CN115877454A - Method and device for generating seismic inversion data and high-resolution seismic data - Google Patents
Method and device for generating seismic inversion data and high-resolution seismic data Download PDFInfo
- Publication number
- CN115877454A CN115877454A CN202310158634.9A CN202310158634A CN115877454A CN 115877454 A CN115877454 A CN 115877454A CN 202310158634 A CN202310158634 A CN 202310158634A CN 115877454 A CN115877454 A CN 115877454A
- Authority
- CN
- China
- Prior art keywords
- seismic data
- training
- resolution
- generating
- sample
- 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
Landscapes
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention provides a method and a device for generating seismic inversion data and high-resolution seismic data. The method comprises the following steps: generating a reflection coefficient curve of each well point in the predetermined area based on the seismic data in the predetermined area; generating a training sample and a label sample based on the reflection coefficient curve; training an initial deep confidence DBN network model by using a training sample to obtain a forward propagation training result model; carrying out backward propagation training on the forward propagation training result model to obtain a high-resolution seismic data generation model; generating high-resolution seismic data corresponding to the actual seismic data to be processed by using a high-resolution seismic data generation model; and converting the high-resolution seismic data into a high-resolution impedance inversion result by using Bayesian sparse pulse inversion. The method can realize frequency increase of actual seismic data, converts the seismic data into high-resolution wave impedance, and has good application effect on the thin interbed of the complicated sedimentary sand shale.
Description
Technical Field
The application relates to the field of big data processing, in particular to a method and a device for generating seismic inversion data and high-resolution seismic data.
Background
Along with the consumption of oil and gas resources, the oil and gas data of seismic exploration are more and more complex, the exploration difficulty is increased day by day, and the requirement on the exploration precision is higher and higher. In the exploration and development process of the lithologic oil and gas reservoir, sand-shale thin interbed layers mostly develop on the front edges of an underwater diversion river channel and an delta, and the single sand layer of the reservoir is thin, generally 2-8 meters, poor in transverse connectivity, fast in lateral pinch-out, generally has the characteristics of sand-shale interbed layers, strong non-mean value and the like, and brings great challenges to the lithologic oil and gas reservoir prediction.
In the prior art, the sand-shale interbed prediction mainly depends on seismic data. The accuracy of the sandstone-shale interbed prediction is closely related to the resolution of the seismic data. In order to meet the requirement of accurate exploration and development, various methods and technologies for improving the resolution of seismic data are also brought forward. At present, the longitudinal resolution of seismic data is improved mainly in two directions, namely, high-resolution acquisition is carried out, and conventional seismic data are digitally processed, so that the frequency bandwidth of the seismic data is widened. The high-resolution seismic acquisition not only has multiplied cost and period, but also is influenced by various factors, and the resolution is still limited to a certain extent. Therefore, a processing method for performing digital processing on conventional seismic data so as to improve the resolution of the seismic data becomes one of the hot spots of current research.
Disclosure of Invention
The invention provides a method and a device for generating seismic inversion data and high-resolution seismic data, which solve the problem caused by low actual resolution of the seismic data.
In a first aspect, the present invention provides a method for generating seismic inversion data, including: generating a reflection coefficient curve of each well point in the predetermined area based on the seismic data in the predetermined area; generating a training sample and a label sample based on the well point reflection coefficient curve; training an initial deep confidence DBN network model by using a training sample to obtain a forward propagation training result model; carrying out backward propagation training on a forward propagation training result model based on the label sample to obtain a high-resolution seismic data generation model; generating high-resolution seismic data corresponding to the actual seismic data to be processed by using a high-resolution seismic data generation model; and converting the high-resolution seismic data into a high-resolution impedance inversion result by using Bayesian sparse pulse inversion.
Optionally, generating a label sample based on the well point reflection coefficient curve includes: using band-pass filters
Performing band-pass filtering on the reflection coefficient curve to obtain a label sample,
wherein the content of the first and second substances,
is frequency and/or>
Four frequencies of low cut, low pass, high pass and high cut of band-pass filtering.
Optionally, generating a training sample based on the well point reflection coefficient curve includes:
obtaining a learning sample by utilizing Ricker wavelets with different main frequencies in a preset quantity to convolution well point reflection coefficient curves; calculating the three-transient attribute of the learning sample by using Hilbert transform; then, the three-transient attribute is used as three dimensions to be expanded into a corresponding learning sample, so that a training sample corresponding to the learning sample is obtained; the formula of the Ricker wavelet is as follows:
，/>
is time, is>
Is the dominant frequency of the wavelet. />
The number and the value of the (A) are preset values,
the number and value of (A) can beThe settings are made as needed.
