CN115576014A - Intelligent identification method for fractured reservoir based on acoustic wave remote detection imaging - Google Patents
Intelligent identification method for fractured reservoir based on acoustic wave remote detection imaging Download PDFInfo
- Publication number
- CN115576014A CN115576014A CN202211317841.6A CN202211317841A CN115576014A CN 115576014 A CN115576014 A CN 115576014A CN 202211317841 A CN202211317841 A CN 202211317841A CN 115576014 A CN115576014 A CN 115576014A
- Authority
- CN
- China
- Prior art keywords
- network
- imaging
- field
- steps
- function
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
- G01V1/40—Seismology; Seismic or acoustic prospecting or detecting specially adapted for well-logging
- G01V1/44—Seismology; Seismic or acoustic prospecting or detecting specially adapted for well-logging using generators and receivers in the same well
- G01V1/48—Processing data
- G01V1/50—Analysing data
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V2210/00—Details of seismic processing or analysis
- G01V2210/60—Analysis
- G01V2210/61—Analysis by combining or comparing a seismic data set with other data
- G01V2210/616—Data from specific type of measurement
- G01V2210/6169—Data from specific type of measurement using well-logging
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V2210/00—Details of seismic processing or analysis
- G01V2210/60—Analysis
- G01V2210/62—Physical property of subsurface
- G01V2210/624—Reservoir parameters
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
- Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
Landscapes
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Acoustics & Sound (AREA)
- Environmental & Geological Engineering (AREA)
- Geology (AREA)
- Remote Sensing (AREA)
- General Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Geophysics (AREA)
- Geophysics And Detection Of Objects (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a fracture type reservoir intelligent identification method based on sound wave remote detection imaging, belonging to the technical field of geophysical acoustic logging; the method comprises the following steps: step 1: preparing a theoretical simulation data set; step 2: manually labeling the field data set; and step 3: constructing a segmentation network model; and 4, step 4: training a neural network; and 5: network knowledge migration; step 6: and (4) processing a field imaging graph. According to the method, the fracture reflector in the imaging graph is intelligently identified through the angle of image segmentation, a segmentation network model is constructed, an end-to-end segmentation network is trained by using simulation data, the end-to-end segmentation network is migrated and learned to an actual data domain, and the fracture type reservoir in the imaging graph can be automatically identified under the condition of strong noise interference.
Description
Technical Field
The invention belongs to the technical field of geophysical acoustic logging, and particularly relates to an intelligent identification method for a fractured reservoir based on acoustic far-detection imaging.
Background
In recent years, with the exploration and development of unconventional oil and gas, more and more attention has been paid to the identification of fractured reservoirs. Acoustic remote sensing techniques can use reflection information to image high resolution of geologic structures outside of a well. However, both monopole sound sources and dipole sound sources can excite and generate direct mode waves, such as sliding longitudinal waves, sliding transverse waves, stoneley waves and bending waves, which propagate along the well wall near the well hole, and interfere with the measurement of the reflection information. Therefore, in the process of processing the acoustic far-detection signal, the direct signal needs to be suppressed by using a reflected wave extraction method. However, under the actual measurement condition, the amplitude of the reflected wave is weak, and aliasing is generated between the time domain and the direct wave in the well, so that the effect of the reflected wave extraction method is reduced, and a large amount of residual signals still exist in the processed wave field. After the subsequent offset imaging algorithm, the residual signal will appear as irregular coherent noise in the imaging image.
The Chinese patent discloses a noise reduction method of an acoustic wave remote detection imaging graph (the application number is 202210845762.6, and the application publication number is CN 115170428A), which comprises the following steps: step 1: preparing a data set; and 2, step: constructing a noise reduction network model; and step 3: preprocessing data; and 4, step 4: training a network model; and 5: and (4) processing a field imaging graph. And removing interference noise in the removed imaging graph by constructing a noise reduction neural network model.
The noise reduction method for the acoustic wave remote detection imaging graph in the patent has the following defects: there are also residual direct wave artifacts distributed vertically near the borehole that interfere with accurate identification of valid reflectors. How to weaken or eliminate the influence of direct wave artifacts on reflector identification and the explanation of effective reservoirs in an imaging graph still remains a problem to be solved, so that the popularization and application of the acoustic wave remote detection technology can be improved.
Disclosure of Invention
Aiming at the defects, the invention provides an intelligent identification method of a fractured reservoir based on sound wave remote detection imaging.
The purpose of the invention is realized as follows: a fracture type reservoir intelligent identification method based on acoustic far detection imaging is characterized in that: the method comprises the following steps:
step 1: preparing a theoretical simulation data set;
step 2: manually labeling the field data set;
and step 3: constructing a segmentation network model;
and 4, step 4: training a neural network;
and 5: network knowledge migration;
and 6: and (4) processing a field imaging graph.
Preferably, the specific operation of preparing the theoretical simulation data set in step 1 is:
step 1-1: setting the size, the longitudinal and transverse wave velocity and the density value of a uniform stratum as a background medium, setting the longitudinal and transverse wave velocity and the density value of the non-uniform medium, determining the size, the shape and the position of the non-uniform medium by using a random function, and overlapping the non-uniform area on the uniform stratum to construct a geological model;
step 1-2: rapidly calculating a scattering sound field of the fracture model beside the well by using a numerical method based on Born approximation;
step 1-3: returning the time domain waveform obtained in the step 1-2 to the declination direction of the inclined phase axis by using an offset imaging algorithm to obtain an offset imaging graph of the geological structure around the well;
step 1-4: setting a background medium in the geological model as 0 and setting a non-uniform area as 1 to obtain a corresponding label image;
step 1-5: repeating the steps 1-1, 1-2, 1-3 and 1-4 to obtain a large number of paired labels and offset imaging graphs and establishing a large-scale theoretical simulation data set.
Preferably, the specific operation of preparing the manual annotation field data set in the step 2 is as follows: and marking the collected actual imaging graph, and only marking a continuous and inclined reflector in the marking process to generate a label image only containing 0 and 1, wherein a pixel 1 represents the reflector, and a pixel 0 represents a background medium.
