Oilfield methane leakage detection method using satellite remote sensing data

Abstract

The invention discloses an oil field methane leakage detection method using satellite remote sensing data, and belongs to the field of atmosphere remote sensing and energy production, the technical scheme is that the oil field methane leakage detection method using the satellite remote sensing data comprises the following steps: screening a methane absorption spectrum data range as a simulation spectrum; acquiring hyperspectral image data in a simulated spectrum range, and preprocessing the hyperspectral image data as an observation spectrum; simulating a spectral absorption curve of gas in an oil field area in a real-time state, and finding an ideal reference spectral line; performing logarithmic difference processing on the screened simulation spectrum and the reference spectrum to obtain prior spectrum data; inputting the observation spectrum and the prior spectrum into a matched filter, and performing inversion to obtain an XCH4 enhanced image; and identifying a plume region in the XCH4 enhanced image by using a convolutional neural network. The oil field methane leakage detection method has the beneficial effect that the oil field methane leakage detection method utilizing the satellite remote sensing data is provided.

Description

Technical Field

[0001] This invention belongs to the fields of atmospheric remote sensing and energy production, and specifically relates to a method for detecting methane leaks in oil fields using satellite remote sensing data. Background Technology

[0002] Since the industrial era, atmospheric methane (CH4) concentration has increased by more than 150%, reaching a global average concentration of 1920 ppb by early 2023. As a key factor in the anthropogenic greenhouse effect, CH4's impact on global warming is undeniable. It is estimated to have contributed nearly one-third of global warming to date, with emissions accounting for one-quarter of all greenhouse gases. The warming effect of a single CH4 molecule on a 20-year timescale is equivalent to 80 times that of carbon dioxide. Therefore, methane emission reduction is crucial for rapidly mitigating climate change. Oil and gas development is a vital source of global energy supply, and related CH4 leaks are the most significant source of methane emissions. Statistics show that CH4 emissions related to oil and gas extraction, processing, and transportation account for more than 30% of total CH4 emissions. The vast majority of oil and gas-related CH4 emissions are leaks, and calculations indicate that the cost of reducing them is lower than the economic value generated by collection. Therefore, methane emission reduction from oil and gas is not only scientifically urgent but also highly feasible in practice. The key to reducing CH4 emissions from oil and gas lies in identifying the leak point. To this end, this application proposes a method for detecting methane leaks in oil fields using satellite remote sensing data. Summary of the Invention

[0003] The purpose of this invention is to provide a method for detecting methane leaks in oil fields using satellite remote sensing data.

[0004] A first aspect of this application provides a method for detecting methane leaks in oil fields using satellite remote sensing data, characterized in that...

[0005] The range of methane absorption spectral data was selected as the simulated spectrum;

[0006] Acquire hyperspectral image data within the simulated spectral range and preprocess the hyperspectral image data to obtain the observed spectrum;

[0007] Simulate the spectral absorption curve of gas under real-time conditions in the oilfield area to find the ideal reference spectral line;

[0008] The selected simulated spectra and reference spectra are subjected to logarithmic difference processing to obtain prior spectral data;

[0009] The observed spectrum and the prior spectrum are input into a matched filter to invert and obtain the XCH4 enhanced image;

[0010] Using convolutional neural networks to identify plume regions in XCH4 enhanced images.

[0011] Furthermore, the range of methane absorption spectral data was selected as the simulated spectrum, specifically:

[0012] Using the Monte Carlo simulation principle, the absorption spectral range with the smallest error was selected from the HITRAN database as the search window, and the optimal combination of spectral bands was selected as the simulated spectrum.

[0013] Furthermore, the hyperspectral image data undergoes preprocessing, including:

[0014] Acquire hyperspectral image data within the optimal combination range;

[0015] Based on the information obtained from the satellite header file, radiometric calibration and atmospheric correction were performed on the hyperspectral image data.

[0016] Cloud masking is used to remove clouds from hyperspectral image data, and data containing clouds are assigned a value of 0.