Optionally, the initial deep belief DBN network model is composed of a predetermined number of RBM layers and a softmax prediction classifier layer from the bottom layer to the top layer.
Optionally, performing back propagation training on the forward propagation training result model to obtain a high-resolution seismic data generation model, including: carrying out back propagation training on the forward propagation training result model by using a cost function J to obtain a high-resolution seismic data generation model; wherein, the first and the second end of the pipe are connected with each other,
In a second aspect, the present invention further provides a method for generating high resolution seismic data, including: generating a reflection coefficient curve of each well point in the predetermined area based on the seismic data in the predetermined area; generating a training sample and a label sample based on the well point reflection coefficient curve; training an initial deep confidence DBN network model by using a training sample to obtain a forward propagation training result model; carrying out backward propagation training on a forward propagation training result model based on the label sample to obtain a high-resolution seismic data generation model; and generating high-resolution seismic data corresponding to the actual seismic data to be processed by using the high-resolution seismic data generation model.
In a third aspect, the invention also provides a device, an electronic device and a computer readable storage medium corresponding to the method.
The device for generating the seismic inversion data comprises: the model acquisition module is used for generating a reflection coefficient curve of each well point in the preset area based on the seismic data in the preset area; generating a training sample and a label sample based on the well point reflection coefficient curve; training an initial deep confidence DBN network model by using a training sample to obtain a forward propagation training result model; carrying out backward propagation training on a forward propagation training result model based on the label sample to obtain a high-resolution seismic data generation model; the seismic data processing module is used for generating high-resolution seismic data corresponding to the actual seismic data to be processed by using the high-resolution seismic data generation model; and the inversion data generation module is used for converting the high-resolution seismic data into a high-resolution impedance inversion result by utilizing Bayesian sparse pulse inversion.
A high resolution seismic data generation apparatus comprising: the curve acquisition unit is used for generating a reflection coefficient curve of each well point in the preset area based on the seismic data in the preset area; the sample generating unit is used for generating a training sample and a label sample based on the well point reflection coefficient curve; the first training unit is used for training the initial deep confidence DBN network model by using the training samples to obtain a forward propagation training result model; the second training unit is used for carrying out backward propagation training on the forward propagation training result model based on the label sample to obtain a high-resolution seismic data generation model; and the data generation unit is used for generating high-resolution seismic data corresponding to the actual seismic data to be processed by using the high-resolution seismic data generation model.
According to the seismic inversion data and high-resolution seismic data generation method and device based on depth confidence DBN deep learning, an artificial intelligence deep learning technology and a big data technology can be fused into seismic physics, complex mathematical relations in the geophysical subject are gradually represented by forming more abstract high layers from low-layer features through learning training of samples, and feature representation of the samples in an original space is converted to a new feature space, so that the problem of classification or prediction is easier, and the problem caused by low actual resolution of seismic data is solved. The frequency band of the seismic data is widened through deep confidence DBN deep learning, high-resolution impedance is obtained through Bayesian sparse pulse inversion, the problem that sand-shale interbed prediction is difficult is solved, and support is provided for oil-gas exploration and development.
Drawings
FIG. 1 is a schematic flow chart of a method for generating seismic inversion data according to an embodiment of the present invention;
FIG. 2A is a schematic illustration of the generation of low and medium frequency learning samples at a well site in one embodiment of the present invention;
FIG. 2B is a schematic diagram of instantaneous amplitude learning samples generated by the Hilbert transform in one embodiment of the invention;
FIG. 2C is a schematic diagram of instantaneous phase learning samples generated by the Hilbert transform in one embodiment of the invention;
FIG. 2D is a schematic diagram of instantaneous frequency learning samples generated by the Hilbert transform in one embodiment of the present invention;
FIG. 2E is a schematic diagram of a tag sample for generating well site broadband in one embodiment of the invention;
FIG. 2F is a schematic illustration of actual seismic data in accordance with an embodiment of the present invention;
FIG. 2G is a spectrogram of actual seismic data in an embodiment of the present invention;
FIG. 2H is a schematic illustration of high resolution seismic data according to one embodiment of the invention;
FIG. 2I is a spectrogram of high resolution seismic data in an embodiment of the present invention;
FIG. 2J is a schematic diagram of the inversion results of a conventional wave impedance inversion;
fig. 2K is a schematic diagram of a high resolution deep learning wave impedance inversion result according to an embodiment of the invention.
