S7, separating the total methane column concentration into methane column concentration increments; S8, calculating the methane emission value of the oil and gas field production area according to the methane column concentration increments.

[0044] In the process of using satellite remote sensing to invert methane column concentration, it is necessary to perform forward inversion on the atmospheric radiation transfer simulation process to solve the simulated spectral data and column weight function under the conditions of prior atmospheric and surface parameters. In step S1, the high spectral resolution radiation transfer model SCIATRAN is used, and the input data include atmospheric profile parameters (including CH4, CO, H2O, temperature and pressure profile data), surface parameters (surface type, surface albedo), absorption pure parameters (HITRAN), solar spectrum and observation parameters (including longitude and latitude of the study area, solar zenith angle, satellite zenith angle, satellite azimuth angle, aerosol type, instrument function), etc. The simulated satellite sensor is the TROPOMI sensor carried on the Sentinel-5P satellite, the simulated band is the 7th band (corresponding to the wavelength range of 2311nm-315.5nm and 2320-2338nm), the spectral resolution is 0.1nm, the linear function is Gaussian, and the FWHM (full width at half maxima) corresponding to the 7th band of the TROPOMI sensor is used. The gas cross-section is integrated using the line-by-line integration mode, and the SCIATRAN radiation transfer model is run to obtain simulated spectral data and column weight function.

[0045] Furthermore, the specific process of the atmospheric radiation transfer simulation by the SCIATRAN radiation transfer model in step S1 is as follows: Assuming the forward model f(x), the inversion process can be expressed as:

[0046]

[0047] The left side of the above equation is the quantity related to the actual measurement. Assuming there are M variables, the above equation can be expressed as:

[0048]

[0049] in, is the state vector, which is a continuous function of the position in an ideal state, but is actually a discrete variable. Assuming that it contains n independent variables, the state variable It can be expressed as:

[0050]

[0051] Since each measurement has an error, it can be expressed as:

[0052]

[0053] Specifically, step S3 includes: inputting the simulated spectral intensity data, the column weight function and the measured spectral intensity data into a weighted modified differential spectral absorption algorithm, and the weighted modified differential spectral absorption algorithm is:

[0054]

[0055] in, is the measured spectral intensity data; V t is the real column concentration of the target gas to be inverted; b t Represent other real atmospheric and surface parameters; is the target gas prior column concentration; are other a priori atmospheric and surface parameters; It represents the simulated spectral intensity under the prior atmospheric and surface parameters; is the target gas column concentration to be fitted; is the jth other atmospheric and surface parameter to be fitted; is the jth other atmospheric and surface prior parameters; F is the number of variables of other atmospheric and surface parameters other than the target gas used; i is the corresponding wavelength, P i is a polynomial about wavelength; unknown parameters and polynomial coefficients were determined by linear least squares method;

[0056] The spectral structure formed by Mie scattering, Rayleigh scattering, aerosol and error is assumed to be a “broadband structure”, which can be expressed by a low-order polynomial P i However, in practical applications, the influence of scattering and aerosols is complex, and the order of the polynomial will also affect the inversion result. Fitting with only the polynomial has a large uncertainty. Therefore, in the embodiment of the present invention, by adjusting the polynomial P i The order of the measured spectral intensity data is subtracted from the adjustment polynomial P i The optimal fitting result obtained after the order of is constructed, and the broadband spectrum structure Y is:

[0057]

[0058] Specifically, step S4 includes: selecting wavelength λ, surface albedo, aerosol optical depth AOD, and aerosol single scattering albedo SSA as four variables, recorded as vector A = (λ, albedo, AOD, SSA), and using FCNN to establish the relationship between the broadband spectral structure Y in the spectral intensity and the feature vector A.

