Patent
Atlas literature
Patent
US 11,740,372 B1Patent drawings and their descriptions. Click a drawing to enlarge it.
FIG. 1B is a flow schematic diagram of a method for identifying a carbon storage box based on a GAN network according to an embodiment of the present disclo- …
FIG. 2 is an effect schematic diagram of pre-stack single- 15 shot seismic data according to an embodiment of the present disclosure;
FIG. 3 is an effect schematic diagram of denoised seismic data according to an embodiment of the present disclosure;
FIG. 4 is an effect schematic diagram of noise removed 20 according to an embodiment of the present disclosure;
FIG. 5B is a flow schematic diagram of filling through the GAN network according to an embodi- ment of the present disclosure;
FIG. 6 is an effect schematic diagram of seismic wave- 25 form data of a stable sedimentary area according to an embodiment of the present disclosure;
FIG. 7 is an effect schematic diagram of a three-dimen- sional wave impedance prediction data volume according to an embodiment of the present disclosure; and …
FIG. 8 is an effect schematic diagram of an abnormal wave impedance data volume according to an embodiment of the present disclosure.
Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.
A method for intelligently identifying a carbon storage box based on a GAN network, comprising: obtaining pre-stack single-shot seismic data and well logging data, then obtaining a near-well geological interpretation result, performing pre-stack time migra-tion and superposition on the pre-stack single-shot seismic data to obtain post-stack seismic data; building an isochronous stratigraphic framework model of a target horizon based on the post-stack seismic data; performing well-to-seismic calibration on the post-stack seismic data and the well logging data to obtain a time-depth conversion relationship; calculating a three-dimensional variance attribute volume based on the post-stack seismic data, delineating seis-mic waveform data of a stable sedimentary area, and removing seismic waveform data points in a fault zone area to obtain stable sedimentary background wave-form data; based on the stable sedimentary background waveform data, through a generator of a background waveform data filling model based on the GAN neural network, obtaining fine stable sedimentary background seismic waveform data, and then obtaining a fine stable sedi-mentary background seismic waveform data inver-tomer, wherein the background waveform data filling model based on the GAN neural network is built by a generator and discriminator; based on the well logging data, the post-stack seismic data and the time-depth conversion relationship, obtaining a three-dimensional wave impedance prediction data vol-ume through a wave impedance value prediction model based on a cross-well seismic waveform structure; calculating the difference between the fine stable sedi-mentary background seismic waveform data inver-tomer and the three-dimensional wave impedance prediction data volume to obtain an abnormal wave impedance data volume; by removing areas lower than the average value in the three-dimensional variance attribute volume, retaining the abnormal wave impedance data in the spatial geo-metric contour of the fault zone to obtain a carbon storage box wave impedance data volume including the geometric structure and internal wave impedance char-acteristics of a carbon storage box; comparing the near-well geological interpretation result with the carbon storage box wave impedance data volume, delineating a characteristic value interval of a hole reservoir bed, a characteristic value interval of a transition zone, and a characteristic value interval of surrounding rock, and obtaining a carbon storage box interpretation model; and based on the carbon storage box interpretation model, obtaining the dredging situation of the carbon storage box, and then obtaining the carbon sequestration box evaluation.
A method for identifying a carbon storage box reservoir bed based on the GAN network of claim 1, wherein a method of obtaining the post-stack seismic data comprises: based on the single-shot seismic data, performing denois-ing to obtain denoised seismic data, which specifically comprises: encoding the single-shot seismic data by a convolution antoencoder, extracting hidden characteristics; the convolution antoencoder is: h"k=ù(Wk1*x+bk1) where, x represents the single-shot seismic data, a con-volution layer extracts hidden characteristics of the single-shot seismic data through multiple convolution kernels, Wk₁ represents a weight matrix of a k1-th convolution kernel, bk1 represents the offset of the k1-th convolution kernel, * represents convolution operation, ù represents a pooling function of the encoder, and h"k1 represents the hidden characteristics extracted by the k1-th convolution kernel; decoding and rebuilding the hidden characteristics by the decoder: xˆ=ΣH(W'k2*g(h"k2)+b'k2), where, g represents a sampling function on the decoder, W'k2 represents a weight matrix of a k2-th convolution kernel in the decoder, b'k2 represents the offset of the k2-th convolution layer, and the decoder decodes and rebuilds the hidden characteristics, and merges rebuilt results into denoised seismic data; and performing pre-stack time migration and superposition on the denoised seismic data to obtain the post-stack seismic data.