Detailed Description
In the conventional technology, resolution-improving processing methods can be adopted, which focus on deconvolution, inverse Q filtering, spectral whitening based on spectral analysis, wavelet-compressing inversion methods, and the like, and these methods usually need to be based on a certain linear or nonlinear formula to increase high-frequency information in seismic data. The neural network has nonlinearity, fault tolerance and strong pattern recognition capability, and practice proves that the neural network is an ideal tool for solving the problems. The common neural network methods include a BP neural network, a support vector machine and the like. These learning methods have limitations that they are prone to local minima, unstable in effect, and greatly different in multiple-realization results. In recent years, deep learning has come to receive wide attention from scholars at home and abroad. With deep learning, a computer model with a multi-processing hierarchy can learn data characterization through multiple levels of abstraction. The methods promote the technical development of voice recognition, visual recognition, target detection and many other fields and achieve good effects.
The invention provides a Deep learning network based on a Deep Belief Network (DBN) as a model, a large data sample between low frequency, intermediate frequency and wide frequency (including low, intermediate and high frequency information) is constructed, a mature network is constructed through sample training and applied to frequency extraction of actual seismic data, then the seismic data with the improved resolution is used as input, and Bayesian sparse pulse inversion is used for converting the seismic data with the high resolution into wave impedance with the high resolution. The method improves the longitudinal resolution ratio while maintaining the transverse resolution ratio, and has good application effect on the sand shale thin interbed with complex deposition.
To facilitate the explanation of the claimed embodiments, the principles of the present invention and some of the concepts involved will be explained first.
The technical scheme of the invention comprises the stages of sample generation, model training and data generation.
In the sample generation stage, firstly, a well point reflection coefficient curve is constructed by utilizing seismic data, and then at least a preset amount of learning samples and label samples are constructed on the basis of the well point reflection coefficient curve; building low-frequency and intermediate-frequency learning samples by utilizing a Rake (Ricker) wavelet and a convolution of a reflection coefficient, and expanding the dimensionality of the learning samples by utilizing Hilbert (Hilbert) transformation to obtain corresponding training samples; and performing band-pass filtering on the reflection coefficient to obtain a label sample, namely obtaining a broadband label sample.
In the model training phase, firstly, training samples are used as input, and broadband labeled samples are used as prediction learning targets. Then, by utilizing a deep learning method, firstly, starting from frequency division data, training from a network bottom layer of an initial deep confidence DBN network model, and carrying out abstract representation on the data layer by layer until the top of the network, which is an unsupervised training process; and then starting from a label sample curve, gradually transmitting the error from the top to the bottom of the initial depth confidence DBN network model in a back propagation mode, and adjusting and optimizing each layer of the network layer by layer to obtain a high-resolution seismic data generation model. And then, the high-resolution seismic data generation model can be further adjusted according to the data processing result, namely, the obtained high-resolution seismic data generation model is continuously optimized in an iterative manner along with the increase of the actual seismic data and the high-resolution seismic data. The high-resolution seismic data generation model provided by the application can be used for converting narrow-band seismic data into high-resolution wide-band seismic data.
In the data generation stage, the actual seismic data can be input into a trained high-resolution seismic data generation model, so that the high-resolution seismic data can be obtained through prediction; and then, converting the high-resolution seismic data into a high-resolution impedance inversion result by using Bayesian sparse pulse inversion.
The technical scheme of the invention is further explained by combining the attached drawings. As shown in fig. 1, the technical solution in an embodiment of the present invention may include the following steps:
In embodiments of the invention, the seismic data may be post-stack seismic data. The seismic data within the predetermined area may refer to seismic data acquired from various well points within the predetermined area. Theoretically, post-stack seismic data
Is the reflection coefficient->
And seismic wavelet->
I.e.:
in practical applications, the actual post-stack seismic data may be acquired by instrumental measurements.
And 102, generating a reflection coefficient curve of each well point in a preset area based on the seismic data.
The well reflection coefficient curve can be obtained by calculating the impedance curve of each well point, and the impedance curve of each well point can be calculated by the speed curve and the density curve of the well point.
When generating a well point reflection coefficient curve, the calculation formula of the reflection coefficient for generating the reflection coefficient curve is as follows:
wherein the content of the first and second substances,
respectively two adjacent time points, is greater than or equal to>
In the shade of>
Is the velocity. The density and velocity in the seismic data for each well point can be measured by a logging tool.