[0059] In step S4, the method of constructing FCNN is specifically as follows:

[0060] 1. Construct feature and label dataset

[0061] Select wavelength λ, surface albedo, aerosol optical thickness AOD, and aerosol single scattering albedo SSA as input features, and the corresponding value in Y as the label. The wavelength is the wavelength value corresponding to the 7th band of the TROPOMI sensor used. Albedo, AOD, and SSA are obtained through satellite products such as MODIS (Moderate Resolution Imaging Spectroradiometer). Albedo, AOD, and SSA are converted to the same spatial scale as the 7th band data of the TROPOMI sensor, and paired with the broadband spectral dataset constructed in step S3 to construct a (feature, label) dataset;

[0062] 2. Set up the network structure

[0063] Use FCNN to train broadband structure models, such as Figure 2 As shown in the figure, FCNN consists of an input layer, two hidden layers and an output layer. The first hidden layer is set with 110 neurons, which are activated by the Relu function. The second hidden layer is set with 50 neurons. The Softmax function is used to make the output result conform to the probability distribution, and the L2 regularization method is used. The formula of the Relu function is: Relu(x) = max(x, 0), x represents the input value of the Relu function; that is, when the input value is greater than 0, the input value is taken as the output result, otherwise it is 0.

[0064] The Softmax function is: y i represents the i-th output, and n represents the total number of outputs;

[0065] The L2 regularization formula is: w is the weight of each link in the network structure, To sum the squares of the weights; REGULARIZER is set to 0.03.

[0066] 3. Configure the fully connected neural network parameters and perform fully connected neural network training

[0067] The loss function uses the cross entropy function, the Adam optimizer is used, the learning rate lr is 0.1, the first-order moment estimate exponential decay rate beta_1 is 0.9, the second-order moment estimate exponential decay rate beta_2 is 0.999, batch_size (the number of data passed to the program for training at a single time) is 32, and epochs (the number of training iterations of all batches during forward propagation and backpropagation) is 1000; 0.2 is selected as the validation data set ratio, and the fully connected neural network is trained;

[0068] Among them, the cross entropy function formula is: H(y - ,y)=-∑y - *lny;y -is the true value, y is the predicted value in probability form, representing the distance between the two probability distributions;

[0069] In the Adam optimizer, the first-order momentum formula is: t =β1·m t-1 +(1-β1)·g t ; g t is the current parameter gradient, t is the current iteration number, and β1 is the hyperparameter;

[0070] The deviation correction formula for the first-order momentum is:

[0071] The second-order momentum formula is: β2 is a hyperparameter;

[0072] The deviation correction formula for the second-order momentum is:

[0073] Then the gradient descent is:

[0074]

[0075] The parameters are updated as follows:

[0076]

[0077] Specifically, step S5 includes: replacing the low-order polynomial P in the weighted modified differential spectral absorption algorithm with the trained fully connected neural network i Fitting the broadband spectral structure, the improved weighted corrected differential spectral absorption algorithm is obtained as follows:

[0078]

[0079] Specifically, step S6 includes: performing least square fitting on the improved weighted corrected differential spectral absorption algorithm to invert and obtain the total column concentration of methane in the study area.

[0080] Specifically, step S7 includes: the total methane column concentration obtained in step S6 is expressed as the sum of the methane column concentration in the stratosphere, the background value of the methane column concentration in the troposphere, and the incremental value of the methane column concentration in the lower part of the troposphere through the relationship between the methane column concentration and the air pressure, and separated into the methane column concentration increment through least squares fitting, specifically including: the total methane column concentration n is expressed as:

[0081] The total column concentration of methane n is expressed as:

[0082] Where n is the total column concentration of methane, XCH4 is the dry air mixing volume ratio, and p d，sfc is the surface pressure, g is the gravitational acceleration constant, m airis the molar mass of dry air; represents the conversion relationship between XCH4 and the total column concentration of methane n. Due to the characteristics of methane in the atmosphere, the total column concentration of methane n is regarded as consisting of the methane column concentration in the stratosphere, the background value of the methane column concentration in the troposphere, and the incremental concentration of methane in the lower troposphere; the total column concentration of methane n is expressed as:

[0083] n=kXCH4 s p d，tp +kXCH4 t (p d，sfc -p d，tp )+Δn; where k is the conversion coefficient, XCH4 s is the average stratospheric methane column concentration, p d，tp is the pressure at the top of the troposphere, XCH4 t is the background value of tropospheric methane column concentration; p d，sfc is the surface pressure, Δn is the increment of tropospheric methane column concentration;

[0084] Then the increment of tropospheric methane column concentration Δn is:

[0085] Δn=n-(c0+c1p d，sfc ); where c0 = k(XCH4 s -XCH4 t ) d，tp ;

[0086] c1=kXCH4 t ;

[0087] Among them, the background value of methane column concentration within a certain range and the air pressure at the top of the troposphere are constants. c0 and c1 are obtained by linear fitting. The total methane column concentration n is obtained by combining the improved weighted corrected differential spectral absorption algorithm, and the methane column concentration increment Δn is calculated.