Preferably, the constructing of the segmentation network model in step 3 includes the following steps:
step 3-1: and (3) constructing an encoder network: an encoder network is built by adopting Effectienet-b 4 as a backbone structure, wherein the encoder network is divided into 6 stages in total, and the output characteristic of each stage is { E } 1 ,E 2 ,E 3 ,E 4 ,E 5 ,E 6 };
Step 3-2: and (3) building a decoder network: outputting E to the encoder network by using the nearest neighbor interpolation mode 6 After upsampling with E 4 Carrying out residual error connection to obtain output characteristic D 4 And then up-sampling the output stage by stage and comparing the up-sampled output with { E } 2 ,E 3 Residual error connection is carried out to respectively obtain output characteristics (D) 2 ,D 3 };
Step 3-3: building a segmentation module: output characteristics for encoder and decoder, respectively { D 3 ,D 4 ,E 6 Carry out 2,4,8 times of up-sampling operation, unify the feature size to the same size, and then combine it with feature D 2 Performing up-sampling by 4 times after common superposition, and finally obtaining a prediction segmentation graph after filtering by using a convolution kernel with the size of 1 multiplied by 1 and the step length of 1 multiplied by 1;
step 3-4: and (4) combining the encoder network, the decoder network module and the segmentation module in the steps 3-1,3-2 and 3-3, and building a characteristic pyramid network for a subsequent image segmentation task.
Preferably, the specific operation of training the neural network in step 4 is:
step 4-1: dividing data by adopting a 10-fold cross validation method, namely dividing 10% of data from a data set to be used as a test set, dividing the rest data into 10 mutually exclusive subsets with the same size, taking each subset as a validation set once, taking other subsets as training sets, repeating for ten times, and taking the average value as an evaluation result;
step 4-2: randomly sampling from the training set according to the set batch sampling number, and inputting the label graph and the offset imaging graph into the segmented neural network model established in the step 3;
step 4-3: normalizing the image:
wherein μ represents the mean of the data and σ is the standard deviation of the data;
step 4-4: performing a series of random data enhancement processing on input data in a training process, wherein the random data enhancement processing comprises horizontal and vertical turning and adding 0-30 dB of white Gaussian noise;
and 4-5: combining network output and corresponding label data, and calculating a loss function by adopting a Dice coefficient:
wherein Y is a label, f (θ) is a predicted segmentation map of the model, and # in the molecule represents an overlapping portion of the former two;
and 4-6: reversely propagating the calculated Dice loss function by using an Adam algorithm, and updating a neural network parameter;
and 4-7: and (3) repeating the steps 4-2 to 4-6 to continuously optimize the network until all data in the training set are sampled, and calculating a lou coefficient on the verification set:
and 4-8: repeating the steps 4-7 until the iteration number exceeds a set threshold, or setting the average lou coefficient increment value on the verification set in the iteration number not to exceed the set threshold;
and 4-9: re-executing the step 4-8 by using the Lov-sz-wing function as a loss function, and performing reinforced training on the network trained by the Dice coefficient to overcome the gradient disappearance phenomenon;
step 4-10: and applying the trained network to a test data set to evaluate the generalization performance of the network.
Preferably, the specific operation of the network knowledge migration in step 5 is:
step 5-1: initializing a new encoder and decoder network in the FPN network by using the parameters of the trained network obtained in the step (4), and randomly initializing the parameters of the segmentation module;
step 5-2: and (3) setting a smaller learning rate by using an Adam optimizer, and performing network optimization on the field data set established in the step (2) by using the Dice coefficient in the step (4-2) as a loss function.
Preferably, the specific operation of the processing of the field imaging graph in the step 6 is as follows:
step 6-1: solidifying and storing the FPN network model parameters trained in the step 5;
step 6-2: and (3) slicing the field imaging image to be processed, inputting the sliced field imaging image to the FPN network stored in the step 6-1, outputting the corresponding prediction segmentation image, and finally combining all output reflectors in the field imaging image.
Preferably, the specific operation of performing fast calculation on the scattering sound field of the fracture model beside the well by using a numerical method based on Born approximation in the step 1-2 is as follows:
step 1-2-1: under weak scattering conditions, the local nonuniformity xi (x) < 1 caused by nonuniform medium speed change and controls the propagation of scattering wave together with the incident wave field, and the equation is as follows:
where u is a scalar displacement potential, V 0 Background velocity for non-homogeneous media;
step 1-2-2: based on the Born approximation principle, the total sound field can be equivalent to the incident field u 0 And a scattered field u 1 Superposition of (2):
u=u 0 +u 1 (2)
wherein the incident field satisfies the homogeneous medium equation:
step 1-2-3: neglecting high order infinitesimal quantitiesThe relation equation of the scattered field and the incident field can be obtained by combining the formulas (1), (2) and (3):
step 1-2-4: further utilizing a Green function and a superposition principle to deduce a scattering sound field in the non-uniform area as follows:
where x' is a point in the inhomogeneous medium region V and G is the incident wave field u 0 A corresponding green's function, R = | x-x' | represents the distance between two points; when the scattering sound field is calculated, firstly, the non-uniform area is discretized into smaller volume units, and then the sound fields of all the subunits are superposed through a time domain integral equation method to obtain the scattering sound field of the non-uniform medium.
Preferably, the specific operation of obtaining the imaging map only containing the reflector by using the offset imaging method in the steps 1 to 3 is as follows:
step 1-3-1: reversely propagating the waveform data to the offset domain by using a Kirchhoff equation through the waveform data obtained in the step 1-2, wherein a certain point zeta in the offset domain 0 At t 0 The backward propagation sound field expression at the moment is as follows:
in the formula (I), the compound is shown in the specification,represents the boundary of the offset domain omega, whereDerivative of the Ω outer normal direction;
step 1-3-2 with Gaussian beam G GB (ζ,t;ζ 0 ,t 0 ) Replacing the Green function in the formula (6) by a superimposed asymptotic form to obtain a certain point zeta in the offset domain 0 At t 0 Time-of-day backward-propagating wavefield expression:
zeta in well s The wave equation for the direct wavefield excited by the sound source at the location is:
wherein f (t) is the acoustic source excitation function of the logging instrument, generally a Rake wavelet; the expression of the direct wavefield obtained by the same derivation steps as the above backward propagating wavefield is:
wherein Re represents a real part of a complex number, f F (ω) is the frequency domain representation of the excitation function, G GB (ζ,ζ s (ii) a ω) represents ζ from the sound source point s A Gaussian beam cluster to a point zeta in the offset domain is an asymptotic solution of a Green function in the wave equation;
asymptotic solution G of Green's function GB (ζ,ζ 0 (ii) a ω) is formed by a series of zeta points 0 And a gaussian beam emerging through the imaging point ζ, wherein the expression of a single gaussian beam ray in the ray center coordinate system is:
s is the arc length of the ray, n is the distance from a point outside the ray to the ray, upsilon(s) is the speed on the ray, p and q are kinetic parameters, and tau(s) is the arrival time on the central ray; and obtaining an asymptotic solution of the green function by superposing the Gaussian beams in different emergent directions:
where Φ (θ; ω) is the initial amplitude of the gaussian beam with an exit angle θ, the equation is as follows:
transforming the green's function represented by the frequency domain to the time domain:
1-3-3, solving the Gaussian beam by using a formula (10), calculating a direct wave field by using a formula (11) and a formula (9), further calculating a backward propagation wave field by using a formula (13) and a formula (7), and then calculating the cross-correlation of the two wave fields for imaging, wherein the formula is as follows:
I(ζ;ζ s )=∫dt 0 u D (ζ,t 0 ;ζ s )u(ζ,t 0 ) (14)
the above formula calculates the direct wave field u D (ζ,t 0 ;ζ s ) And the backward propagating wave field u (ζ, t) 0 ) The correlation between the two images is used to image the extrawell geologic structure.