And 103, generating a learning sample and a label sample based on the well point reflection coefficient curve.
After the well point reflection coefficient curve is generated, learning samples and label samples may be generated based on the well point reflection coefficient curve. The learning samples are used to further generate training samples, and the label samples are used for subsequent back propagation training.
The generation methods of the label samples and the learning samples can be generated in various ways, and the generation methods of the label samples and the learning samples are described below by taking some specific implementation manners as examples.
In one implementation, the tag samples may be obtained by band-pass filtering the reflection coefficient curve with a band-pass filter.
When the tag samples are generated using a band pass filter, the band pass filter is formulated as:
wherein the content of the first and second substances,
is frequency and/or>
The four frequencies of low cut, low pass, high pass and high cut of the band-pass filter can be selected according to the requirement.
In one implementation, the learning samples can be obtained by using Ricker wavelet convolution well point reflection coefficient curves with different dominant frequencies, wherein the number and specific frequency of the dominant frequencies can be set as required.
When a Ricker wavelet convolution well point reflection coefficient curve is used for generating a learning sample, the formula of the Ricker wavelet is as follows:
wherein the content of the first and second substances,
is time, is>
Is the dominant frequency of the wavelet. />
The number and value of (a) can be set as required.
For example, ricker wavelet dominant frequency
Can be set to>
These 6 different wavelets have dominant frequencies. Then, convolution is carried out on the main frequencies of different wavelets and the reflection coefficient curves of all well points to obtain low and intermediate frequency learning sample data->
Wherein is present>
For the number of well points, nt is the number of time sample points, as shown in the first 6 plots in the first row of the study sample in FIG. 2A. In fig. 2A, the ordinate of each sample represents time, the abscissa represents waveform, and different samples correspond to different frequencies.
The bandpass filter parameters may be set to [0,3,120,130]Namely, the four frequencies of low-cut, low-pass, high-pass and high-cut are 0Hz,3Hz,120Hz and 130Hz, respectively. Further constructing a broadband label sample
The label sample is shown in FIG. 2E as the label sample. In fig. 2E, the ordinate of the sample represents time, and the abscissa represents a waveform. The four frequencies of low cut, low pass, high pass and high cut can also take other frequency values, and the application is not limited to this.
And 104, performing dimensionality extension on the learning sample to obtain a training sample.
After learning samples are obtained, the Hilbert transformation can be used for calculating the three-transient attributes (instantaneous amplitude, instantaneous frequency and instantaneous phase) of the learning samples, and then the three-transient attributes are expanded into the corresponding learning samples as three dimensions, so that the training samples corresponding to the learning samples are obtained.
The training samples after attribute expansion are shown in fig. 2B, 2C, and 2D. In fig. 2B, 2C, 2D, the ordinate of each sample is time, the abscissa of fig. 2B represents the instantaneous amplitude, the abscissa of fig. 2C represents the instantaneous phase, the abscissa of fig. 2D represents the instantaneous frequency, and the six different samples correspond to 6 different frequencies, 10Hz, 20Hz, 30Hz, 40Hz, 50Hz and 60Hz, respectively.
And 105, training an initial deep confidence DBN network model by using the training sample to obtain a forward propagation training result model.
After the initial deep belief DBN network model is obtained, the initial deep belief DBN network model may be trained using training samples, and the training process of the initial deep belief DBN network model may include two processes of forward propagation training and backward propagation training.
After the training samples are generated, the training samples can be used for carrying out forward propagation training on the initial deep confidence DBN network model, so that a forward propagation training result model is obtained.
The initial depth confidence DBN network model can be composed of a K-layer Restricted Boltzmann Machine (RBM) layer and a softmax prediction classifier layer from the bottom layer to the top layer, wherein the K value is a positive integer, and the softmax is called a normalized exponential function. For convenience of description, the RBM layers of the initial deep belief DBN network model from bottom to top can be represented as layer 1 to layer K, respectively.
When training the initial deep confidence DBN network model by using the training samples, training each RBM layer from bottom to top layer by using a contrast divergence method, wherein the training data of the 1 st layer is the training samples, and the training data of the n-th layer is the output of the n-1 st layer, wherein n =2,3, \8230, K. That is, after completing the layer 1 training by using the training sample, using the 1 st feature data extracted from the trained layer 1 as the layer 2 training data, and then training the layer 2 by using a contrast divergence method; and after the training of the layer 2 is finished, using the layer 2 characteristic data generated based on the layer 1 characteristic data as training data, then training the layer 3 by using a contrast divergence method, and so on until the training of the layer K is finished.