[0088] Specifically, step S8 includes: using the horizontal flux divergence method, combined with the methane column concentration increment Δn obtained in step S7, solving the specific formula of the methane emission value E as follows:

[0089] E is methane emission (Emission), F is methane flux (Flux), Δn is the methane column concentration increment, v is the horizontal wind field (HorizontalWindField), using the average wind field from the surface to a height of 500m; where v = (u, v), u and v are two orthogonal wind speed components, u is the south-north wind speed component, and v is the east-west wind speed component; is the horizontal divergence operator, x and y are two orthogonal directions, i and j are unit vectors in the south-north direction and east-west direction respectively. Finally, the methane emission E and its spatial distribution in the oil and gas field production area are calculated by the fourth-order central-finite difference algorithm.

[0090] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for calculating methane emissions in oil and gas field production areas, characterized in that: The method comprises the following steps: obtain simulated spectral intensity data and column weight functions; Obtain measured spectral intensity data; The simulated spectral intensity data, column weight function and measured spectral intensity data are input into the weighted corrected differential spectral absorption algorithm to construct a broadband spectral structure; Training a fully connected neural network based on the broadband spectral structure and selected features; The trained fully connected neural network is used to improve the weighted correction differential spectral absorption algorithm; The total column concentration of methane is obtained according to the improved weighted corrected differential spectral absorption algorithm; separating the total column concentration of methane into methane column concentration increments; Calculate the methane emissions from the oil and gas field production area based on the methane column concentration increment.

2. The method according to claim 1, characterized in that: Acquisition of simulated spectral intensity data and column weight functions involves: The atmospheric profile parameters, surface parameters, solar spectrum, pure absorption parameters and observation parameters of the study area are input into the radiation transfer model. After simulation by the forward radiation transfer model, the simulated spectral intensity of the shortwave infrared band and the column weight functions of the target gas and interfering gas, temperature and air pressure are obtained.

3. The method according to claim 2, characterized in that Atmospheric profile parameters include CH4, CO, H2O, temperature and pressure profile data; surface parameters include surface type and surface albedo; observation parameters include longitude and latitude of the study area, solar zenith angle, satellite zenith angle, satellite azimuth angle, aerosol type and instrument function.

4. The method according to claim 1, characterized in that The simulated spectral intensity data, column weight function and measured spectral intensity data are input into the weighted corrected differential spectral absorption algorithm to construct a broadband spectral structure including: The simulated spectral intensity data, column weight function and measured spectral intensity data are input into the weighted corrected differential spectral absorption algorithm. The weighted corrected differential spectral absorption algorithm is: in, is the measured spectral intensity data; V t is the real column concentration of the target gas to be inverted; b t Represent other real atmospheric and surface parameters; is the target gas prior column concentration; are other a priori atmospheric and surface parameters; It represents the simulated spectral intensity under the prior atmospheric and surface parameters; is the target gas column concentration to be fitted; is the jth other atmospheric and surface parameter to be fitted; is the jth other atmospheric and surface prior parameters; F is the number of variables of other atmospheric and surface parameters other than the target gas used; i is the corresponding wavelength, P i is a polynomial about wavelength; unknown parameters and polynomial coefficients were determined by linear least squares method; The measured spectral intensity data is subtracted from the adjustment polynomial P i The optimal fitting result obtained after the order of is constructed, and the broadband spectrum structure Y is:

5. The method according to claim 4, characterized in that Training a fully connected neural network based on broadband spectral structure and feature vectors includes: Select wavelength λ, surface albedo, aerosol optical depth AOD and aerosol single scattering albedo SSA as features, record them as feature vector A = (λ, albedo, AOD, SSA), pair them with broadband spectral structure Y, and construct feature and label dataset; Set up a fully connected neural network, including input layer, hidden layer and output layer; Configure network parameters and train the fully connected neural network.