Preferably, the specific operation of building the encoder network in the step 3-1 is as follows:
step 3-1-1: adopting an Effectienet-b 4 network as a backbone, wherein the structure is a continuous convolution structure, and the input of each stage is the output of the previous stage;
step 3-1-2: the core component of the above-mentioned Effecientnet-b4 network is a mobile convolution module (MBConv), which first performs channel expansion on an input feature by using a convolution kernel of 1 × 1, with an expansion coefficient r, further performs feature normalization by using a batch normalization layer (BatchNorm), and then performs nonlinear mapping by using a Swish activation function:
Swish(x)=x·sigmoid(βx) (15)
where β is a constant or trainable parameter, typically set to a constant of 1.
Step 3-1-3: in order to reduce the parameter quantity of the network, the expanded characteristics of the channels are filtered by adopting deep separable convolution, each channel of the input characteristics is respectively processed by utilizing a two-dimensional convolution kernel, the convolution graph of each channel is obtained, then the output values of a plurality of channels are mapped into a single channel by utilizing 1 multiplied by 1 convolution, and on the premise that the calculated quantity is slightly lower than that of a standard convolution kernel with the same size, richer multilevel characteristics are extracted.
Step 3-1-4: the method comprises the following steps of further refining features by utilizing a channel attention mechanism, compressing the spatial dimension of the features by utilizing global average pooling operation, and changing the spatial dimension into one-dimensional features; and then, performing channel compression on the one-dimensional features by using a convolution kernel, expanding the one-dimensional features to an original channel by using extraction operation, and finally multiplying the one-dimensional features by input features.
Step 3-1-5: and scaling the features processed by the channel attention mechanism to an original input channel by utilizing a 1 x 1 projection convolution kernel, and then performing residual connection with the original input channel to inhibit the gradient disappearance phenomenon of the deep network.
The invention has the beneficial effects that: 1. the simulation data set is used for training the network and is migrated to the actual data set, so that the problem that the actual data set is rare and the effective training of the network is difficult to support is effectively solved; meanwhile, by building a segmentation network model and transferring learning to an actual data domain, the fracture type reservoir in the imaging graph can be automatically identified under the condition of strong noise interference.
2. The invention trains an end-to-end reflector identification network, does not need complex parameter setting and expert knowledge support, can overcome noise interference in an imaging graph, intelligently identifies the reflector therein, and is beneficial to the field popularization and application of the sound wave remote detection technology.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
FIG. 2 is an offset imaging plot of a formation model.
FIG. 3 is a label diagram corresponding to the stratigraphic model.
FIG. 4 is a plot of collected field offset imaging.
FIG. 5 is a manual annotation result of a field offset imaging plot.
Fig. 6 is a schematic diagram of an Effecientnet-b4 backbone network structure.
Fig. 7 is a schematic diagram of a constructed segmentation network model.
FIG. 8 is a schematic diagram of a mobile convolution module.
Fig. 9 is a training set learning graph.
FIG. 10 is a validation set learning graph.
FIG. 11 is a diagram of the results of the processing of the trained network for live imaging.
Detailed Description
To make the objects, technical solutions and advantages of the present invention clearer, the following description is made with reference to the accompanying drawings and specific embodiments. The specific embodiments described herein are merely illustrative of the invention and are not intended to be limiting. The invention is further summarized with reference to the attached drawings.
As shown in fig. 1, a fracture type reservoir intelligent identification method based on acoustic far-detection imaging includes the following steps:
step 1: preparing a theoretical simulation data set;
step 1-1: setting homogeneous stratum as background medium, size of 20 × 20m, longitudinal wave speed of 6220m/s, transverse wave speed of 3208m/s and density of 2700kg/m 3 (ii) a The longitudinal wave velocity of the crack type inhomogeneous medium is 4010m/s, the transverse wave velocity is 2102m/s, and the density is 2650kg/m 3 (ii) a Superposing the inhomogeneous region on the homogeneous stratum to construct a geological model;
step 1-2: rapidly calculating a scattering sound field of the fracture model beside the well by using a numerical method based on Born approximation;
step 1-2-1: under weak scattering conditions, the local nonuniformity xi (x) < 1 caused by nonuniform medium speed change and controls the propagation of scattering wave together with the incident wave field, and the equation is as follows:
where u is a scalar displacement potential, V 0 Background velocity for non-homogeneous media;
step 1-2-2: based on the Born approximation principle, the total sound field can be equivalent to the incident field u 0 And a scattered field u 1 Superposition of (2):
u=u 0 +u 1 (2)
wherein the incident field satisfies the homogeneous medium equation:
step 1-2-3: neglecting high order infinitesimal quantitiesThe relation equation of the scattered field and the incident field can be obtained by combining the formulas (1), (2) and (3):
step 1-2-4: further utilizing a Green function and a superposition principle to deduce a scattering sound field in the non-uniform area as follows:
where x' is a point in the inhomogeneous medium region V and G is the incident wave field u 0 Corresponding green's function, R = | x-x' | represents the distance between two points; in the above calculation of scattered sound fieldWhen the method is used, firstly, the non-uniform area is discretized into smaller volume units, and then sound fields of all the subunits are superposed through a time domain integral equation method to obtain a scattering sound field of the non-uniform medium.