And 106, performing backward propagation training on the forward propagation training result model based on the label sample to obtain a high-resolution seismic data generation model.
After the forward propagation training result model is generated, the forward propagation training result model may be further back-propagated using a cost function. The Back Propagation (BP) is a short for error back propagation, and the back propagation training is to distribute an output error to each layer to obtain an error signal of each layer, and then use the error signal as a basis for correcting a parameter or a weight in each layer, where the output error is a difference between an actual output and an expected output. In the back propagation training, the parameters or weights in the forward propagation training result model can be adjusted downward by using a random gradient to achieve the goal of minimizing the cost value (cost) of the (cost function) cost function.
In the back propagation training, the following cost function may be used:
wherein the content of the first and second substances,
is a cost function>
For the number of batches, <' >>
Number of samples for a label->
For the ith training sample, ->
Represents the ith label sample, < >>
For connecting weights>
Is serial number of->
Is selected based on the weight value corresponding to the neuron(s) in (4)>
Is serial number of->
B is a bias amount, based on the weight value corresponding to the neuron(s) in (a)>
Is serial number of->
Is correspondingly biased, is greater than or equal to>
Is serial number>
The neuron of (b) corresponds to an offset, and P is a likelihood function.
When the back propagation training is performed, a plurality of times of iterative training can be performed. For example, the iterative training may be performed for a predetermined number of times, or may be performed until the model convergence condition is satisfied without performing the iterative training for a predetermined number of times.
After obtaining the high resolution seismic data generation model, the high resolution seismic data generation model may be used to treat the actual seismic data
Processing is performed such that the corresponding high resolution seismic data->
. The difference between actual seismic data and high resolution seismic data may be as shown in fig. 2F, 2G, 2H, 2I. Wherein, fig. 2F is a schematic diagram of actual seismic data, fig. 2H is a schematic diagram of high-resolution seismic data, the abscissa in the diagram represents the track number, and the ordinate represents the time; FIG. 2G is a spectrum diagram of actual seismic data, and FIG. 2I is a spectrum diagram of seismic data at high resolutionThe frequency spectrum of seismic data is plotted with frequency on the abscissa and amplitude on the ordinate.
108, utilizing Bayesian sparse pulse inversion to invert the high-resolution seismic data
And converting into a high-resolution impedance inversion result.
After the high-resolution seismic data are generated, a Bayesian formula can be used for establishing a reflection coefficient posterior probability density by combining a likelihood function of a seismic convolution model and prior distribution of wave impedance for inversion, and the high-resolution seismic data are subjected to inversion
And converting into a high-resolution impedance inversion result.
A wave impedance model can be added in the inversion process for constraint, and low-frequency and high-frequency components of an inversion result are compensated, wherein the wave impedance model is as follows:
wherein: i denotes the wave impedance value to be inverted,
m represents the number of samples calculated,
variance, representing noise in the seismic data, based on the variance in the seismic data>
Represents the wave impedance model variance, < >>
Is high resolution seismic data, <' > is>
Is wave impedance model data, i.e. formation reflection coefficients>
Indicates the fifth->
Individual stratum reflection coefficient->
Is a wavelet matrix.
Using the same actual seismic data, the results of using conventional wave impedance inversion are shown in fig. 2J, while the results of high resolution deep learning wave impedance inversion based on the present application are shown in fig. 2K. The abscissa in the figure represents the track number and the ordinate represents time.
According to the seismic inversion data generation method based on the depth confidence DBN deep learning, an artificial intelligence deep learning technology and a big data technology can be fused into seismic physics, complex mathematical relations in the geophysical subject are gradually represented by forming a more abstract high-level representation from low-level features through learning and training of samples, and feature representation of the samples in an original space is converted into a new feature space, so that the problem of classification or prediction is easier, and the problem caused by low actual resolution of the seismic data is solved. The frequency band of the seismic data is widened through deep belief DBN deep learning, high-resolution impedance is obtained through Bayesian sparse pulse inversion, the problem that sand-shale interbed prediction is difficult is solved, and support is provided for oil-gas exploration and development.
It should be noted that, the step numbers in the above schemes are only for clearly illustrating the technical schemes, and do not represent the limitation of the technical schemes of the present invention. In practical technical applications, the method may further include more or fewer steps, or some steps or combinations of steps may form independent solutions, for example, the foregoing steps 101 to 106 may form a complete model generation method solution, and the foregoing steps 101 to 107 may form a high-resolution seismic data generation method solution, which are within the scope of the present invention.