[0017] Furthermore, finding the ideal reference spectral line specifically includes:

[0018] Calculate the center wavelength and full width at half maximum (FWHM) of the optimal combination range;

[0019] Using the HITRAN molecular absorption library, unit gas absorption spectra at a resolution of 0.1 cm⁻¹ were obtained;

[0020] Based on the full width at half maximum (FWHM), the spectral response function of the hyperspectral remote sensor band is simulated using a Gaussian function to obtain the reference spectral line for stably recorded light intensity under oilfield conditions.

[0021] Furthermore, the matched filter is selected as an iterative log-normal matched filter, specifically:

[0022]

[0023] Furthermore, convolutional neural networks were used to identify plume regions in XCH4 enhanced images, including:

[0024] A dataset was constructed to create a convolutional neural network for identifying plume regions in XCH4 enhanced imagery;

[0025] Construct a convolutional neural network to identify plume regions in XCH4 enhanced images;

[0026] Train the convolutional neural network;

[0027] The dataset was processed using PCA on the XCH4 enhanced images obtained from the inversion.

[0028] The data processed by PCA is input into a convolutional neural network for recognition.

[0029] Furthermore, the dataset is processed using PCA on the inverted XCH4 enhanced imagery, including:

[0030] Select the principal component singular vector that best describes the spectral changes of the scene;

[0031] Determine the number of the first three optimal singular vectors;

[0032] These vectors are concatenated with the methane Jacobian to construct a matrix J of dimension 4 × the number of PRISMA bands. This matrix is ​​then used with the logarithm y of the measured radiation intensity to find the vector W that minimizes the cost function for each pixel in a linear least squares fit.

[0033] Furthermore, the convolutional neural network is trained, including:

[0034] The encoder, which captures context in hyperspectral images, consists of convolutional layers and max-pooling layers;

[0035] The decoder, which can locate the features captured by the encoder, consists of convolutional layers and upsampling layers.

[0036] A second aspect of this application provides a computer device, characterized in that it includes: a processor and a memory, wherein the processor is configured to execute a program for detecting methane leaks in an oil field stored in the memory, so as to implement the method for detecting methane leaks in an oil field.

[0037] A third aspect of the embodiments of this application provides a storage medium, characterized in that the storage medium stores one or more programs, which can be executed by one or more processors to implement the oilfield methane leakage detection method.

[0038] The beneficial effects of the technical solution provided by this invention are as follows: High-resolution ΔXCH4 images are retrieved from oil and gas production areas using an iterative logarithmic matched filtering algorithm and differential absorption spectroscopy. A convolutional neural network is used to identify and extract CH4 plumes from noisy ΔXCH4 images, generating a binarized mask. By obtaining refined CH4 plume data, CH4 leakage phenomena at the oil and gas production facility level can be identified, thereby providing information for implementing CH4 emission reduction measures. Attached Figure Description

[0039] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings listed below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1This is a flowchart of an oilfield methane leak detection method using satellite remote sensing data, as described in an embodiment of the present invention.

[0041] Figure 2 This is a principle block diagram of an oilfield methane leak detection method using satellite remote sensing data, as described in an embodiment of the present invention.

[0042] Figure 3 Schematic diagram of iterative logarithmic matched filtering in this embodiment of the invention;

[0043] Figure 4 This is a schematic diagram of the neural network for automatically identifying plume enhancement regions according to the present invention;

[0044] Figure 5 A methane leak plume map of a joint station in an atmospheric oilfield was obtained based on a method for detecting methane leaks in oilfields using satellite remote sensing data. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. Of course, the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0046] Example 1:

[0047] See Figures 1-5 This embodiment provides a method for detecting methane leaks in oil fields using satellite remote sensing data, characterized in that...

[0048] S1. Select the range of methane absorption spectrum data as the simulated spectrum;

[0049] S11. Optical depth calculated from spectral lines in the HITRAN database at spectral intervals of 0.1 cm⁻¹. Water vapor, carbon dioxide, and methane have fundamental vibrational absorption in the short-wave infrared (SWIR) spectrum. Satellite imaging spectrometers can utilize the CH₄ absorption characteristics in SWIR, which has weak (~1700 nm) and strong (~2300 nm) absorption windows.