Step 1-3: returning the obtained time domain waveform to the declination direction of the same inclination phase axis by using an offset imaging algorithm to obtain a label imaging graph of the geological structure around the well;
step 1-3-1: reversely propagating the waveform data to an offset domain by using a Kirchhoff equation through the waveform data obtained in the step 1-2, wherein zeta at a certain point in the offset domain 0 At t 0 The backward propagation sound field expression at the moment is as follows:
in the formula (I), the compound is shown in the specification,represents the boundary of the offset domain omega, whereDerivative of the Ω outer normal direction;
step 1-3-2 with Gaussian beam G GB (ζ,t;ζ 0 ,t 0 ) Replacing the Green function in the formula (6) by a superimposed asymptotic form to obtain a certain point zeta in the offset domain 0 At t 0 Time-of-day backward-propagating wavefield expression:
zeta in well s The wave equation for the direct wavefield excited by the sound source at the location is:
wherein f (t) is the acoustic source excitation function of the logging instrument, generally a Rake wavelet; the expression for the direct wavefield is obtained by the same derivation steps as for the above backward propagating wavefield as:
wherein Re represents a real part of a complex number, f F (ω) is the frequency domain representation of the excitation function, G GB (ζ,ζ s (ii) a ω) represents ζ from the sound source point s A Gaussian beam cluster to a point zeta in the offset domain is an asymptotic solution of a Green function in the wave equation;
asymptotic solution G of Green's function GB (ζ,ζ 0 (ii) a ω) is formed by a series of zeta points 0 And a gaussian beam emerging through the imaging point ζ, wherein the expression of a single gaussian beam ray in the ray center coordinate system is:
s is the arc length of the ray, n is the distance from a point outside the ray to the ray, upsilon(s) is the speed on the ray, p and q are kinetic parameters, and tau(s) is the arrival time on the central ray; and obtaining an asymptotic solution of the green function by superposing the Gaussian beams in different emergent directions:
where Φ (θ; ω) is the initial amplitude of the gaussian beam with an exit angle θ, the equation is as follows:
transforming the green's function represented by the frequency domain to the time domain:
1-3-3, solving the Gaussian beam by using a formula (10), calculating a direct wave field by using a formula (11) and a formula (9), further calculating a backward propagation wave field by using a formula (13) and a formula (7), and then calculating the cross-correlation of the two wave fields for imaging, wherein the formula is as follows:
l(ζ;ζ s )=∫dt 0 u D (ζ,t 0 ;ζ s )u(ζ,t 0 ) (14)
the above formula calculates the direct wave field u D (ζ,t 0 ;ζ s ) And the backward propagating wave field u (ζ, t) 0 ) The correlation between the two images is used to image the extrawell geologic structure, and the result is shown in fig. 2.
Step 1-4: setting a background medium in the geological model as 0 and setting a non-uniform area as 1 to obtain a corresponding label image, as shown in fig. 3;
step 1-5: repeating the steps 1-1, 1-2, 1-3 and 1-4 to obtain 15000 paired label graphs and corresponding offset imaging graphs, and constructing a large-scale theoretical simulation data set.
And 2, step: manually labeling the field data set;
step 2-1: collecting a field imaging chart shown in FIG. 4;
step 2-2: the field imaging image is manually labeled, only continuous and inclined reflectors are labeled in the labeling process, a label image only containing 0 and 1 is generated, wherein the pixel 1 represents the reflector, the pixel 0 is a background medium, and the result is shown in fig. 5.
Step 2-3: slicing the field imaging image and the corresponding label image to obtain slices with the size of 256 multiplied by 256;
step 2-4: and (3) repeating the steps 2-1, 2-2 and 2-3 to obtain 1315 imaging graphs and manual label slices, and obtaining a manual labeling field data set.
And step 3: constructing a segmentation network model;
step 3-1: constructing an encoder network;
step 3-1-1: an encoder network is built by using the Effectienet-b 4 as a backbone structure, and as shown in FIG. 6, the structure comprises 6 stages in total, and the stages are pairedThe corresponding output characteristics are respectively { E } 1 ,E 2 ,E 3 ,E 4 ,E 5 ,E 6 ]. The first stage adopts a standard convolution kernel to carry out down-sampling operation, the second stage comprises 2 groups of mobile convolution modules with the size of 3 multiplied by 3 and the channel of 24 and 4 groups of mobile convolution modules with the size of 3 multiplied by 3 and the channel of 32, the third stage comprises 4 groups of mobile convolution modules with the size of 5 multiplied by 5 and the channel of 56, the fourth stage comprises 6 groups of mobile convolution modules with the size of 3 multiplied by 3 and the channel of 112, the fifth stage comprises 6 groups of mobile convolution modules with the size of 5 multiplied by 5 and the channel of 160, and the sixth stage comprises 8 groups of mobile convolution modules with the size of 5 multiplied by 5 and the channel of 272 and 2 groups of mobile convolution modules with the size of 3 multiplied by 3 and the channel of 448;
step 3-1-2: the mobile convolution module (MBConv) is a core module of the above-mentioned effecentnet-b 4 backbone network, and as shown in fig. 8, it first performs channel expansion on input features by using a convolution kernel of 1 × 1, and sets an expansion coefficient r to 6, and further performs feature normalization by using a batch normalization layer (BatchNorm), and then performs nonlinear mapping by using a Swish activation function:
Swish(x)=x·sigmoid(βx) (15)
where β is a constant or trainable parameter set to a constant of 1.
Step 3-1-3: in order to reduce the parameter quantity of the network, the expanded characteristics of the channels are filtered by adopting deep separable convolution, each channel of the input characteristics is respectively processed by utilizing a two-dimensional convolution kernel, the convolution graph of each channel is obtained, then the output values of a plurality of channels are mapped into a single channel by utilizing 1 multiplied by 1 convolution, and on the premise that the calculated quantity is slightly lower than that of a standard convolution kernel with the same size, richer multilevel characteristics are extracted.
Step 3-1-4: the method comprises the following steps of further refining features by utilizing a channel attention mechanism, compressing the spatial dimension of the features by utilizing global average pooling, and converting the spatial dimension into one-dimensional features; and then, compressing the one-dimensional feature by using a convolution kernel with a compression ratio of 8, expanding the one-dimensional feature to an original channel by using extraction operation, and finally multiplying the one-dimensional feature by the input feature.
Step 3-1-5: and scaling the features processed by the channel attention mechanism to an original input channel by utilizing a 1 x 1 projection convolution kernel, and then performing residual connection with the original input channel to inhibit the gradient disappearance phenomenon of the deep network.