Corresponding to the method, the invention also provides a technical scheme of a device for generating the seismic inversion data and a device for generating the high-resolution seismic data.
The apparatus for generating seismic inversion data may include: the device comprises a model acquisition module, a seismic data processing module and an inversion data generation module.
The model acquisition module is used for generating a reflection coefficient curve of each well point in a preset area based on seismic data in the preset area; generating a training sample and a label sample based on the well point reflection coefficient curve; training an initial deep confidence DBN network model by using a training sample to obtain a forward propagation training result model; carrying out backward propagation training on a forward propagation training result model based on the label sample to obtain a high-resolution seismic data generation model; the seismic data processing module is used for generating high-resolution seismic data corresponding to the actual seismic data to be processed by using the high-resolution seismic data generation model; and the inversion data generation module is used for converting the high-resolution seismic data into a high-resolution impedance inversion result by utilizing Bayesian sparse pulse inversion.
The high resolution seismic data generation apparatus may include: the device comprises a curve acquisition unit, a sample generation unit, a first training unit, a second training unit and a data generation unit.
The curve acquisition unit is used for generating a reflection coefficient curve of each well point in the preset area based on the seismic data in the preset area; the sample generating unit is used for generating a training sample and a label sample based on the well point reflection coefficient curve; the first training unit is used for training an initial deep confidence DBN network model by using a training sample to obtain a forward propagation training result model; the second training unit is used for carrying out backward propagation training on the forward propagation training result model based on the label sample to obtain a high-resolution seismic data generation model; and the data generation unit is used for generating high-resolution seismic data corresponding to the actual seismic data to be processed by using the high-resolution seismic data generation model.
In accordance with the foregoing method, the present invention also provides an electronic device, which includes a processor, coupled to a memory, configured to read instructions from the memory, and execute the method for generating seismic inversion data according to the instructions, or execute the method for generating high-resolution seismic data according to the instructions.
Corresponding to the method, the invention also provides a technical scheme of the computer readable storage medium. A computer readable storage medium comprising instructions which, when executed on a computer, cause the computer to perform the method of generating seismic inversion data as provided in the various embodiments above, or to perform the method of generating high resolution seismic data as provided in the various embodiments above.
Those skilled in the art will readily appreciate that the techniques of the embodiments of the present invention may be implemented as software plus a required general purpose hardware platform. Based on such understanding, the technical solutions in the embodiments of the present invention may be essentially or partially implemented in the form of a software product, which may be stored in a storage medium, such as ROM/RAM, magnetic disk, optical disk, etc., and includes several instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method according to the embodiments or some parts of the embodiments.
The same and similar parts in the various embodiments in this specification may be referred to each other. In particular, for the above embodiments of the present invention, since they are substantially similar to the method embodiments, the description is simple, and the relevant points can be referred to the description in the method embodiments. The above-described embodiments of the present invention do not limit the scope of the present invention.
Claims (10)
1. A method for generating seismic inversion data, comprising:
generating a reflection coefficient curve of each well point in the predetermined area based on the seismic data in the predetermined area;
generating a training sample and a label sample based on the well point reflection coefficient curve;
training an initial deep confidence DBN network model by using a training sample to obtain a forward propagation training result model;
carrying out backward propagation training on a forward propagation training result model based on the label sample to obtain a high-resolution seismic data generation model;
generating high-resolution seismic data corresponding to the actual seismic data to be processed by using a high-resolution seismic data generation model;
and converting the high-resolution seismic data into a high-resolution impedance inversion result by using Bayesian sparse pulse inversion.
2. The method of claim 1, wherein generating a label sample based on the well point reflection coefficient curve comprises:
using band-pass filters
Performing band-pass filtering on the well point reflection coefficient curve to obtain a label sample,
wherein the content of the first and second substances,
is frequency->
Four frequencies of low cut, low pass, high pass and high cut of band-pass filtering.
3. The method of claim 1, wherein generating training samples based on the well point reflection coefficient curve comprises:
utilizing Ricker wavelets with different main frequencies in a preset quantity to convolute the well point reflection coefficient curve to obtain a learning sample;
calculating the three-transient attribute of the learning sample by using Hilbert transform;
expanding the three-transient attribute as three dimensions to a corresponding learning sample so as to obtain a training sample corresponding to the learning sample;
the formula of the Ricker wavelet is as follows:
wherein is present>
Is time, is>
Is the wavelet dominant frequency>
The number and value of the above-mentioned two groups are preset values.