6. The method according to claim 5, characterized in that The first hidden layer is set with multiple neurons, which are activated by the Relu function; the second hidden layer is set with multiple neurons, and the output results are made to conform to the probability distribution through the Softmax function, using the L2 regularization formula; Among them, the formula of the Relu function is: Relu(x)=max(x,0), x represents the input value of the Relu function; That is, when the input value is greater than 0, the input value is taken as the output result, otherwise it is 0; The Softmax function is: y i represents the i-th output, and n represents the total number of outputs; The L2 regularization formula is: w is the weight of each link in the network structure, To sum up the squares of the weights; Using the cross entropy function and Adam optimizer, the proportion of the validation data set is selected to train the fully connected neural network; the cross entropy function formula is: H(y - ,y)=-Σy - *ln y;y - is the true value, y is the predicted value in probability form, representing the distance between the two probability distributions; In the Adam optimizer, the first-order momentum formula is: t =β1·m t-1 +(1-β1)·g t ; g t is the current parameter gradient, t is the current iteration number, and β1 is the hyperparameter; The deviation correction formula for the first-order momentum is: The second-order momentum formula is: β2 is a hyperparameter; The deviation correction formula for the second-order momentum is: Then the gradient descent is: The parameters are updated as follows:

7. The method according to claim 1, characterized in that The total column concentration of methane is obtained according to the improved weighted corrected differential spectral absorption algorithm, including: The trained fully connected neural network is used to replace the low-order polynomial P in the weighted modified differential spectral absorption algorithm. i Fitting the broadband spectral structure, the improved weighted corrected differential spectral absorption algorithm is obtained as follows: Among them, FCNN is a fully connected neural network, and the total column concentration of methane in the study area is obtained by least squares fitting.

8. The method according to claim 7, characterized in that Separation of the total methane column concentration into methane column concentration increments involves: The total methane column concentration is expressed as the sum of the methane column concentration in the stratosphere, the background value of the methane column concentration in the troposphere, and the incremental value of the methane column concentration in the lower troposphere through the relationship between the methane column concentration and the air pressure. It is separated into the methane column concentration increment through least squares fitting.

9. The method according to claim 8, characterized in that The total methane column concentration is expressed as the sum of the methane column concentration in the stratosphere, the background value of the methane column concentration in the troposphere, and the incremental value of the methane column concentration in the lower troposphere through the relationship between the methane column concentration and the air pressure. Through the least squares fitting, the methane column concentration increment is separated into: The total column concentration of methane n is expressed as: Where n is the total column concentration of methane, XCH4 is the dry air mixing volume ratio, and p d，sfc is the surface pressure, g is the gravitational acceleration constant, m air is the molar mass of dry air; It represents the conversion relationship between XCH4 and the total column concentration of methane n. Due to the characteristics of methane in the atmosphere, the total column concentration of methane n is regarded as consisting of the methane column concentration in the stratosphere, the methane background concentration in the troposphere and the methane incremental concentration in the lower troposphere; the total column concentration of methane n is expressed as: n=kXCH4 s p d，tp +kXCH4 t (p d，sfc -p d，tp )+Δn; Where k is the conversion factor, XCH4 s is the average stratospheric methane column concentration, p d，tp is the pressure at the top of the troposphere, XCH4 t is the background value of tropospheric methane column concentration; p d，sfc is the surface pressure, Δn is the increment of tropospheric methane column concentration; Then the increment of tropospheric methane column concentration Δn is: Δn=n-(c0+c1p d，sdfc ); where c0 = k(XCH4 s -XCH4 t ) d，tp ; c1 = kXCH4 t ; Among them, the background value of methane column concentration within a certain range and the air pressure at the top of the troposphere are constants. c0 and c1 are obtained by linear fitting. The total methane column concentration n is obtained by combining the improved weighted corrected differential spectral absorption algorithm, and the methane column concentration increment Δn is calculated.

10. The method according to any one of claims 1 to 9, characterized in that: Calculating the methane emission value of the oil and gas field production area based on the methane column concentration increment includes: using the horizontal flux divergence method, combined with the methane column concentration increment Δn, to calculate the methane emission E of the oil and gas field production area: Among them, E is the methane emission, F is the methane flux, Δn is the methane column concentration increment, and v is the horizontal wind field, using the average wind field from the surface to a height of 500m. is the horizontal divergence operator; Where v = (u, v), u is the wind speed component in the south-north direction, and v is the wind speed component in the east-west direction; x and y are two orthogonal directions, i and j are unit vectors in the south-north and east-west directions respectively; the methane emission E in the oil and gas field production area is calculated by the fourth-order central finite difference algorithm.

Images