Step 3-2: building a decoder network: outputting E to the encoder network by using the nearest neighbor interpolation mode 6 After upsampling with E 4 Residual error connection is carried out to obtain output characteristic D 4 And then up-sampling the output stage by stage and comparing the up-sampled output with { E } 2 ,E 3 Residual error connection is carried out to respectively obtain output characteristics (D) 2 ,D 3 };
Step 3-3: building a segmentation module: features { D } of the output of the encoder and decoder, respectively 3 ,D 4 ,E 6 Carry out 2,4,8 times of up-sampling operation, unify the feature size to the same size, and then combine it with feature D 2 Performing up-sampling by 4 times after common superposition, and finally obtaining a prediction segmentation graph after filtering by using a convolution kernel with the size of 1 multiplied by 1 and the step length of 1 multiplied by 1;
step 3-4: and (4) combining all modules in the steps 3-1,3-2,3-3, and constructing a characteristic pyramid network as shown in fig. 7 for a subsequent image segmentation task.
And 4, step 4: training a network model;
step 4-1: dividing data by adopting a 10-fold cross validation method, namely dividing 10% of data from a data set to be used as a test set, dividing the rest data into 10 mutually exclusive subsets with the same size, taking each subset as a validation set once, taking other subsets as a training set, repeating for ten times, and taking the average value as an evaluation result.
Step 4-2: randomly sampling from the training set according to the set batch sampling number, and inputting the label graph and the offset imaging graph into the segmented neural network model established in the step 3;
step 4-3: normalizing the image:
wherein μ represents the mean of the data and σ is the standard deviation of the data;
step 4-4: in the training process, a series of random data enhancement processing is carried out on input data, including horizontal and vertical turning, and 0-30 dB of white Gaussian noise is added.
And 4-5: combining network output and corresponding label data, and calculating a loss function by adopting a Dice coefficient:
where Y is a label, f (θ) is a predicted segmentation map of the model, and # in the molecule represents an overlapping portion of the former two.
And 4-6: reversely propagating the calculated Dice loss function by using an Adam algorithm, and updating a neural network parameter;
and 4-7: and (3) repeating the steps 4-2 to 4-6 to continuously optimize the network until all data in the training set are sampled, and calculating a Dice loss function and a lou coefficient on the verification set:
and 4-8: repeating the steps 4-7 until the iteration number exceeds a set threshold, or setting the average Iou coefficient increment value on the verification set in the iteration number not to exceed the set threshold; as shown by the dashed line in fig. 9, the Dice coefficient on the validation set rapidly decreases from 1 until after the 55 th round decreases to 0.02, and does not decrease, and the dashed line in fig. 10 shows the corresponding lou coefficient, and finally the Iou coefficient reaches 0.936.
And 4-9: re-executing the step 4-8 by using the Lov-sz-wing function as a loss function, and performing reinforced training on the network trained by the Dice coefficient to overcome the gradient disappearance phenomenon; the results are shown in the implementations of fig. 9 and 10, with the final Iou coefficient on the validation set reaching 0.947.
Step 4-10: and applying the trained network to a test data set to evaluate the generalization performance of the network.
And 5: network knowledge migration:
step 5-1: and (5) initializing a new encoder and decoder network in the FPN network by using the parameters of the trained network obtained in the step (4), and randomly initializing the parameters of the segmentation module.
Step 5-2: and (3) setting a smaller learning rate by using an Adam optimizer, and performing network optimization on the field data set established in the step 2 by using the Dice coefficient adopted in the step 4-5 as a loss function.
Step 6: processing a field imaging graph:
step 6-1: performing parameter solidification and storage on the FPN network model trained in the step 5;
step 6-2: and slicing the field imaging image to be processed, inputting the image to the network to generate a corresponding prediction segmentation image, and finally combining all output reflectors in the image. As shown in fig. 11, the third is the evaluation result of the network trained in step 4, which can only learn the low-level semantic meaning of reflector amplitude anomaly, but cannot effectively identify other coherent noise. And the fourth step is the evaluation result of the network trained in the step 5, so that higher semantics can be learned, the difference between random noise and the reflector can be accurately identified, the prediction result is more continuous, and on the basis of the continuous reflector prediction result, information such as the position, the inclination angle and the like of the reflector can be accurately evaluated.
The working principle is as follows: the acoustic wave remote detection well logging technology utilizes a well sound source to excite elastic waves to an underground stratum, the elastic waves are collected by a well receiver after being reflected by an underground heterogeneous body, a high-resolution imaging graph of the underground heterogeneous structure is obtained through a migration imaging algorithm, in order to intelligently identify a crack reflector in the imaging graph, firstly, a paired label and imaging graph data set is established by combining a Born approximate scattering sound field calculation method and a Gaussian beam migration imaging method, manual marking is carried out on the collected field imaging graph data set to establish a field data set, a characteristic pyramid segmentation network model is established according to the random distribution characteristics of the scale and the position of the underground heterogeneous structure, the characteristic information of the reflector in the imaging graph data set is fully extracted by utilizing a simulation data set, then, the high-level semantic information of the reflector is learned in the field imaging graph data set by utilizing migration learning, the interference of background noise is effectively suppressed on the basis, and the crack reflector in the imaging graph is intelligently identified.
The above description is only an embodiment of the present invention, and is not intended to limit the present invention. Various modifications and alterations to this invention will become apparent to those skilled in the art. Any modification, equivalent replacement, improvement or the like made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.
Claims (10)
1. A fracture type reservoir intelligent identification method based on acoustic far detection imaging is characterized in that: the method comprises the following steps:
step 1: preparing a theoretical simulation data set;
step 2: manually labeling the field data set;
and step 3: constructing a segmentation network model;
and 4, step 4: training a neural network;
and 5: network knowledge migration;
step 6: and (4) processing a field imaging graph.
2. The intelligent identification method for the fractured reservoir based on the acoustic far-detection imaging is characterized by comprising the following steps of: the specific operation of preparing the theoretical simulation data set in the step 1 is as follows:
step 1-1: setting the size, the longitudinal and transverse wave velocity and the density value of a uniform stratum as a background medium, setting the longitudinal and transverse wave velocity and the density value of the non-uniform medium, determining the size, the shape and the position of the non-uniform medium by using a random function, and overlapping the non-uniform area on the uniform stratum to construct a geological model;
step 1-2: rapidly calculating a scattering sound field of the fracture model beside the well by using a numerical method based on Born approximation;
step 1-3: returning the time domain waveform obtained in the step 1-2 to the declination direction of the inclined phase axis by using an offset imaging algorithm to obtain an offset imaging graph of the geological structure around the well;
step 1-4: setting a background medium in the geological model as 0 and setting a non-uniform area as 1 to obtain a corresponding label image;
step 1-5: repeating the steps 1-1, 1-2, 1-3 and 1-4 to obtain a large number of paired labels and offset imaging graphs and establishing a large-scale theoretical simulation data set.