4. The method of claim 1, wherein the initial deep belief DBN network model consists of a predetermined number of layers of restricted boltzmann machine RBM layers and one layer of softmax predictive classifier layers from bottom to top.
5. The method of claim 1, wherein back-propagating the forward-propagated training result model to obtain the high-resolution seismic data generation model comprises:
carrying out back propagation training on the forward propagation training result model by using a cost function J to obtain a high-resolution seismic data generation model; wherein the content of the first and second substances,
6. A method of generating high resolution seismic data, comprising:
generating a reflection coefficient curve of each well point in the predetermined area based on the seismic data in the predetermined area;
generating a training sample and a label sample based on the well point reflection coefficient curve;
training an initial deep confidence DBN network model by using a training sample to obtain a forward propagation training result model;
carrying out backward propagation training on a forward propagation training result model based on the label sample to obtain a high-resolution seismic data generation model;
and generating high-resolution seismic data corresponding to the actual seismic data to be processed by using the high-resolution seismic data generation model.
7. An apparatus for generating seismic inversion data, comprising:
the model acquisition module is used for generating a reflection coefficient curve of each well point in the preset area based on the seismic data in the preset area; generating a training sample and a label sample based on the well point reflection coefficient curve; training an initial deep confidence DBN network model by using a training sample to obtain a forward propagation training result model; carrying out backward propagation training on a forward propagation training result model based on the label sample to obtain a high-resolution seismic data generation model;
the seismic data processing module is used for generating high-resolution seismic data corresponding to the actual seismic data to be processed by using the high-resolution seismic data generation model;
and the inversion data generation module is used for converting the high-resolution seismic data into a high-resolution impedance inversion result by utilizing Bayesian sparse pulse inversion.
8. An apparatus for generating high resolution seismic data, comprising:
the curve acquisition unit is used for generating a reflection coefficient curve of each well point in the preset area based on the seismic data in the preset area;
the sample generating unit is used for generating a training sample and a label sample based on the well point reflection coefficient curve;
the first training unit is used for training an initial deep confidence DBN network model by using a training sample to obtain a forward propagation training result model;
the second training unit is used for carrying out backward propagation training on the forward propagation training result model based on the label sample to obtain a high-resolution seismic data generation model;
and the data generation unit is used for generating high-resolution seismic data corresponding to the actual seismic data to be processed by using the high-resolution seismic data generation model.
9. An electronic device, comprising a processor, coupled to a memory, configured to read instructions from the memory and execute the method according to the instructions, or execute the method according to the instructions, according to any one of claims 1 to 5, or execute the method according to claim 6.
10. A computer-readable storage medium comprising instructions which, when executed on a computer, cause the computer to perform the method of any one of claims 1 to 5, or perform the method of claim 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310158634.9A CN115877454B (en) | 2023-02-24 | 2023-02-24 | Method and device for generating seismic inversion data and high-resolution seismic data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310158634.9A CN115877454B (en) | 2023-02-24 | 2023-02-24 | Method and device for generating seismic inversion data and high-resolution seismic data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115877454A true CN115877454A (en) | 2023-03-31 |
| CN115877454B CN115877454B (en) | 2023-05-12 |
Family
ID=85761618
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310158634.9A Active CN115877454B (en) | 2023-02-24 | 2023-02-24 | Method and device for generating seismic inversion data and high-resolution seismic data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115877454B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| RU2011125437A (en) * | 2011-06-20 | 2012-12-27 | Общество с ограниченной ответственностью "РН-КрасноярскНИПИнефть" | METHOD FOR DETERMINING PLACES OF OPENING WELLS IN THE DEVELOPMENT OF HYDROCARBON DEPOSITS |