A method for intelligently identifying a carbon storage box based on a GAN network of claim 1, wherein a method of obtaining the time-depth conversion relationship com-prises: based on the post-stack seismic data, tracing peak points of a reflection event, constructing a continuous surface of the reflection event, and then determining the reflec-tion event where the marker layer is located to build the isochronous three-dimensional distribution of the marker layer; performing product operation based on a sonic time difference curve and a density curve in the well logging data of each known well site to obtain a wave imped-ance curve, and further calculating a reflection coeffi-cient curve; building a Ricker wavelet on the basis of the main seismic frequency of a target interval, and performing convo-lution calculation of the Ricker wavelet and the reflec-tion coefficient curve to obtain a synthetic seismic record; making the depth data of the maker layer at a wellbore of each drilling well position model correspond to a three-dimensional distribution model of the maker layer, calculating the correlation between the synthetic seismic record and the post-stack seismic data of a seismic trace near the well, and when the waveform correlation is higher than the first correlation threshold, the well-to-seismic calibration is completed to finally obtain the time-depth conversion relationship between the well logging depth and the two-way travel time of seismic reflection waves; Td=THo+2Σn i=1Ti·∆H, where, THo represents the two-way travel time of the seismic data corresponding to the depth of a sonic well logging marker layer; Ti represents sonic time differ-ence; ∆H represents a well logging curve data sampling interval; and Td represents the two-way travel time of a seismic wave.
A method for identifying a carbon storage box based on a GAN network of claim 1, wherein a method of obtaining the seismic waveform data of the stable sedimentary area comprises: based on the post-stack seismic data, calculating the seismic waveform variance attribute data volume: letting the data of each sampling point in the post-stack seismic data be Sijk, p represents a seismic gird wire size, q represents a seismic grid trace number, and k represents a sampling point serial number of a seismic record sampled at 1 ms; calculating a mean square error of sampling point data in a preset sampling area: p+1 q+1 k+1 p+1 q+1 k+1 2 Qpqk = Spqk-1/9 Spqk p-1 q-1 k-1 p-1 q-1 k-1 translating the sampling areas vertically and horizontally, transversely calculating the data mean square errors of all the sampling areas to obtain a three-dimensional variance attribute volume; and slicing the three-dimensional variance attribute volume, obtaining the distribution characteristics of variance attribute data on a plane, and taking the area where the variance attribute value is lower than the total energy average value as the sedimentary stable area, wherein data corresponding to the sedimentary stable area is the stable sedimentary background seismic waveform data.
A method for identifying a carbon storage box based on a GAN network of claim 1, wherein the background wave-form data filling model based on the GAN neural network specially comprises: a generator and a discriminator; wherein the generator comprises four groups of down-sampling convolution layers and pooling layers, a full connection layer and four up-sampling convolution layers, and four kinds of convolution kernels with the scale of 5×5 are used in each down-sampling convo-lution layer; and the discriminator comprises four up-sampling convolu-tion layers.
Patents and literature cited by this patent (applicant and examiner references).
Cited patents · 6
Patent
Atlas literature
Patent
US 11,740,372 B1Patent drawings and their descriptions. Click a drawing to enlarge it.
FIG. 1B is a flow schematic diagram of a method for identifying a carbon storage box based on a GAN network according to an embodiment of the present disclo- …
FIG. 2 is an effect schematic diagram of pre-stack single- 15 shot seismic data according to an embodiment of the present disclosure;
FIG. 3 is an effect schematic diagram of denoised seismic data according to an embodiment of the present disclosure;
FIG. 4 is an effect schematic diagram of noise removed 20 according to an embodiment of the present disclosure;
FIG. 5B is a flow schematic diagram of filling through the GAN network according to an embodi- ment of the present disclosure;
FIG. 6 is an effect schematic diagram of seismic wave- 25 form data of a stable sedimentary area according to an embodiment of the present disclosure;
FIG. 7 is an effect schematic diagram of a three-dimen- sional wave impedance prediction data volume according to an embodiment of the present disclosure; and …
FIG. 8 is an effect schematic diagram of an abnormal wave impedance data volume according to an embodiment of the present disclosure.
Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.