3. The intelligent identification method for the fractured reservoir based on the acoustic far-detection imaging is characterized by comprising the following steps of: the specific operation of preparing the manual annotation field data set in the step 2 is as follows:
and marking the collected actual imaging graph, and only marking a continuous and inclined reflector in the marking process to generate a label image only containing 0 and 1, wherein a pixel 1 represents the reflector, and a pixel 0 represents a background medium.
4. The intelligent identification method for the fractured reservoir based on the acoustic far-detection imaging is characterized by comprising the following steps of: the step 3 of constructing the segmentation network model comprises the following steps:
step 3-1: and (3) constructing an encoder network: an encoder network is built by adopting Effectienet-b 4 as a backbone structure, wherein the encoder network is divided into 6 stages in total, and the output characteristic of each stage is { E } 1 ,E 2 ,E 3 ,E 4 ,E 5 ,E 6 };
Step 3-2: and (3) building a decoder network: outputting E to the encoder network by using the nearest neighbor interpolation mode 6 After upsampling with E 4 Residual error connection is carried out to obtain output characteristic D 4 And then up-sampling the output stage by stage and comparing the up-sampled output with { E } 2 ,E 3 Performing residual connection to obtain output characteristics (D) respectively 2 ,D 3 };
Step 3-3: build segmentationA module: output characteristics for encoder and decoder, respectively { D 3 ,D 4 ,E 6 Perform an upsampling operation 2,4,8 times, unify feature sizes to the same size, and then match it with feature D 2 Performing up-sampling by 4 times after common superposition, and finally obtaining a prediction segmentation graph after filtering by using a convolution kernel with the size of 1 multiplied by 1 and the step length of 1 multiplied by 1;
step 3-4: and (4) combining the encoder network, the decoder network module and the segmentation module in the steps 3-1,3-2 and 3-3, and building a characteristic pyramid network for a subsequent image segmentation task.
5. The intelligent fractured reservoir identification method based on acoustic far-detection imaging according to claim 4, wherein the method comprises the following steps: the specific operation of training the neural network in the step 4 is as follows:
step 4-1: dividing data by adopting a 10-fold cross validation method, namely dividing 10% of data set as a test set, dividing the rest data into 10 mutually exclusive subsets with the same size, taking each subset as a validation set, taking other subsets as a training set, repeating for ten times, and taking the average value as an evaluation result;
step 4-2: randomly sampling from the training set according to a set batch sampling number, and inputting the label graph and the offset imaging graph into the segmented neural network model established in the step 3;
step 4-3: normalizing the image:
wherein μ represents the mean of the data and σ is the standard deviation of the data;
step 4-4: performing a series of random data enhancement processing on input data in a training process, wherein the random data enhancement processing comprises horizontal and vertical turning and adding 0-30 dB of white Gaussian noise;
and 4-5: combining network output and corresponding label data, and calculating a loss function by adopting a Dice coefficient:
wherein Y is a label, f (θ) is a predicted segmentation map of the model, and # in the molecule represents an overlapping portion of the former two;
and 4-6: reversely propagating the calculated Dice loss function by using an Adam algorithm, and updating a neural network parameter;
and 4-7: and (3) repeating the steps 4-2 to 4-6 to continuously optimize the network until all data in the training set are sampled, and calculating a Iou coefficient on the verification set:
and 4-8: repeating the steps 4-7 until the iteration number exceeds a set threshold, or setting the average Iou coefficient increment value on the verification set in the iteration number not to exceed the set threshold;
and 4-9: re-executing the step 4-8 by using the Lov-sz-wing function as a loss function, and performing reinforced training on the network trained by the Dice coefficient to overcome the gradient disappearance phenomenon;
step 4-10: and applying the trained network to a test data set to evaluate the generalization performance of the network.
6. The intelligent identification method for the fractured reservoir based on the acoustic far-detection imaging is characterized by comprising the following steps of: the specific operation of the network knowledge migration in the step 5 is as follows:
step 5-1: and (4) initializing a new encoder and decoder network in the FPN network by using the parameters of the trained network obtained in the step (4), and randomly initializing the parameters of the segmentation module.
Step 5-2: and (3) setting a smaller learning rate by using an Adam optimizer, and performing network optimization on the field data set established in the step 2 by using the Dice coefficient in the step 4-2 as a loss function.
7. The intelligent identification method for the fractured reservoir based on the acoustic far-detection imaging is characterized by comprising the following steps of: the specific operation of the field imaging graph processing in the step 6 is as follows:
step 6-1: solidifying and storing the parameters of the FPN network model trained in the step 5;
step 6-2: and (3) slicing the field imaging image to be processed, inputting the sliced field imaging image to the FPN network stored in the step 6-1, outputting the corresponding prediction segmentation image, and finally combining all output reflectors in the field imaging image.
8. The intelligent identification method for the fractured reservoir based on the acoustic far-detection imaging is characterized by comprising the following steps of: the specific operation of rapidly calculating the scattering sound field of the fracture model beside the well by using a numerical method based on Born approximation in the step 1-2 is as follows:
step 1-2-1: under weak scattering conditions, the local nonuniformity xi (x) < 1 caused by nonuniform medium speed change and controls the propagation of scattering wave together with the incident wave field, and the equation is as follows:
where u is a scalar displacement potential, V 0 Background velocity for non-homogeneous media;
step 1-2-2: based on the Born approximation principle, the total sound field can be equivalent to the incident field u 0 And a scattered field u 1 Superposition of (2):
u=u 0 +u 1 (2)
wherein the incident field satisfies the homogeneous medium equation:
step 1-2-3: ignoring high order nothingPoor small quantityThe relation equation of the scattered field and the incident field can be obtained by combining the formulas (1), (2) and (3):
step 1-2-4: further utilizing a Green function and a superposition principle to deduce a scattering sound field in the non-uniform area as follows:
where x' is a point in the inhomogeneous medium region V and G is the incident wave field u 0 A corresponding green's function, R = | x-x' | represents the distance between two points; when the scattering sound field is calculated, firstly, the non-uniform area is discretized into smaller volume units, and then the sound fields of all the subunits are superposed through a time domain integral equation method to obtain the scattering sound field of the non-uniform medium.