| CN104280767A (en) * | 2013-07-12 | 2015-01-14 | 中国石油天然气集团公司 | Sparse-spike inversion method based on Cauchy distribution |
| CN108802812A (en) * | 2017-04-28 | 2018-11-13 | 中国石油天然气股份有限公司 | A kind of formation lithology inversion method of well shake fusion |
| US20210349227A1 (en) * | 2020-04-28 | 2021-11-11 | Xi'an Jiaotong University | Method of stripping strong reflection layer based on deep learning |
| CN115598697A (en) * | 2022-10-31 | 2023-01-13 | 中国石油大学(北京)（Cn） | Thin-layer structure high-resolution seismic inversion method, device, medium and equipment |
-
2023
- 2023-02-24 CN CN202310158634.9A patent/CN115877454B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| RU2011125437A (en) * | 2011-06-20 | 2012-12-27 | Общество с ограниченной ответственностью "РН-КрасноярскНИПИнефть" | METHOD FOR DETERMINING PLACES OF OPENING WELLS IN THE DEVELOPMENT OF HYDROCARBON DEPOSITS |
| CN104280767A (en) * | 2013-07-12 | 2015-01-14 | 中国石油天然气集团公司 | Sparse-spike inversion method based on Cauchy distribution |
| CN108802812A (en) * | 2017-04-28 | 2018-11-13 | 中国石油天然气股份有限公司 | A kind of formation lithology inversion method of well shake fusion |
| US20210349227A1 (en) * | 2020-04-28 | 2021-11-11 | Xi'an Jiaotong University | Method of stripping strong reflection layer based on deep learning |
| CN115598697A (en) * | 2022-10-31 | 2023-01-13 | 中国石油大学(北京)（Cn） | Thin-layer structure high-resolution seismic inversion method, device, medium and equipment |
Non-Patent Citations (8)
| Title |
|---|
| 叶云飞；刘春成；刘志斌等: "宽频地震数据瞬时振幅属性优势分析及在南海深水区的应用" * |
| 张兴岩;潘冬明;张华;张立霞;: "波阻抗反演在提高煤层分辨率上的应用" * |
| 张兴岩;潘冬明;张华;张立霞;: "波阻抗反演在提高煤层分辨率上的应用", 煤炭科学技术 * |
| 张国印;王志章;林承焰;王伟方;李令;李诚;: "基于小波变换和卷积神经网络的地震储层预测方法及应用", 中国石油大学学报(自然科学版) * |
| 李国和，郑阳，李莹等: "基于深度信念网络的多采样点岩性识别" * |
| 正文第246-230页: "叠后地震反演方法联合应用研究" * |
| 逄丽民;: "馆下段岩性油藏研究――以陈家庄凸起东段为例", 油气地球物理 * |
| 陆文凯,张善文: "基于频率搬移的地震资料约束测井资料外推", 地球物理学报 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115877454B (en) | 2023-05-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11226423B1 (en) | Model-driven deep learning-based seismic super-resolution inversion method | |
| CN107607992B (en) | Multi-wave matching method based on convolutional neural network | |
| CN110334580A (en) | The equipment fault classification method of changeable weight combination based on integrated increment | |
| CN110069886A (en) | The identification of cable initial failure and classification method based on VMD and CNN | |
| Sun et al. | Deep learning for low-frequency extrapolation of multicomponent data in elastic FWI | |
| CN110895348B (en) | Method, system and storage medium for extracting low-frequency information of seismic elastic impedance | |
| CN114662414B (en) | Oil reservoir production prediction method based on graph wavelet neural network model | |
| CN109409442A (en) | Convolutional neural networks model selection method in transfer learning | |
| CN111722283B (en) | Stratum velocity model building method | |
| CN104730576A (en) | Curvelet transform-based denoising method of seismic signals | |
| CN109507726A (en) | The inversion method and system of time-domain elastic wave multi-parameter Full wave shape | |
| CN114966860A (en) | Seismic data denoising method based on convolutional neural network | |
| CN115598697A (en) | Thin-layer structure high-resolution seismic inversion method, device, medium and equipment | |
| Li et al. | Using GAN priors for ultrahigh resolution seismic inversion | |
| Yang et al. | Well-log information-assisted high-resolution waveform inversion based on deep learning | |
| CN112285776B (en) | Seismic velocity automatic picking method based on deep learning | |
| CN115877454B (en) | Method and device for generating seismic inversion data and high-resolution seismic data | |
| CN115267911B (en) | Model and data driving deep learning-based earthquake multiple suppression method | |
| CN113419278B (en) | Well-seismic joint multi-target simultaneous inversion method based on state space model and support vector regression | |
| CN113253350B (en) | Porosity inversion method based on joint dictionary | |
| Taner et al. | Reservoir characterization by calibration of selforganized map clusters | |
| Luo et al. | Low-Frequency Prediction Based on Multiscale and Cross-Scale Deep Networks in Full-Waveform Inversion | |
| Zhang et al. | Broadband reconstruction of seismic signal with generative recurrent adversarial network | |
| Wu et al. | An Unsupervised Inversion Method for Seismic Brittleness Parameters Driven by the Physical Equation | |
| CN113820741B (en) | Seismic inversion initial model construction method based on deep learning |
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 |