A method for intelligently identifying a carbon storage box based on a GAN network, comprising: obtaining pre-stack single-shot seismic data and well logging data, then obtaining a near-well geological interpretation result, performing pre-stack time migra-tion and superposition on the pre-stack single-shot seismic data to obtain post-stack seismic data; building an isochronous stratigraphic framework model of a target horizon based on the post-stack seismic data; performing well-to-seismic calibration on the post-stack seismic data and the well logging data to obtain a time-depth conversion relationship; calculating a three-dimensional variance attribute volume based on the post-stack seismic data, delineating seis-mic waveform data of a stable sedimentary area, and removing seismic waveform data points in a fault zone area to obtain stable sedimentary background wave-form data; based on the stable sedimentary background waveform data, through a generator of a background waveform data filling model based on the GAN neural network, obtaining fine stable sedimentary background seismic waveform data, and then obtaining a fine stable sedi-mentary background seismic waveform data inver-tomer, wherein the background waveform data filling model based on the GAN neural network is built by a generator and discriminator; based on the well logging data, the post-stack seismic data and the time-depth conversion relationship, obtaining a three-dimensional wave impedance prediction data vol-ume through a wave impedance value prediction model based on a cross-well seismic waveform structure; calculating the difference between the fine stable sedi-mentary background seismic waveform data inver-tomer and the three-dimensional wave impedance prediction data volume to obtain an abnormal wave impedance data volume; by removing areas lower than the average value in the three-dimensional variance attribute volume, retaining the abnormal wave impedance data in the spatial geo-metric contour of the fault zone to obtain a carbon storage box wave impedance data volume including the geometric structure and internal wave impedance char-acteristics of a carbon storage box; comparing the near-well geological interpretation result with the carbon storage box wave impedance data volume, delineating a characteristic value interval of a hole reservoir bed, a characteristic value interval of a transition zone, and a characteristic value interval of surrounding rock, and obtaining a carbon storage box interpretation model; and based on the carbon storage box interpretation model, obtaining the dredging situation of the carbon storage box, and then obtaining the carbon sequestration box evaluation.
A method for identifying a carbon storage box reservoir bed based on the GAN network of claim 1, wherein a method of obtaining the post-stack seismic data comprises: based on the single-shot seismic data, performing denois-ing to obtain denoised seismic data, which specifically comprises: encoding the single-shot seismic data by a convolution antoencoder, extracting hidden characteristics; the convolution antoencoder is: h"k=ù(Wk1*x+bk1) where, x represents the single-shot seismic data, a con-volution layer extracts hidden characteristics of the single-shot seismic data through multiple convolution kernels, Wk₁ represents a weight matrix of a k1-th convolution kernel, bk1 represents the offset of the k1-th convolution kernel, * represents convolution operation, ù represents a pooling function of the encoder, and h"k1 represents the hidden characteristics extracted by the k1-th convolution kernel; decoding and rebuilding the hidden characteristics by the decoder: xˆ=ΣH(W'k2*g(h"k2)+b'k2), where, g represents a sampling function on the decoder, W'k2 represents a weight matrix of a k2-th convolution kernel in the decoder, b'k2 represents the offset of the k2-th convolution layer, and the decoder decodes and rebuilds the hidden characteristics, and merges rebuilt results into denoised seismic data; and performing pre-stack time migration and superposition on the denoised seismic data to obtain the post-stack seismic data.
A method for intelligently identifying a carbon storage box based on a GAN network of claim 1, wherein a method of obtaining the time-depth conversion relationship com-prises: based on the post-stack seismic data, tracing peak points of a reflection event, constructing a continuous surface of the reflection event, and then determining the reflec-tion event where the marker layer is located to build the isochronous three-dimensional distribution of the marker layer; performing product operation based on a sonic time difference curve and a density curve in the well logging data of each known well site to obtain a wave imped-ance curve, and further calculating a reflection coeffi-cient curve; building a Ricker wavelet on the basis of the main seismic frequency of a target interval, and performing convo-lution calculation of the Ricker wavelet and the reflec-tion coefficient curve to obtain a synthetic seismic record; making the depth data of the maker layer at a wellbore of each drilling well position model correspond to a three-dimensional distribution model of the maker layer, calculating the correlation between the synthetic seismic record and the post-stack seismic data of a seismic trace near the well, and when the waveform correlation is higher than the first correlation threshold, the well-to-seismic calibration is completed to finally obtain the time-depth conversion relationship between the well logging depth and the two-way travel time of seismic reflection waves; Td=THo+2Σn i=1Ti·∆H, where, THo represents the two-way travel time of the seismic data corresponding to the depth of a sonic well logging marker layer; Ti represents sonic time differ-ence; ∆H represents a well logging curve data sampling interval; and Td represents the two-way travel time of a seismic wave.
A method for identifying a carbon storage box based on a GAN network of claim 1, wherein a method of obtaining the seismic waveform data of the stable sedimentary area comprises: based on the post-stack seismic data, calculating the seismic waveform variance attribute data volume: letting the data of each sampling point in the post-stack seismic data be Sijk, p represents a seismic gird wire size, q represents a seismic grid trace number, and k represents a sampling point serial number of a seismic record sampled at 1 ms; calculating a mean square error of sampling point data in a preset sampling area: p+1 q+1 k+1 p+1 q+1 k+1 2 Qpqk = Spqk-1/9 Spqk p-1 q-1 k-1 p-1 q-1 k-1 translating the sampling areas vertically and horizontally, transversely calculating the data mean square errors of all the sampling areas to obtain a three-dimensional variance attribute volume; and slicing the three-dimensional variance attribute volume, obtaining the distribution characteristics of variance attribute data on a plane, and taking the area where the variance attribute value is lower than the total energy average value as the sedimentary stable area, wherein data corresponding to the sedimentary stable area is the stable sedimentary background seismic waveform data.
A method for identifying a carbon storage box based on a GAN network of claim 1, wherein the background wave-form data filling model based on the GAN neural network specially comprises: a generator and a discriminator; wherein the generator comprises four groups of down-sampling convolution layers and pooling layers, a full connection layer and four up-sampling convolution layers, and four kinds of convolution kernels with the scale of 5×5 are used in each down-sampling convo-lution layer; and the discriminator comprises four up-sampling convolu-tion layers.
Patents and literature cited by this patent (applicant and examiner references).
Cited patents · 6
Patent
Atlas literature
Patent
US 11,740,372 B1Patent drawings and their descriptions. Click a drawing to enlarge it.
FIG. 1B is a flow schematic diagram of a method for identifying a carbon storage box based on a GAN network according to an embodiment of the present disclo- …
FIG. 2 is an effect schematic diagram of pre-stack single- 15 shot seismic data according to an embodiment of the present disclosure;
FIG. 3 is an effect schematic diagram of denoised seismic data according to an embodiment of the present disclosure;
FIG. 4 is an effect schematic diagram of noise removed 20 according to an embodiment of the present disclosure;
FIG. 5B is a flow schematic diagram of filling through the GAN network according to an embodi- ment of the present disclosure;
FIG. 6 is an effect schematic diagram of seismic wave- 25 form data of a stable sedimentary area according to an embodiment of the present disclosure;
FIG. 7 is an effect schematic diagram of a three-dimen- sional wave impedance prediction data volume according to an embodiment of the present disclosure; and …
FIG. 8 is an effect schematic diagram of an abnormal wave impedance data volume according to an embodiment of the present disclosure.
Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.
A method for intelligently identifying a carbon storage box based on a GAN network, comprising: obtaining pre-stack single-shot seismic data and well logging data, then obtaining a near-well geological interpretation result, performing pre-stack time migra-tion and superposition on the pre-stack single-shot seismic data to obtain post-stack seismic data; building an isochronous stratigraphic framework model of a target horizon based on the post-stack seismic data; performing well-to-seismic calibration on the post-stack seismic data and the well logging data to obtain a time-depth conversion relationship; calculating a three-dimensional variance attribute volume based on the post-stack seismic data, delineating seis-mic waveform data of a stable sedimentary area, and removing seismic waveform data points in a fault zone area to obtain stable sedimentary background wave-form data; based on the stable sedimentary background waveform data, through a generator of a background waveform data filling model based on the GAN neural network, obtaining fine stable sedimentary background seismic waveform data, and then obtaining a fine stable sedi-mentary background seismic waveform data inver-tomer, wherein the background waveform data filling model based on the GAN neural network is built by a generator and discriminator; based on the well logging data, the post-stack seismic data and the time-depth conversion relationship, obtaining a three-dimensional wave impedance prediction data vol-ume through a wave impedance value prediction model based on a cross-well seismic waveform structure; calculating the difference between the fine stable sedi-mentary background seismic waveform data inver-tomer and the three-dimensional wave impedance prediction data volume to obtain an abnormal wave impedance data volume; by removing areas lower than the average value in the three-dimensional variance attribute volume, retaining the abnormal wave impedance data in the spatial geo-metric contour of the fault zone to obtain a carbon storage box wave impedance data volume including the geometric structure and internal wave impedance char-acteristics of a carbon storage box; comparing the near-well geological interpretation result with the carbon storage box wave impedance data volume, delineating a characteristic value interval of a hole reservoir bed, a characteristic value interval of a transition zone, and a characteristic value interval of surrounding rock, and obtaining a carbon storage box interpretation model; and based on the carbon storage box interpretation model, obtaining the dredging situation of the carbon storage box, and then obtaining the carbon sequestration box evaluation.
A method for identifying a carbon storage box reservoir bed based on the GAN network of claim 1, wherein a method of obtaining the post-stack seismic data comprises: based on the single-shot seismic data, performing denois-ing to obtain denoised seismic data, which specifically comprises: encoding the single-shot seismic data by a convolution antoencoder, extracting hidden characteristics; the convolution antoencoder is: h"k=ù(Wk1*x+bk1) where, x represents the single-shot seismic data, a con-volution layer extracts hidden characteristics of the single-shot seismic data through multiple convolution kernels, Wk₁ represents a weight matrix of a k1-th convolution kernel, bk1 represents the offset of the k1-th convolution kernel, * represents convolution operation, ù represents a pooling function of the encoder, and h"k1 represents the hidden characteristics extracted by the k1-th convolution kernel; decoding and rebuilding the hidden characteristics by the decoder: xˆ=ΣH(W'k2*g(h"k2)+b'k2), where, g represents a sampling function on the decoder, W'k2 represents a weight matrix of a k2-th convolution kernel in the decoder, b'k2 represents the offset of the k2-th convolution layer, and the decoder decodes and rebuilds the hidden characteristics, and merges rebuilt results into denoised seismic data; and performing pre-stack time migration and superposition on the denoised seismic data to obtain the post-stack seismic data.
A method for intelligently identifying a carbon storage box based on a GAN network of claim 1, wherein a method of obtaining the time-depth conversion relationship com-prises: based on the post-stack seismic data, tracing peak points of a reflection event, constructing a continuous surface of the reflection event, and then determining the reflec-tion event where the marker layer is located to build the isochronous three-dimensional distribution of the marker layer; performing product operation based on a sonic time difference curve and a density curve in the well logging data of each known well site to obtain a wave imped-ance curve, and further calculating a reflection coeffi-cient curve; building a Ricker wavelet on the basis of the main seismic frequency of a target interval, and performing convo-lution calculation of the Ricker wavelet and the reflec-tion coefficient curve to obtain a synthetic seismic record; making the depth data of the maker layer at a wellbore of each drilling well position model correspond to a three-dimensional distribution model of the maker layer, calculating the correlation between the synthetic seismic record and the post-stack seismic data of a seismic trace near the well, and when the waveform correlation is higher than the first correlation threshold, the well-to-seismic calibration is completed to finally obtain the time-depth conversion relationship between the well logging depth and the two-way travel time of seismic reflection waves; Td=THo+2Σn i=1Ti·∆H, where, THo represents the two-way travel time of the seismic data corresponding to the depth of a sonic well logging marker layer; Ti represents sonic time differ-ence; ∆H represents a well logging curve data sampling interval; and Td represents the two-way travel time of a seismic wave.
A method for identifying a carbon storage box based on a GAN network of claim 1, wherein a method of obtaining the seismic waveform data of the stable sedimentary area comprises: based on the post-stack seismic data, calculating the seismic waveform variance attribute data volume: letting the data of each sampling point in the post-stack seismic data be Sijk, p represents a seismic gird wire size, q represents a seismic grid trace number, and k represents a sampling point serial number of a seismic record sampled at 1 ms; calculating a mean square error of sampling point data in a preset sampling area: p+1 q+1 k+1 p+1 q+1 k+1 2 Qpqk = Spqk-1/9 Spqk p-1 q-1 k-1 p-1 q-1 k-1 translating the sampling areas vertically and horizontally, transversely calculating the data mean square errors of all the sampling areas to obtain a three-dimensional variance attribute volume; and slicing the three-dimensional variance attribute volume, obtaining the distribution characteristics of variance attribute data on a plane, and taking the area where the variance attribute value is lower than the total energy average value as the sedimentary stable area, wherein data corresponding to the sedimentary stable area is the stable sedimentary background seismic waveform data.
A method for identifying a carbon storage box based on a GAN network of claim 1, wherein the background wave-form data filling model based on the GAN neural network specially comprises: a generator and a discriminator; wherein the generator comprises four groups of down-sampling convolution layers and pooling layers, a full connection layer and four up-sampling convolution layers, and four kinds of convolution kernels with the scale of 5×5 are used in each down-sampling convo-lution layer; and the discriminator comprises four up-sampling convolu-tion layers.
Patents and literature cited by this patent (applicant and examiner references).
Cited patents · 6
Patent
Atlas literature
Patent
US 11,740,372 B1Patent drawings and their descriptions. Click a drawing to enlarge it.
FIG. 1B is a flow schematic diagram of a method for identifying a carbon storage box based on a GAN network according to an embodiment of the present disclo- …
FIG. 2 is an effect schematic diagram of pre-stack single- 15 shot seismic data according to an embodiment of the present disclosure;
FIG. 3 is an effect schematic diagram of denoised seismic data according to an embodiment of the present disclosure;
FIG. 4 is an effect schematic diagram of noise removed 20 according to an embodiment of the present disclosure;
FIG. 5B is a flow schematic diagram of filling through the GAN network according to an embodi- ment of the present disclosure;
FIG. 6 is an effect schematic diagram of seismic wave- 25 form data of a stable sedimentary area according to an embodiment of the present disclosure;
FIG. 7 is an effect schematic diagram of a three-dimen- sional wave impedance prediction data volume according to an embodiment of the present disclosure; and …
FIG. 8 is an effect schematic diagram of an abnormal wave impedance data volume according to an embodiment of the present disclosure.
Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.
A method for intelligently identifying a carbon storage box based on a GAN network, comprising: obtaining pre-stack single-shot seismic data and well logging data, then obtaining a near-well geological interpretation result, performing pre-stack time migra-tion and superposition on the pre-stack single-shot seismic data to obtain post-stack seismic data; building an isochronous stratigraphic framework model of a target horizon based on the post-stack seismic data; performing well-to-seismic calibration on the post-stack seismic data and the well logging data to obtain a time-depth conversion relationship; calculating a three-dimensional variance attribute volume based on the post-stack seismic data, delineating seis-mic waveform data of a stable sedimentary area, and removing seismic waveform data points in a fault zone area to obtain stable sedimentary background wave-form data; based on the stable sedimentary background waveform data, through a generator of a background waveform data filling model based on the GAN neural network, obtaining fine stable sedimentary background seismic waveform data, and then obtaining a fine stable sedi-mentary background seismic waveform data inver-tomer, wherein the background waveform data filling model based on the GAN neural network is built by a generator and discriminator; based on the well logging data, the post-stack seismic data and the time-depth conversion relationship, obtaining a three-dimensional wave impedance prediction data vol-ume through a wave impedance value prediction model based on a cross-well seismic waveform structure; calculating the difference between the fine stable sedi-mentary background seismic waveform data inver-tomer and the three-dimensional wave impedance prediction data volume to obtain an abnormal wave impedance data volume; by removing areas lower than the average value in the three-dimensional variance attribute volume, retaining the abnormal wave impedance data in the spatial geo-metric contour of the fault zone to obtain a carbon storage box wave impedance data volume including the geometric structure and internal wave impedance char-acteristics of a carbon storage box; comparing the near-well geological interpretation result with the carbon storage box wave impedance data volume, delineating a characteristic value interval of a hole reservoir bed, a characteristic value interval of a transition zone, and a characteristic value interval of surrounding rock, and obtaining a carbon storage box interpretation model; and based on the carbon storage box interpretation model, obtaining the dredging situation of the carbon storage box, and then obtaining the carbon sequestration box evaluation.
A method for identifying a carbon storage box reservoir bed based on the GAN network of claim 1, wherein a method of obtaining the post-stack seismic data comprises: based on the single-shot seismic data, performing denois-ing to obtain denoised seismic data, which specifically comprises: encoding the single-shot seismic data by a convolution antoencoder, extracting hidden characteristics; the convolution antoencoder is: h"k=ù(Wk1*x+bk1) where, x represents the single-shot seismic data, a con-volution layer extracts hidden characteristics of the single-shot seismic data through multiple convolution kernels, Wk₁ represents a weight matrix of a k1-th convolution kernel, bk1 represents the offset of the k1-th convolution kernel, * represents convolution operation, ù represents a pooling function of the encoder, and h"k1 represents the hidden characteristics extracted by the k1-th convolution kernel; decoding and rebuilding the hidden characteristics by the decoder: xˆ=ΣH(W'k2*g(h"k2)+b'k2), where, g represents a sampling function on the decoder, W'k2 represents a weight matrix of a k2-th convolution kernel in the decoder, b'k2 represents the offset of the k2-th convolution layer, and the decoder decodes and rebuilds the hidden characteristics, and merges rebuilt results into denoised seismic data; and performing pre-stack time migration and superposition on the denoised seismic data to obtain the post-stack seismic data.
A method for intelligently identifying a carbon storage box based on a GAN network of claim 1, wherein a method of obtaining the time-depth conversion relationship com-prises: based on the post-stack seismic data, tracing peak points of a reflection event, constructing a continuous surface of the reflection event, and then determining the reflec-tion event where the marker layer is located to build the isochronous three-dimensional distribution of the marker layer; performing product operation based on a sonic time difference curve and a density curve in the well logging data of each known well site to obtain a wave imped-ance curve, and further calculating a reflection coeffi-cient curve; building a Ricker wavelet on the basis of the main seismic frequency of a target interval, and performing convo-lution calculation of the Ricker wavelet and the reflec-tion coefficient curve to obtain a synthetic seismic record; making the depth data of the maker layer at a wellbore of each drilling well position model correspond to a three-dimensional distribution model of the maker layer, calculating the correlation between the synthetic seismic record and the post-stack seismic data of a seismic trace near the well, and when the waveform correlation is higher than the first correlation threshold, the well-to-seismic calibration is completed to finally obtain the time-depth conversion relationship between the well logging depth and the two-way travel time of seismic reflection waves; Td=THo+2Σn i=1Ti·∆H, where, THo represents the two-way travel time of the seismic data corresponding to the depth of a sonic well logging marker layer; Ti represents sonic time differ-ence; ∆H represents a well logging curve data sampling interval; and Td represents the two-way travel time of a seismic wave.
A method for identifying a carbon storage box based on a GAN network of claim 1, wherein a method of obtaining the seismic waveform data of the stable sedimentary area comprises: based on the post-stack seismic data, calculating the seismic waveform variance attribute data volume: letting the data of each sampling point in the post-stack seismic data be Sijk, p represents a seismic gird wire size, q represents a seismic grid trace number, and k represents a sampling point serial number of a seismic record sampled at 1 ms; calculating a mean square error of sampling point data in a preset sampling area: p+1 q+1 k+1 p+1 q+1 k+1 2 Qpqk = Spqk-1/9 Spqk p-1 q-1 k-1 p-1 q-1 k-1 translating the sampling areas vertically and horizontally, transversely calculating the data mean square errors of all the sampling areas to obtain a three-dimensional variance attribute volume; and slicing the three-dimensional variance attribute volume, obtaining the distribution characteristics of variance attribute data on a plane, and taking the area where the variance attribute value is lower than the total energy average value as the sedimentary stable area, wherein data corresponding to the sedimentary stable area is the stable sedimentary background seismic waveform data.
A method for identifying a carbon storage box based on a GAN network of claim 1, wherein the background wave-form data filling model based on the GAN neural network specially comprises: a generator and a discriminator; wherein the generator comprises four groups of down-sampling convolution layers and pooling layers, a full connection layer and four up-sampling convolution layers, and four kinds of convolution kernels with the scale of 5×5 are used in each down-sampling convo-lution layer; and the discriminator comprises four up-sampling convolu-tion layers.
Patents and literature cited by this patent (applicant and examiner references).
Cited patents · 6
Cited non-patent literature · 3
Cited non-patent literature · 3
Cited non-patent literature · 3
Cited non-patent literature · 3