9. The intelligent identification method for the fractured reservoir based on the acoustic far-detection imaging is characterized by comprising the following steps of: the specific operation of obtaining the imaging diagram only containing the reflector by using the offset imaging method in the steps 1 to 3 is as follows:
step 1-3-1: reversely propagating the waveform data to the offset domain by using a Kirchhoff equation through the waveform data obtained in the step 1-2, wherein a certain point zeta in the offset domain 0 At t 0 The backward propagation sound field expression at the moment is as follows:
in the formula (I), the compound is shown in the specification,represents the boundary of the offset domain omega, whereDerivative of the Ω outer normal direction;
step 1-3-2 with Gaussian beam G GB (ζ,t;ζ 0 ,t 0 ) Replacing the Green function in the formula (6) by a superimposed asymptotic form to obtain a certain point zeta in the offset domain 0 At t 0 Time-of-day backward-propagating wavefield expression:
zeta in well s The wave equation for the direct wavefield excited by the sound source at the location is:
wherein f (t) is the acoustic source excitation function of the logging instrument, generally a Rake wavelet; the expression for the direct wavefield is obtained by the same derivation steps as for the above backward propagating wavefield as:
wherein Re represents a real part of a complex number, f F (ω) is the frequency domain representation of the excitation function, G GB (ζ,ζ s (ii) a ω) represents the sound source point ζ s A Gaussian beam cluster to a point zeta in the offset domain is an asymptotic solution of a Green function in the wave equation;
asymptotic solution G of Green's function GB (ζ,ζ 0 (ii) a ω) is formed by a series of zeta points 0 And a gaussian beam emerging through the imaging point ζ, wherein the expression of a single gaussian beam ray in the ray center coordinate system is:
s is the arc length of the ray, n is the distance from a point outside the ray to the ray, upsilon(s) is the speed on the ray, p and q are kinetic parameters, and tau(s) is the arrival time on the central ray; and obtaining an asymptotic solution of the green function by superposing the Gaussian beams in different emergent directions:
G GB (ζ,ζ 0 ;ω)=∫ 0 π ω(θ;ω)u(s,n,ω)dθ (11)
where Φ (θ; ω) is the initial amplitude of the gaussian beam with an exit angle θ, the equation is as follows:
transforming the green's function represented by the frequency domain to the time domain:
1-3-3, solving the Gaussian beam by using a formula (10), calculating a direct wave field by using a formula (11) and a formula (9), further calculating a backward propagation wave field by using a formula (13) and a formula (7), and then calculating the cross-correlation of the two wave fields for imaging, wherein the formula is as follows:
I(ζ;ζ s )=∫dt 0 u D (ζ,t 0 ;ζ s )u(ζ,t 0 ) (14)
the above formula calculates the direct wave field u D (ζ,t 0 ;ζ s ) And the backward propagating wave field u (ζ, t) 0 ) The correlation between the two images is used to image the extrawell geologic structure.
10. The intelligent fractured reservoir identification method based on acoustic far-detection imaging according to claim 4, wherein the method comprises the following steps: the specific operation of constructing the encoder network in the step 3-1 is as follows:
step 3-1-1: adopting an Effectienet-b 4 network as a backbone, wherein the structure is a continuous convolution structure, and the input of each stage is the output of the previous stage;
step 3-1-2: the core component of the above-mentioned Effecientnet-b4 network is a mobile convolution module (MBConv), which first performs channel expansion on an input feature by using a convolution kernel of 1 × 1, with an expansion coefficient r, further performs feature normalization by using a batch normalization layer (BatchNorm), and then performs nonlinear mapping by using a Swish activation function:
Swish(x)=x·sigmoid(βx) (15)
where β is a constant or trainable parameter, typically set to a constant of 1;
step 3-1-3: filtering the expanded characteristics of the channels by adopting depth separable convolution, respectively processing each channel of the input characteristics by utilizing a two-dimensional convolution kernel to obtain a convolution map of each channel, mapping output values of a plurality of channels into a single channel by utilizing 1 multiplied by 1 convolution, and extracting richer multilevel characteristics on the premise that the calculated amount is slightly lower than that of a standard convolution kernel with the same size;
step 3-1-4: the method comprises the following steps of further refining features by utilizing a channel attention mechanism, compressing the spatial dimension of the features by utilizing global average pooling operation, and changing the spatial dimension into one-dimensional features; then, performing channel compression on the one-dimensional features by using a convolution kernel, expanding the one-dimensional features to an original channel by using extraction operation, and finally multiplying the one-dimensional features by input features;
step 3-1-5: and scaling the features processed by the channel attention mechanism to an original input channel by utilizing a 1 x 1 projection convolution kernel, and then performing residual connection with the original input channel to inhibit the gradient disappearance phenomenon of the deep network.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211317841.6A CN115576014B (en) | 2022-10-26 | 2022-10-26 | Crack type reservoir intelligent identification method based on acoustic wave far detection imaging |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211317841.6A CN115576014B (en) | 2022-10-26 | 2022-10-26 | Crack type reservoir intelligent identification method based on acoustic wave far detection imaging |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115576014A true CN115576014A (en) | 2023-01-06 |
CN115576014B CN115576014B (en) | 2023-07-11 |
Family
ID=84586287
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211317841.6A Active CN115576014B (en) | 2022-10-26 | 2022-10-26 | Crack type reservoir intelligent identification method based on acoustic wave far detection imaging |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115576014B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116069035A (en) * | 2023-02-10 | 2023-05-05 | 济宁鲁科检测器材有限公司 | Laser radar-based flaw detection robot weld path recognition device and method |
Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100305895A1 (en) * | 2009-05-28 | 2010-12-02 | Drange Geir A M | Distributing a clock in a subterranean survey data acquisition system |
WO2017030725A1 (en) * | 2015-08-18 | 2017-02-23 | Schlumberger Technology Corporation | Reservoir simulations with fracture networks |
CN111580166A (en) * | 2020-05-21 | 2020-08-25 | 中国石油大学(华东) | Fracture effectiveness evaluation method based on sound wave remote detection and rock mechanics |