[0050] S12. According to calculations based on the HITRAN database, methane has strong absorption between 2200 nm and 2400 nm, and a second weak absorption near 1667 nm, which is similar to that of satellite imaging spectrometers.

[0051] S13. Using a SWIR band matched filter will result in greater enhancement from retrieval artifacts and clutter at 1700 nm, while reducing enhancement from plume pixels that have passed through a 2300 nm band matched filter. The enhancement will be even lower in cases of overlapping CO2 and CH4 plumes. SWIR band matched filter values ​​greater than the 2300 nm band matched filter value have 145 interference factors. Most pixels meeting this condition belong to the negative portion of the normal distribution of the 2300 nm band matched filter value, which is not considered part of the CH4 absorption band in this spectral window. However, for the SWIR band matched filter, due to the appearance of artifacts and new clutter noise below the 1700 nm band matched filter value, these values ​​can be changed back to the original 2300 nm band matched filter value as a penalty enhancement effect. If the selected band range is too short, it will cause the gas to "falsely absorb" due to surface reflection, resulting in a decrease in inversion accuracy. If the selected band range is too long, it may cause interference from other intermediate gases, which will also cause a decrease in inversion accuracy. For example, due to the interference of other trace gases, such as H2O and CO2 that have not been removed from the band, the absorption characteristics of these gases appear in the SWIR spectral region, which may interfere with the CH4 absorption value.

[0052] S14. Using the Monte Carlo simulation principle, the absorption spectral range with the smallest error is selected as the search window in the HITRAN database. There are n spectral lines between 1600nm and 2400nm. Selecting s spectral lines yields a total of the following combinations:

[0053]

[0054] Where s∈(0,N], the mean square error of H2O and CO2 band interference in each band is obtained by filtering out all the band combinations corresponding to τ, the minimum error interval Smin is found, and the band combination corresponding to this interval is recorded as 1650nm-1750nm and 2100nm-2250nm.

[0055] Furthermore, filters were set up in the 1650nm-1750nm and 2100nm-2250nm bands, and the results showed that the error was minimized. This band was then multiplied by a scaling factor. Where, σ 2300nm and σ SWIR This represents the root mean square error of the absorption coefficient of methane in this band under various conditions. This scaling method minimizes interference from other trace gases caused by the infrared band and the 2300nm filter, as well as spurious absorption phenomena from ground surface reflection, allowing the processed band combination to serve as simulated spectral data.

[0056] S2. Acquire hyperspectral image data within the simulated spectral range and preprocess the hyperspectral image data as the observation spectrum;

[0057] Hyperspectral image data within the ranges of 1650nm-1750nm and 2100nm-2250nm were acquired and merged into image cubes (containing two-dimensional image dimensions and spectral information curve dimensions). Based on information obtained from the satellite header file, the data underwent radiometric calibration, atmospheric correction, and cloud removal using cloud masks, with data containing clouds assigned a value of 0. Each cubic grid of the filtered and preprocessed hyperspectral image data was then treated as an independent observation spectrum.

[0058] S3. Simulate the spectral absorption curve of the gas under real-time conditions in the oilfield area and find the ideal reference spectral line;

[0059] The reference spectral lines should meet the following conditions: they should not be absorbed or should be absorbed very weakly in the atmosphere to ensure that the reference spectral lines can stably record the light intensity, and they should cover the absorption wavelength range of the gas CH4 to be measured so that changes in the gas to be measured can be effectively reflected in the reference spectral lines.

[0060] S31. Simulate the spectral absorption curve of the gas in the oilfield area under real-time conditions using the HITRAN database, and find the ideal reference spectral line.

[0061] S32. Calculate the center wavelength (MED) and full width at half maximum (FWHM) of the bands in the 1650nm-1750nm and 2100nm-2250nm band combinations selected in S1.

[0062] S33. The method for obtaining the reference spectral line is to use the HITRAN molecular absorption library and its simulation program to obtain the unit gas absorption spectrum at a resolution of 0.1 cm⁻¹. Then, based on the full width at half maximum (FWHM), the spectral response function of the hyperspectral remote sensor band is simulated using a Gaussian function to obtain the reference spectral line for stably recorded light intensity under oilfield conditions.

[0063]

[0064] Where λ is the spectral wavelength, α is the attenuation coefficient assigned a value of 2, and FWHM is the full width at half maximum (FWHM) of 2100nm-2300nm, calculated using the following formula:

[0065] FWHM=λ R -λ L (3)

[0066] Where, λ R ;λ L These are the right and left intersection points where the half-peak value of the reference spectral line intersects the curve.

[0067] S4. Perform logarithmic difference processing on the selected simulated spectrum and the reference spectrum to obtain prior spectral data, including:

[0068] Perform logarithmic difference processing on the simulated spectrum selected in step S1 and the reference spectrum generated in step S3 to obtain prior spectral data, so as to eliminate the influence of atmospheric scattering and optical thickness. The calculation formula of the prior spectrum is as follows:

[0069]

[0070] where I c (λ): Simulated spectrum of methane absorption, I m (λ): Reference spectral line, D(λ): Prior spectral line. The differential spectrum contains the absorption changes caused by methane. In this case, the interference of other factors on the simulated spectrum of methane absorption is excluded.

[0071] S5. Input the observed spectrum and the prior spectrum into a matched filter, and inversely retrieve the enhanced XCH4 image;

[0072] With the help of the matched filter, estimate in the two-dimensional spatial tiles of the absorption spectrum to extract the spectral background value and the enhanced value. The matched filter can accurately extract the enhanced signal of the CH4 gas concentration of interest from the noise background.

[0073] According to Beer-Lambert law, the enhancement of the column concentration of gas molecules ΔXCH4 will act on the observed spectrum x m :

[0074]

[0075] where, x m represents the radiance observed by the sensor, that is, the observed spectrum, x r represents the reference radiance, k represents the absorption coefficient per unit gas concentration, obtained by looking up the table when calculating the simulated spectrum, i and j are the horizontal and vertical retrieval index coordinates of the two-dimensional observed hyperspectral image, and L is the length of each column of the image.

[0076] To simplify the calculation, expand formula (5) into a first-order form using Taylor formula. Since the AHSI sensor collects data in a column-by-column scanning mode, during inversion, one column is used as a unit for inversion; use the mean value of the radiance of this column to approximate the reference radiance, that is:

[0077] x m (i,j) = μ(j) - k L×1 ·μ L×1 (j)ΔXCH4(i,j) + ∈ (6)

[0078] Let t = kμ, then the residual ∈ can be expressed as:

[0079] ∈ = x m (i,j)-μ(j)-t L×1 ·ΔXCH4(i,j) (7)

[0080] In order to obtain an accurate ΔXCH4, this patent uses the least squares method to solve it. If the sum of the weighted squares of the residuals between the prior value (prior spectrum) and the observed value (observed spectrum) of the spectral intensity of each band is minimized, then the optimal estimate of the concentration enhancement of the gas molecular column is obtained, as shown in formula (8):

[0081] argmin:(∈ - Σ∈) (8)

[0082] This embodiment also provides the optimal solution for ΔXCH4, specifically: taking the logarithm of both sides of formula (5) and performing least squares operations using formulas (6), (7), and (8) to obtain the LMF matched filter detector:

[0083]

[0084] The mean and covariance matrix are updated using the column concentration enhancement value calculated by formula (9), and the iteration continues until convergence. Outliers are considered to have statistical significance exceeding twice the noise level (2σ threshold, p<0.05). The iteration terminates when there are no outliers or when the number of iterations exceeds five. Then, the mean and covariance are updated for the last time. The preprocessed hyperspectral image and the calculated prior spectrum are input into the iterative logarithmic matched filter ILMF, where each grid of the hyperspectral image is used as the observed spectrum and matched with the prior spectrum to obtain the XCH4 enhanced image, i.e., the ΔXCH4 image. The specific principle is as follows: Figure 2 As shown.

[0085] S6. Identify plume regions in XCH4 enhanced images using convolutional neural networks, including:

[0086] S61. Construct a dataset for a convolutional neural network to identify plume regions in XCH4 enhanced imagery. The dataset is a generated synthetic dataset, created by combining spaceborne imaging spectrometer images (including but not limited to Hyperion, PRISMA, and GF-5A / B AHSI imagery data from EO-1) with synthetic plumes simulated by WRF-LES. SWIR spectral radiometry and RGB bands from the spaceborne imaging spectrometer Level-1b data are used. These datasets have pixel quality and cloud masking information. Typically, 100 different spaceborne imaging spectrometer background images are used.

[0087] S62. Construct a UNet model convolutional neural network to identify feather regions in XCH4 enhanced images;

[0088] S63. Training the UNet model convolutional neural network includes:

[0089] The encoder, which captures context in hyperspectral images, consists of convolutional layers and max-pooling layers;

[0090] The decoder, which can locate the features captured by the encoder, consists of convolutional layers and upsampling layers.

[0091] In the model architecture of this patent, an additional 1×1 convolutional layer with 64 filters is introduced at the beginning because methane plumes are associated with anomalies in certain SWIR bands of hyperspectral imagery. The inclusion of visible bands helps the neural network distinguish between plumes and non-plumes by providing background information about the image. Methane plumes can be identified based on the typical spatial structure formed by atmospheric turbulence and advection, as well as the variation in methane absorption bands compared to the background landscape. It is for the latter reason that the additional 1×1 convolutional layer is considered to contribute to improving the model's accuracy. The model constructed in this patent establishes an intermediate prediction layer where the presence or absence of plumes in the image is determined by binary classification of the entire image. At each stage of the model, the input is a concatenation of the input satellite image and all previous outputs. To optimize the training of the model weights, the model is trained separately for each part, so that the weights of all other parts are not updated. The parts of the model are trained sequentially, and the loss function for predicting the plume mask is as follows, where BC is the binary cross-entropy:

[0092]

[0093] S64. The dataset is processed using PCA on the XCH4 enhanced image obtained from the inversion, specifically as follows:

[0094] The relationship between the spectral intensity of each point in the ΔXCH4 image generated in step S5 and the spectral intensity of the satellite radiance can be represented by a methane Jacobian vector, which describes the intensity I in frequency band k. k The logarithm of the scene changes with the enhancement of the methane column. The spectral changes of the scene (after cloud removal) background can be approximated by several principal components of all measured spectra obtained by applying principal component analysis (PCA). PCA is performed on the logarithms of the scene measured spectra, and the principal component singular vectors that best describe the spectral changes of the scene are selected. The number of the top three optimal singular vectors is determined. These vectors are concatenated with the methane Jacobian to construct a matrix J of dimension 4 × the number of PRISMA bands. This matrix is ​​used with the logarithm y of the measured radiant intensity to find the vector W that minimizes the cost function in a linear least squares fit for each pixel:

[0095]

[0096] S65. Input the PCA-processed data into a convolutional neural network for recognition, such as... Figure 4 As shown.

[0097] Example 2:

[0098] This embodiment provides a computer device, characterized in that it includes a processor and a memory, wherein the processor is used to execute a program for detecting methane leaks in an oil field stored in the memory, so as to implement the method for detecting methane leaks in an oil field.

[0099] An electronic device includes at least one processor, memory, at least one network interface, and other user interfaces. The various components of the electronic device are coupled together via a bus system. It is understood that the bus system is used to enable communication between these components. In addition to a data bus, the bus system also includes a power bus, a control bus, and a status signal bus.

[0100] The user interface may include a display, keyboard, or clicking device (e.g., mouse, trackball, touchpad, or touchscreen). It is understood that the memory in this embodiment may be volatile memory or non-volatile memory, or may include both.

[0101] In this embodiment of the invention, the processor executes the method steps provided in each method embodiment by calling a program or instruction stored in the memory, specifically a program or instruction stored in an application program.

[0102] In some implementations, the memory stores elements such as executable units or data structures, or subsets thereof, or extended sets thereof: operating systems and applications.

[0103] The operating system includes various system programs, such as the framework layer, core library layer, and driver layer, used to implement various basic business functions and handle hardware-based tasks. The application programs include various applications, such as media players and browsers, used to implement various application functions. The program implementing the method of this invention can be included in the application programs.

[0104] Example 3:

[0105] This embodiment provides a storage medium, characterized in that the storage medium stores one or more programs, which can be executed by one or more processors to implement the oilfield methane leakage detection method.

[0106] The method steps described in conjunction with Embodiment 1 disclosed herein can be implemented using hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0107] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting methane leaks in oil fields using satellite remote sensing data, characterized in that, The range of methane absorption spectral data was selected as the simulated spectrum; Acquire hyperspectral image data within the simulated spectral range and preprocess the hyperspectral image data to obtain the observed spectrum; Simulate the spectral absorption curve of gas under real-time conditions in the oilfield area to find the ideal reference spectral line; The selected simulated spectra and reference spectra are subjected to logarithmic difference processing to obtain prior spectral data; The observed spectrum and the prior spectrum are input into a matched filter to invert and obtain the XCH4 enhanced image; Using convolutional neural networks to identify plume regions in XCH4 enhanced images.

2. The oilfield methane leak detection method of claim 1, wherein, The range of methane absorption spectral data was selected as the simulated spectrum, specifically: Using the Monte Carlo simulation principle, the absorption spectral range with the smallest error was selected from the HITRAN database as the search window, and the optimal combination of spectral bands was selected as the simulated spectrum.

3. The oilfield methane leak detection method of claim 2, wherein, Preprocessing of hyperspectral image data includes: Acquire hyperspectral image data within the optimal combination range; Based on the information obtained from the satellite header file, radiometric calibration and atmospheric correction were performed on the hyperspectral image data. Cloud masking is used to remove clouds from hyperspectral image data, and data containing clouds are assigned a value of 0.

4. The method for detecting methane leaks in oil fields according to claim 2, characterized in that, Finding the ideal reference spectral line specifically includes: Calculate the center wavelength and full width at half maximum (FWHM) of the optimal combination range; Using the HITRAN molecular absorption library, unit gas absorption spectra at a resolution of 0.1 cm⁻¹ were obtained; Based on the full width at half maximum (FWHM), the spectral response function of the hyperspectral remote sensor band is simulated using a Gaussian function to obtain the reference spectral line for stably recorded light intensity under oilfield conditions.

5. The method for detecting methane leaks in oil fields according to claim 1, characterized in that, The matched filter is an iterative log-normal matched filter, specifically:

6. The method for detecting methane leaks in oil fields according to claim 1, characterized in that, Using convolutional neural networks to identify plume regions in XCH4 enhanced images, including: A dataset was constructed to create a convolutional neural network for identifying plume regions in XCH4 enhanced imagery; Construct a convolutional neural network to identify plume regions in XCH4 enhanced images; Train the convolutional neural network; The dataset was processed using PCA on the XCH4 enhanced images obtained from the inversion. The data processed by PCA is input into a convolutional neural network for recognition.

7. The method for detecting methane leaks in oil fields according to claim 1, characterized in that, The dataset is processed using PCA on the XCH4 enhanced images obtained from the inversion, including: Select the principal component singular vector that best describes the spectral changes of the scene; Determine the number of the first three optimal singular vectors; These vectors are concatenated with the methane Jacobian to construct a matrix J of dimension 4 × the number of PRISMA bands. This matrix is ​​then used with the logarithm y of the measured radiation intensity to find the vector W that minimizes the cost function for each pixel in a linear least squares fit.

8. The method for detecting methane leaks in oil fields according to claim 1, characterized in that, Training a convolutional neural network includes: The encoder, which captures context in hyperspectral images, consists of convolutional layers and max-pooling layers; The decoder, which can locate the features captured by the encoder, consists of convolutional layers and upsampling layers.

9. A computer device, characterized in that, include: A processor and a memory, the processor being configured to execute a program for detecting methane leaks in an oil field stored in the memory, to implement the method for detecting methane leaks in an oil field according to any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores one or more programs, which can be executed by one or more processors to implement the oilfield methane leakage detection method according to any one of claims 1 to 7.

Images