CN111913217A (en) * | 2020-08-10 | 2020-11-10 | 中海石油(中国)有限公司 | Fracture type stratum seismic scattering wave field characteristic simulation method based on imaginary number domain Rytov approximation |
CN112711072A (en) * | 2020-12-23 | 2021-04-27 | 西安交通大学 | Res U-net-based three-dimensional seismic data fault identification method |
CN113156515A (en) * | 2021-04-16 | 2021-07-23 | 中国石油大学(华东) | Acoustic wave remote detection imaging noise reduction processing method and device |
CN113534251A (en) * | 2020-04-22 | 2021-10-22 | 中国石油天然气股份有限公司 | Seismic data fracture identification method and device based on convolutional neural network |
WO2021258096A1 (en) * | 2020-06-19 | 2021-12-23 | Exxonmobil Upstream Research Company | Classifying geologic features in seismic data through image analysis by neural network |
CN114140415A (en) * | 2021-11-25 | 2022-03-04 | 长江大学 | Ultrasonic logging image crack extraction method and device based on deep learning |
WO2022052367A1 (en) * | 2020-09-10 | 2022-03-17 | 中国科学院深圳先进技术研究院 | Neural network optimization method for remote sensing image classification, and terminal and storage medium |
CN114609682A (en) * | 2020-12-09 | 2022-06-10 | 中国石油天然气股份有限公司 | Method and system for extracting parameters of electrical imaging seam hole and computer readable storage medium |
CN115166831A (en) * | 2022-07-28 | 2022-10-11 | 中国石油大学(华东) | Acoustic logging method for guiding fracturing and sidetracking operations |
CN115170428A (en) * | 2022-07-18 | 2022-10-11 | 江苏科技大学 | Noise reduction method for acoustic wave remote detection imaging graph |
-
2022
- 2022-10-26 CN CN202211317841.6A patent/CN115576014B/en active Active
Patent Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100305895A1 (en) * | 2009-05-28 | 2010-12-02 | Drange Geir A M | Distributing a clock in a subterranean survey data acquisition system |
WO2017030725A1 (en) * | 2015-08-18 | 2017-02-23 | Schlumberger Technology Corporation | Reservoir simulations with fracture networks |
CN113534251A (en) * | 2020-04-22 | 2021-10-22 | 中国石油天然气股份有限公司 | Seismic data fracture identification method and device based on convolutional neural network |
CN111580166A (en) * | 2020-05-21 | 2020-08-25 | 中国石油大学(华东) | Fracture effectiveness evaluation method based on sound wave remote detection and rock mechanics |
WO2021258096A1 (en) * | 2020-06-19 | 2021-12-23 | Exxonmobil Upstream Research Company | Classifying geologic features in seismic data through image analysis by neural network |
CN111913217A (en) * | 2020-08-10 | 2020-11-10 | 中海石油(中国)有限公司 | Fracture type stratum seismic scattering wave field characteristic simulation method based on imaginary number domain Rytov approximation |
WO2022052367A1 (en) * | 2020-09-10 | 2022-03-17 | 中国科学院深圳先进技术研究院 | Neural network optimization method for remote sensing image classification, and terminal and storage medium |
CN114609682A (en) * | 2020-12-09 | 2022-06-10 | 中国石油天然气股份有限公司 | Method and system for extracting parameters of electrical imaging seam hole and computer readable storage medium |
CN112711072A (en) * | 2020-12-23 | 2021-04-27 | 西安交通大学 | Res U-net-based three-dimensional seismic data fault identification method |
CN113156515A (en) * | 2021-04-16 | 2021-07-23 | 中国石油大学(华东) | Acoustic wave remote detection imaging noise reduction processing method and device |
CN114140415A (en) * | 2021-11-25 | 2022-03-04 | 长江大学 | Ultrasonic logging image crack extraction method and device based on deep learning |
CN115170428A (en) * | 2022-07-18 | 2022-10-11 | 江苏科技大学 | Noise reduction method for acoustic wave remote detection imaging graph |
CN115166831A (en) * | 2022-07-28 | 2022-10-11 | 中国石油大学(华东) | Acoustic logging method for guiding fracturing and sidetracking operations |
Non-Patent Citations (2)
Title |
---|
ZHANG GONG 等: "fracture identification based on remote detection acoustic reflection logging", vol. 12, no. 4, pages 473 - 481, XP035598717, DOI: 10.1007/s11770-015-0522-0 * |
董少群 等: "人工智能在致密储层裂缝测井识别中的应用", pages 1 - 23 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116069035A (en) * | 2023-02-10 | 2023-05-05 | 济宁鲁科检测器材有限公司 | Laser radar-based flaw detection robot weld path recognition device and method |
CN116069035B (en) * | 2023-02-10 | 2023-09-05 | 济宁鲁科检测器材有限公司 | Laser radar-based flaw detection robot weld path recognition device and method |
Also Published As
Publication number | Publication date |
---|---|
CN115576014B (en) | 2023-07-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112232392B (en) | Data interpretation and identification method for three-dimensional ground penetrating radar | |
CN107688201B (en) | RBM-based seismic prestack signal clustering method | |
EA004551B1 (en) | Method for seismic facies interpretation using textural analysis and neutral networks | |
CN104614763A (en) | Method and system for inverting elastic parameters of multi-wave AVO reservoir based on reflectivity method | |
CN111562611A (en) | Semi-supervised depth learning seismic data inversion method based on wave equation drive | |
US11009617B2 (en) | Method for fast calculation of seismic attributes using artificial intelligence | |
CN114117906A (en) | Multi-scale unsupervised seismic wave velocity inversion method based on observation data self-encoding | |
CN112731522B (en) | Intelligent recognition method, device and equipment for seismic stratum and storage medium | |
CN115576014B (en) | Crack type reservoir intelligent identification method based on acoustic wave far detection imaging | |
CN116400426B (en) | Electromagnetic method-based data survey system | |
CN111929723B (en) | Velocity model super-resolution method under seismic data constraint based on multi-task learning | |
CN115345790A (en) | Ground penetrating radar image enhancement method based on window self-attention neural network | |
CN111461988B (en) | Seismic velocity model super-resolution technology based on multi-task learning | |
CN116660996B (en) | Drifting type shallow sea local earth sound parameter prediction method based on deep learning | |
CN116256722B (en) | Method and device for suppressing multiple interference of B-scan image of ground penetrating radar | |
EP3175263A1 (en) | Re-ordered interpolation and convolution for faster staggered-grid processing | |
CN115170428A (en) | Noise reduction method for acoustic wave remote detection imaging graph | |
CN116068644A (en) | Method for improving resolution and noise reduction of seismic data by using generation countermeasure network | |
US10338250B2 (en) | Method of removing incoherent noise | |
CN102012517A (en) | Underground medium imaging method and device | |
Guo et al. | Denoising and wavefield separation method for DAS VSP via deep learning | |
Lee et al. | Efficient seismic numerical modeling technique using the yolov2-based expanding domain method | |
CN116088049B (en) | Least square inverse time migration seismic imaging method and device based on wavelet transformation | |
CN113222976B (en) | Space-time image texture direction detection method and system based on DCNN and transfer learning | |
CN111722287B (en) | Seismic phase characteristic identification waveform inversion method based on progressive data assimilation method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |