CN111487679B - Transverse wave velocity prediction method, device and equipment - Google Patents
Transverse wave velocity prediction method, device and equipment Download PDFInfo
- Publication number
- CN111487679B CN111487679B CN202010321843.7A CN202010321843A CN111487679B CN 111487679 B CN111487679 B CN 111487679B CN 202010321843 A CN202010321843 A CN 202010321843A CN 111487679 B CN111487679 B CN 111487679B
- Authority
- CN
- China
- Prior art keywords
- seismic
- wave velocity
- target
- neural network
- data
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 75
- 238000012549 training Methods 0.000 claims abstract description 159
- 238000013528 artificial neural network Methods 0.000 claims abstract description 112
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 48
- 230000015572 biosynthetic process Effects 0.000 claims abstract description 39
- 125000004122 cyclic group Chemical group 0.000 claims abstract description 29
- 230000006870 function Effects 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 16
- 230000015654 memory Effects 0.000 claims description 13
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 6
- 230000002159 abnormal effect Effects 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 238000010586 diagram Methods 0.000 description 13
- 238000004422 calculation algorithm Methods 0.000 description 10
- 230000008569 process Effects 0.000 description 10
- 238000012360 testing method Methods 0.000 description 10
- 230000000306 recurrent effect Effects 0.000 description 9
- 238000009795 derivation Methods 0.000 description 6
- 230000005540 biological transmission Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 5
- 239000011435 rock Substances 0.000 description 5
- 238000004590 computer program Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 230000007787 long-term memory Effects 0.000 description 4
- 230000006403 short-term memory Effects 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 241000522649 Zornia Species 0.000 description 2
- 230000002742 anti-folding effect Effects 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000001364 causal effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000007795 chemical reaction product Substances 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000006386 memory function Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
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
- G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
- G01V1/282—Application of seismic models, synthetic seismograms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- 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
- G01V2210/66—Subsurface modeling
- G01V2210/665—Subsurface modeling using geostatistical modeling
-
- 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
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Remote Sensing (AREA)
- Acoustics & Sound (AREA)
- Environmental & Geological Engineering (AREA)
- Geology (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Geophysics (AREA)
- Geophysics And Detection Of Objects (AREA)
Abstract
The application provides a method, a device and equipment for predicting transverse wave velocity, wherein the method comprises the following steps: acquiring data to be predicted, wherein the data to be predicted comprises: seismic records, seismic wavelets, longitudinal wave velocities, and formation densities; inputting the data to be predicted into a target transverse wave velocity prediction model to obtain a predicted value of the transverse wave velocity; the target transverse wave prediction model is obtained by unsupervised pre-training of a target neural network structure constructed by utilizing a convolutional neural network structure and a gated cyclic neural network structure. In the embodiment of the application, the seismic records with the heaviest proportion in seismic exploration are considered, so that the data used for prediction have diversity and representativeness, the constructed target neural network structure can be trained better according to the characteristics of the data related in the transverse wave velocity prediction, and the reliability and the accuracy of the trained target transverse wave velocity prediction model are effectively improved.
Description
Technical Field
The application relates to the technical field of seismic exploration, in particular to a method, a device and equipment for predicting shear wave velocity.
Background
Longitudinal and transverse wave velocities are key parameters in seismic exploration and play an important role in underground structure explanation, reservoir description and other works. Longitudinal wave velocity can be obtained through conventional acoustic logging, while transverse wave logging (such as dipole acoustic logging) is high in cost, and generally, actual seismic data often lack transverse wave velocity information.
The shear wave velocity prediction method in the prior art is a shear wave velocity prediction method based on a GRU (Gate recovery Unit) neural network, and the shear wave velocity is mainly predicted by using reservoir parameters such as longitudinal wave velocity, density and natural gamma. However, in the prior art, the transverse wave velocity prediction method based on the GRU neural network only utilizes logging data of reservoir parameters including longitudinal wave velocity, density, natural gamma and the like, the used data is relatively single, the applicability to a complex working scene of seismic exploration is not strong, and the accuracy of a prediction result is influenced. In addition, the transverse wave velocity prediction method based on the GRU neural network in the prior art cannot obtain accurate reservoir parameter values at a non-logging position, so that the method has certain limitation.
In addition, the transverse wave velocity prediction method based on the GRU neural network belongs to a supervised deep learning method, a training set needs to be established before prediction, a very accurate training set is usually difficult to establish in practical application, and the prediction accuracy is limited by the quantity and accuracy of the training set. But it is often difficult to establish a very accurate training set. Therefore, the technical scheme in the prior art cannot comprehensively, efficiently and accurately predict the transverse wave speed.
In view of the above problems, no effective solution has been proposed.
Disclosure of Invention
The embodiment of the application provides a shear wave velocity prediction method, a shear wave velocity prediction device and shear wave velocity prediction equipment, and aims to solve the problem that the shear wave velocity cannot be comprehensively, efficiently and accurately predicted in the prior art.
The embodiment of the application provides a method for predicting transverse wave velocity, which comprises the following steps: acquiring data to be predicted, wherein the data to be predicted comprises: seismic records, seismic wavelets, longitudinal wave velocities, and formation densities; inputting the seismic record, the seismic wavelets, the longitudinal wave velocity and the stratum density in the data to be predicted into a target transverse wave velocity prediction model to obtain a predicted value of the transverse wave velocity; the target transverse wave prediction model is obtained by unsupervised pre-training of a target neural network structure constructed by utilizing a convolutional neural network structure and a gated cyclic neural network structure.
In one embodiment, before acquiring the data to be predicted, the method further comprises: obtaining a plurality of groups of training sample data, wherein each group of training sample data comprises: seismic records, seismic wavelets, longitudinal wave velocities, and formation densities; constructing the target neural network structure by using the convolutional neural network structure and the gated cyclic neural network structure; carrying out unsupervised pre-training on the target neural network structure by utilizing the multiple groups of training sample data to obtain a target shear wave velocity prediction model, wherein the target shear wave velocity prediction model comprises the following steps: and the corresponding objective function between the shear wave velocity and the seismic record, the seismic wavelet, the longitudinal wave velocity and the stratum density.
In one embodiment, constructing the target neural network structure using at least two of a fully-connected neural network structure, a convolutional neural network structure, and a recurrent neural network structure comprises: and constructing the target neural network structure by utilizing the convolutional neural network structure and the gated cyclic neural network structure.
In one embodiment, the unsupervised pre-training of the target neural network structure using the plurality of sets of training sample data to obtain a target shear wave velocity prediction model includes: obtaining target training sample data in the multiple groups of training sample data; inputting a first seismic record and seismic wavelets in the target training sample data into an anti-folding layer of a convolutional neural network structure in the target shear wave velocity prediction model to obtain a first reflection coefficient; sequentially inputting the first reflection coefficient, the longitudinal wave velocity and the stratum density in the target training sample data into a convolution layer, a pooling layer and a gated cyclic neural network structure of the convolutional neural network structure to obtain a predicted value of the transverse wave velocity; determining whether preset training requirements are met or not according to the predicted value of the shear wave velocity, the longitudinal wave velocity, the stratum density, the seismic wavelets and the first seismic record; and under the condition that the preset training requirement is met, ending the unsupervised pre-training of the target neural network structure to obtain the target shear wave velocity prediction model.
In one embodiment, determining whether a preset training requirement is met according to the predicted value of the shear wave velocity, the longitudinal wave velocity, the stratigraphic density, the seismic wavelets and the first seismic record comprises: according to the predicted value of the transverse wave velocity, the longitudinal wave velocity and the formation density, a second reflection coefficient is obtained by utilizing a Zoeppritz equation; obtaining a second seismic record based on a seismic convolution model according to the second reflection coefficient and the seismic wavelets; determining a Euclidean norm between the first seismic record and the second seismic record according to the first seismic record and the second seismic record; and determining whether the maximum value of the Euclidean norm reaches a preset training requirement.
In one embodiment, after determining whether a preset training requirement is met according to the predicted value of the shear wave velocity, the longitudinal wave velocity, the stratigraphic density, the seismic wavelets and the first seismic record, the method further comprises: and under the condition that the preset training requirement is not met, reversely updating the parameter matrix and the paranoim vector layer by layer from an output layer in the target neural network structure according to the Euclidean norm between the first seismic record and the second seismic record.
In one embodiment, the data to be predicted input into the target shear wave velocity prediction model are seismic records of double wavelength, seismic wavelets, longitudinal wave velocity and formation density.
In one embodiment, obtaining multiple sets of training sample data comprises: obtaining a plurality of sets of training data from the geophysical model, wherein each set of training data comprises: compressional velocity, shear velocity, and formation density; according to the longitudinal wave velocity, the transverse wave velocity and the stratum density in each group of training data, calculating by utilizing a Zornia pritz equation and a seismic convolution model to obtain a seismic record corresponding to each group of training data; and respectively carrying out normalization processing on each group of training data and the corresponding seismic records to obtain a plurality of groups of training sample data comprising the seismic records, the seismic wavelets, the longitudinal wave velocity and the stratum density.
In one embodiment, obtaining multiple sets of training sample data comprises: acquiring well side channel seismic records and logging data; denoising the well side channel seismic record to obtain a denoised seismic record; processing abnormal values of the logging data to obtain target logging data; extracting target seismic wavelets through well seismic calibration according to the target logging data and the denoised seismic records; normalizing the target logging data, the denoised seismic records and the target seismic wavelets to obtain normalized logging data, seismic records and seismic wavelets; and generating training sample data comprising the seismic record, the seismic wavelet, the longitudinal wave velocity and the stratum density according to the normalized logging data, the seismic record and the seismic wavelet.
The embodiment of the present application further provides a device for predicting a shear wave velocity, including: the device comprises an acquisition module, a prediction module and a prediction module, wherein the acquisition module is used for acquiring data to be predicted, and the data to be predicted comprises: seismic records, seismic wavelets, longitudinal wave velocities, and formation densities; the prediction module is used for inputting the seismic record, the seismic wavelets, the longitudinal wave velocity and the stratum density in the data to be predicted into a target transverse wave velocity prediction model to obtain a predicted value of the transverse wave velocity; the target shear wave prediction model is obtained by unsupervised pre-training by using a target neural network structure constructed by a convolutional neural network structure and a gated cyclic neural network structure.
The embodiment of the application also provides a shear wave velocity prediction device, which comprises a processor and a memory for storing processor-executable instructions, wherein the processor executes the instructions to realize the steps of the shear wave velocity prediction method.
The embodiment of the application provides a shear wave velocity prediction method, which can be used for obtaining data to be predicted, wherein the data to be predicted comprises: seismic records, seismic wavelets, longitudinal wave velocities, and formation densities. The data to be predicted comprises seismic records and well logging data, and the data used for prediction has diversity and representativeness by considering the heaviest seismic records in seismic exploration, so that the accuracy of the transverse wave velocity prediction can be improved to a certain extent. Further, the seismic records, the seismic wavelets, the longitudinal wave velocity and the stratum density in the data to be predicted can be input into a target transverse wave velocity prediction model to obtain a predicted value of the transverse wave velocity; the target transverse wave prediction model is obtained by unsupervised pre-training of a target neural network structure constructed by utilizing a convolutional neural network structure and a gated cyclic neural network structure. Because different neural network structures have different characteristics and the types of data suitable for processing are different, the target neural network structure constructed by the convolutional neural network structure and the gated cyclic neural network structure can be utilized, so that the constructed target neural network structure can be trained better according to the characteristics of data involved in shear wave velocity prediction, and the reliability of the trained target shear wave velocity prediction model is effectively improved. In addition, under the condition that marked, accurate and comprehensive training data is not needed when the unsupervised pre-training mode is adopted for training, the prediction precision is not limited by the precision of the training data, and the applicability and the prediction efficiency of prediction by using a target transverse wave prediction model are effectively improved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application, are incorporated in and constitute a part of this application, and are not intended to limit the application. In the drawings:
FIG. 1 is a schematic diagram illustrating steps of a method for predicting shear wave velocity according to an embodiment of the present application;
FIG. 2 is a schematic diagram of seismic waves reflected and transmitted across an interface according to an embodiment of the present application;
FIG. 3 is a schematic diagram of a target neural network architecture provided in accordance with an embodiment of the present application;
FIG. 4 is a schematic diagram of training sample data generated according to an embodiment of the present application;
FIG. 5 is a parameter diagram of a Marmousi model provided in an embodiment of the present application;
FIG. 6 is a schematic diagram of seismic wavelets provided in accordance with an embodiment of the present application;
FIG. 7 is a schematic illustration of a convolution synthesized seismic record provided in accordance with an embodiment of the present application;
FIG. 8 is a graphical illustration of the relative error of the predicted results for 10 wells in model data prediction versus real data provided in accordance with an embodiment of the present application;
FIG. 9 is a graphical illustration of correlation coefficients of predicted results for 10 wells in model data prediction with real data, provided in accordance with an embodiment of the present application;
FIG. 10 is a graphical illustration of the relative error of the predicted results for 10 wells in actual data prediction versus actual data provided in accordance with an embodiment of the present application;
FIG. 11 is a graph illustrating correlation coefficients of 10 well predictions with real data in actual data prediction according to an embodiment of the present application;
FIG. 12 is a schematic diagram illustrating a comparison between a predicted value and an actual value of shear wave velocity provided according to an embodiment of the present application;
fig. 13 is a schematic diagram of absolute errors of a predicted value of shear wave velocity and an actual shear wave velocity provided according to an embodiment of the present application;
fig. 14 is a schematic structural diagram of a shear wave velocity prediction apparatus according to an embodiment of the present application;
fig. 15 is a schematic structural diagram of a shear wave velocity prediction apparatus provided according to an embodiment of the present application.
Detailed Description
The principles and spirit of the present application will be described with reference to a number of exemplary embodiments. It should be understood that these embodiments are given solely for the purpose of enabling those skilled in the art to better understand and to practice the present application, and are not intended to limit the scope of the present application in any way. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
As will be appreciated by one skilled in the art, embodiments of the present application may be embodied as a system, apparatus, device, method or computer program product. Accordingly, the present disclosure may be embodied in the form of: entirely hardware, entirely software (including firmware, resident software, micro-code, etc.), or a combination of hardware and software.
Although the flow described below includes operations that occur in a particular order, it should be appreciated that the processes may include more or less operations that are performed sequentially or in parallel (e.g., using parallel processors or a multi-threaded environment).
Referring to fig. 1, the present embodiment can provide a method for predicting a transverse wave velocity. The transverse wave velocity prediction method can be used for predicting the transverse wave velocity by using data such as seismic records, seismic wavelets, longitudinal wave velocity, stratum density and the like. The shear wave velocity prediction method may include the following steps.
S101: acquiring data to be predicted, wherein the data to be predicted comprises: seismic records, seismic wavelets, longitudinal wave velocities, and formation densities.
In this embodiment, data to be predicted may be obtained in advance, the data to be predicted may be data for predicting a shear wave velocity, and the data to be predicted may include: seismic records, seismic wavelets, longitudinal wave velocities and formation densities at the area to be predicted.
In this embodiment, the seismic records, seismic wavelets, longitudinal wave velocities, and formation densities described above may be determined from well side channel seismic records and log data. The method for acquiring the data to be predicted comprises the following steps: the method includes the steps of receiving data input into a target transverse wave prediction model by a user, or searching from a file according to a preset path, wherein the data can be determined according to actual conditions, and the method is not limited in the application.
S102: inputting the seismic record, the seismic wavelets, the longitudinal wave velocity and the stratum density in the data to be predicted into a target transverse wave velocity prediction model to obtain a predicted value of the transverse wave velocity; the target transverse wave prediction model is obtained by unsupervised pre-training of a target neural network structure constructed by utilizing a convolutional neural network structure and a gated cyclic neural network structure.
In this embodiment, the acquired data to be predicted may be input into a pre-trained target shear wave prediction model, so as to obtain a predicted value of the shear wave velocity. The target shear wave prediction model can be obtained by unsupervised pre-training of a target neural network structure constructed by a convolutional neural network structure and a gated cyclic neural network structure.
In this embodiment, the target neural network structure may be pre-trained in an unsupervised manner. The difference between unsupervised pre-training and supervised pre-training is that: samples in the pre-training stage do not need to be manually labeled with data, data used for training does not contain output targets, and unsupervised pre-training can automatically learn rules or some valuable information in the training data through a learning algorithm. Under the condition that not too many well-labeled, accurate and comprehensive training data exist, a better training result can be obtained by adopting unsupervised pre-training, the precision of the training result cannot be reduced, and the applicability of the training result is effectively improved.
In this embodiment, the neural network structure is generally divided into three types, namely, a fully connected neural network structure (FC), a convolutional neural network structure (CNN), and a recurrent neural network structure (RNN), and different neural network structures have different characteristics and different types of data suitable for processing. Wherein, the fully-connected neural network structure is a most basic neural network structure; the convolutional neural network structure comprises a convolutional layer and a pooling layer, has the characteristics of local sensing and weight sharing, and is suitable for image processing and recognition tasks; the recurrent neural network structure comprises a signal feedback structure, has dynamic characteristics and memory function, and is suitable for processing time sequence series.
Therefore, the target neural network structure constructed by at least two neural network structures can be selected from the fully-connected neural network structure, the convolutional neural network structure and the cyclic neural network structure according to the characteristics of data related to the transverse wave velocity prediction, so that the constructed target neural network structure can better process the characteristics of the data related to the transverse wave velocity prediction, and the reliability of a target transverse wave velocity prediction model obtained based on the training of the target neural network structure is effectively improved. In one embodiment, since shear wave velocity is continuous in time, a target neural network structure may be constructed using a convolutional neural network structure and a gated circular neural network structure.
In one embodiment, in order to reliably establish the target shear wave velocity prediction model, the following steps may be further included before acquiring the data to be predicted.
S01: acquiring a plurality of groups of training sample data, wherein each group of training sample data comprises: seismic records, seismic wavelets, longitudinal wave velocities, and formation densities.
In this embodiment, a plurality of sets of training sample data may be obtained in advance, and each set of training sample data may at least include: seismic records, seismic wavelets, longitudinal wave velocities, and formation densities. The number of training samples can be 600000 or 590010, but it can be any other possible value, and the specific value can be determined according to the actual situation.
In this embodiment, the manner of acquiring multiple sets of training sample data may include: training sample data input by a user is received, or the training sample data can be obtained by querying according to a preset path. It is understood that, the training sample data may also be obtained in other possible manners, for example, the training sample data is searched in the web page according to a certain search condition, which may be determined specifically according to an actual situation, and the present application does not limit this.
S02: and constructing a target neural network structure by using the convolutional neural network structure and the gated cyclic neural network structure.
In the present embodiment, since data involved in the shear wave velocity prediction is continuous in time, a target neural network structure can be constructed using a recurrent neural network structure. And because the characteristic of weight sharing of the convolutional neural network can also improve learning precision, the convolutional neural network structure can also be adopted when the target neural network structure is constructed, and in a preferred scheme, the convolutional neural network structure and the cyclic neural network structure can be utilized to construct the target neural network structure.
The conventional recurrent neural network can effectively analyze and process a short time sequence array, but cannot analyze and process a time sequence array with an overlong dimension, otherwise, the phenomenon of gradient disappearance or gradient explosion can be generated. An improved structure of the recurrent neural network is a long and short term memory neural network (LSTM) which can realize the addition or deletion of information through the structure of a gate (a forgetting gate, an input gate and an output gate), thereby improving the problem that the conventional recurrent neural network can not process longer time sequence series. However, the repetitive structure of the long-term and short-term memory neural network is too complex, and a large amount of time is required for sample training, so that the long-term and short-term memory neural network has many variants. The gated recurrent neural network (GRU) can replace a forgetting gate, an input gate and an output gate in the long-time memory neural network by using a reset gate and an update gate.
Although the gated cyclic neural network has the same prediction precision as the long-term and short-term memory neural network, the gated cyclic neural network is easier to train and has higher efficiency. In a preferred embodiment, the target neural network structure can be constructed by using a gated recurrent neural network.
The convolutional neural network structure comprises a convolutional layer and a pooling layer, the convolutional layer in the convolutional neural network can be replaced by a deconvolution layer, the physical significance of seismic record and seismic wavelet deconvolution is given to the convolutional layer, training of the calculation process can be avoided, and meanwhile training efficiency and training precision are improved. In addition, the characteristic of weight sharing of the convolutional neural network can also avoid repeated training and error training of the same characteristic in the training process, and the training efficiency and the training precision are improved. Based on the method, the target neural network structure constructed by the convolutional neural network structure and the gated cyclic neural network structure can have higher training precision and training efficiency.
S03: carrying out unsupervised pre-training on a target neural network structure by utilizing a plurality of groups of training sample data to obtain a target shear wave velocity prediction model, wherein the target shear wave velocity prediction model comprises the following steps: and the corresponding objective function between the transverse wave velocity and the seismic record, the seismic wavelet, the longitudinal wave velocity and the stratum density.
In the present embodiment, a plurality of factors that affect the shear wave velocity can be determined in advance from the convolution theory, the zernillitz equation, and the like. In one embodiment, if the effect of the seismic wavelets on the seismic frequency band is considered and the noise is assumed to be zero, the seismic convolution model may be expressed as:
d＝W*R
wherein d is a seismic record; w is seismic wavelet; r is a reflection coefficient.
As the seismic waves propagate in the formation, they are transmitted and refracted by the formation interfaces, as shown in FIG. 2. For isotropic media, the physical phenomena in fig. 2 can be described by the Zoeppritz (Zoeppritz) equation, which is expressed as:
wherein, theta 1 、θ 2 、θ 3 And theta 4 Respectively representing a reflection angle of a reflected longitudinal wave, a reflection angle of a transmitted longitudinal wave, a transmission angle of a reflected transverse wave and a transmission angle of a transmitted transverse wave; v P1 、V S1 And ρ 1 Respectively representing the longitudinal wave velocity, the transverse wave velocity and the density of an upper layer of the interface; v P2 、V S2 And ρ 2 Respectively representing the longitudinal wave velocity, the transverse wave velocity and the density of the lower layer of the interface; r is pp 、R ps 、T pp And T ps The longitudinal wave reflection coefficient, converted transverse wave reflection coefficient, longitudinal wave transmission coefficient and converted transverse wave transmission coefficient are respectively expressed.
From the Zoeppritz equation, it can be seen that the reflection coefficient can be represented by longitudinal wave velocity, transverse wave velocity, density, reflection angle, and transmission angle. On the premise of only considering the incidence of the longitudinal wave, according to Snell's law, the reflection coefficient of the longitudinal wave can be expressed as:
wherein, V P Is the velocity of the longitudinal wave; v S Is the transverse wave velocity; rho is the formation density; r PP Is the longitudinal wave reflection coefficient; f (theta) is R PP To [ V ] P ,V S ,ρ]' is related to the incident angle theta of the longitudinal wave.
Thus, when considering only the incidence of longitudinal waves, a seismic record can be expressed as:
wherein d is a seismic record; r PP Is the longitudinal wave reflection coefficient; w is seismic wavelet; v P Is the velocity of the longitudinal wave; v S Is the transverse wave velocity; rho is the formation density; f (theta) is R PP To [ V ] P ,V S ,ρ]' of the mapping matrix. Therefore, the shear wave velocity can be expressed by using seismic records, seismic wavelets, longitudinal wave velocity and density, and the expression is as follows:
V S ＝g(d,W,V P ,ρ)
wherein g is an objective function which relates the shear wave velocity with the seismic record, the seismic wavelet, the longitudinal wave velocity and the stratum density. During the process of unsupervised pre-training of the target neural network structure, the relation between the transverse wave velocity and the seismic record, the seismic wavelet, the longitudinal wave velocity and the stratum density, namely the target function, can be obtained through autonomous learning fitting. Therefore, the trained target shear wave velocity prediction model may include: and the corresponding objective function between the transverse wave velocity and the seismic record, the seismic wavelet, the longitudinal wave velocity and the stratum density.
In an embodiment, the constructed target neural network structure may be as shown in fig. 3, and taking target training sample data in a plurality of sets of training sample data as an example, the target neural network structure in fig. 3 may be subjected to unsupervised pre-training according to the following steps.
S31: and inputting the first seismic record and the seismic wavelets in the target training sample data into an anti-folding layer of a convolutional neural network structure in a target transverse wave velocity prediction model to obtain a first reflection coefficient.
In this embodiment, the first seismic record may be an original seismic record in the target training sample data. In some embodiments, the first seismic record and seismic wavelet input into the target shear velocity prediction model may be double-wavelength data, e.g., 100ms for seismic wavelets in the target training sample data, and 200ms for each input of seismic wavelet data. Correspondingly, the longitudinal wave velocity and the formation density input into the target shear wave velocity prediction model can also be data with twice the wavelength.
S32: and sequentially inputting the first reflection coefficient, the longitudinal wave velocity and the stratum density in the target training sample data into a pleated layer, a pooling layer and a gated cyclic neural network structure of the convolutional neural network structure to obtain a predicted value of the transverse wave velocity.
S33: and determining whether the preset training requirement is met or not according to the predicted value of the shear wave velocity, the longitudinal wave velocity, the stratum density, the seismic wavelet and the first seismic record.
In the unsupervised pre-training process, when a predicted value of the transverse wave velocity is obtained, the predicted value needs to be verified first to determine whether the predicted precision meets the requirement. The preset training requirement may be used to represent a desired training precision, and may be specifically determined according to an actual situation, which is not limited in the present application.
In the present embodiment, the second reflection coefficient may be obtained by using the zoepez equation based on the predicted value of the shear velocity, the longitudinal velocity input in step S32, and the formation density. And obtaining a second seismic record based on the seismic convolution model according to the second reflection coefficient and the seismic wavelet in the step S31. The second reflection coefficient and the second reflection coefficient may be theoretical values calculated according to a theoretical formula.
Further, the euclidean norm between the first seismic record and the second seismic record may be determined based on the first seismic record and the second seismic record. In one embodiment, it may be determined whether the maximum value of the euclidean norm meets a preset training requirement. The preset training requirement here may be less than or equal to a preset value, and the preset value may be a positive number, for example: 10 -5 It is understood that other possible values may be used, and the specific values may be set according to practical situations, which is not limited in the present application.
In this embodiment, the Norm (Norm) is a function that gives each vector in a certain vector space (or matrix) a length or magnitude. The euclidean norm, i.e., the L2 norm, may be the distance between two points in space, and since the L2 norm has a smoother characteristic relative to the L1 norm in machine learning, in model prediction, it tends to have a better prediction characteristic than the L1 norm. When encountering two features that are helpful to prediction, the L1-norm tends to select a larger one, while the L2-norm tends to combine the two. Thus, in some embodiments, the L2 norm may be applied to a loss function to prevent the problem of over-model fitting.
S34: and under the condition that the preset training requirement is met, ending the unsupervised pre-training of the target neural network structure to obtain a target shear wave velocity prediction model.
In this embodiment, a plurality of sets of test data may be obtained, where each set of test data includes: seismic records, seismic wavelets, longitudinal wave velocities, and formation densities. The pre-trained shear wave velocity prediction model can be tested by using a plurality of groups of test data, the test steps are the same as those of the steps S31-S33, and repeated parts are not described again. The ratio of the number of training sample data to the number of test data may be 4.
In this embodiment, when it is determined that the preset training requirement is not met, the parameter matrix and the bias vector are reversely updated layer by layer in the target neural network structure from the output layer according to the euclidean norm between the first seismic record and the second seismic record, so as to obtain the target neural network structure with reversely updated weights, and then the steps S31 to S34 are repeated to continue training the target neural network structure with reversely updated weights until the preset training model is met.
In this embodiment, the process of updating the weight in the reverse direction is based on a back propagation algorithm, that is, the error is transmitted from the output layer to the front layer by layer, so as to update the parameter matrix and the bias vector layer by layer, and the core of the back propagation algorithm is a chain derivation rule. It should be noted that the derivation of the back-propagation algorithm is based on activation functions, error calculation methods, network connection structure and optimization algorithms. If the activation functions are different, the error calculation modes are different, the network connection structures are different, and the optimization algorithms are different, the specific back propagation algorithms obtained through derivation are different. However, the derivation method is the same, and the derivation can be performed by applying the chain derivation rule. In the present embodiment, the back propagation algorithm is derived from an activation function (sigmoid function and relu function), a sum of squares error, a target neural network structure, and a stochastic gradient descent optimization algorithm.
In one embodiment, in order to make the obtained multiple sets of training sample data more representative and more widely applicable, the training sample data may be generated by using model data, and specifically, the multiple sets of training sample data may be obtained according to the following steps.
S11: obtaining a plurality of sets of training data from the geophysical model, wherein each set of training data comprises: compressional velocity, shear velocity, and formation density.
S12: calculating to obtain the seismic records corresponding to each group of training data by utilizing a Zornia pritz equation and a seismic convolution model according to the longitudinal wave velocity, the transverse wave velocity and the stratum density in each group of training data;
s13: and respectively carrying out normalization processing on each group of training data and the corresponding seismic records to obtain a plurality of groups of training sample data comprising the seismic records, the seismic wavelets, the longitudinal wave velocity and the stratum density.
In this embodiment, the selected geophysical model is a Marmousi model. The Marmousi model is a two-dimensional acoustic wave model, and the model may include: two-dimensional compressional velocity, shear velocity, and formation density. The Marmousi original model only supports propagation of compression (P) waves, and the Marmousi2 model supports not only compression waves but also transverse waves and conversion waves. The specific model can be determined according to actual conditions, and is not limited in the present application.
In this embodiment, the reflection coefficient may be calculated by using a Zoeppritz equation according to the longitudinal wave velocity, the transverse wave velocity, and the formation density in each set of training data, and then the reflection coefficient and the seismic wavelets are convolved into a seismic record by using a seismic convolution model, so as to obtain seismic data corresponding to each set of training data.
The sets of training data and corresponding seismic records may be normalized separately according to the following formula:
wherein x is min And x max Are respectively x i Minimum and maximum values of.
In an embodiment, training sample data may also be generated by using actual data, and specifically, the multiple sets of training sample data may be obtained according to the following steps.
S14: and acquiring well side channel seismic records and logging data.
S15: and denoising the well side channel seismic record to obtain the denoised seismic record.
S16: and processing abnormal values of the logging data to obtain target logging data.
S17: and obtaining target seismic wavelets through well seismic calibration extraction according to the target logging data and the denoised seismic records.
S18: and normalizing the target logging data, the denoised seismic records and the target seismic wavelets to obtain normalized logging data, seismic records and seismic wavelets.
S19: and generating training sample data comprising the seismic record, the seismic wavelet, the longitudinal wave velocity and the stratum density according to the normalized logging data, the seismic record and the seismic wavelet.
In this embodiment, the well-seismic calibration is a bridge connecting an earthquake and a geology, the well-seismic calibration is a foundation and a premise for performing horizon interpretation and reservoir description, and structural and reservoir information included in seismic data can be better researched after the well-seismic calibration is performed. The logging data may include a longitudinal wave velocity and a formation density, and the generated training sample data may be as shown in fig. 4, where an abscissa in fig. 4 is a corresponding numerical value, and an ordinate is a depth.
In one embodiment, the above manners in S11-S13 and S14-S19 may be adopted to generate multiple sets of training sample data, so that the generated training sample data is rich and diverse and has training value.
In one embodiment, the effectiveness and practicability of the trained target shear wave prediction model can be verified by testing the model data and the actual data. Firstly, the Marmousi model can be used for predicting the shear wave velocity, the test of the Marmousi model is only used for verifying the validity, and other similar models can obtain the same result. The parameters of the Marmousi model can be as shown in fig. 5, including compressional velocity, shear velocity and formation density, with 13601 traces in the lateral direction, each trace comprising 2801 points.
Further, a reflection coefficient can be calculated by using a Zoeppritz equation according to the Marmousi model parameters in FIG. 5, and the reflection coefficient is convoluted with a Ricker seismic wavelet (shown in FIG. 6) with a main frequency of 60Hz and a wavelength of 100ms to form a seismic record (shown in FIG. 7). Inputting seismic records with 2-time wavelength, seismic wavelets, longitudinal wave velocity and stratum density every time, respectively utilizing a traditional transverse wave velocity prediction method, a transverse wave velocity prediction method based on an independent CNN, a transverse wave velocity prediction method based on an independent GRU and a target transverse wave velocity prediction model based on an unsupervised GRU-CNN to predict the transverse wave velocity, and utilizing correlation coefficients and relative errors to represent the accuracy of the transverse wave velocity prediction.
The conventional shear wave velocity prediction method may include: empirical formula methods and petrophysical modeling methods. The empirical formula method is used for obtaining a longitudinal wave velocity relation formula and a transverse wave velocity relation formula through statistical analysis of longitudinal wave velocity and transverse wave velocity, and the transverse wave velocity is predicted according to the longitudinal wave velocity. Different rocks have different longitudinal and transverse wave velocity fitting formulas. The empirical formula method is easy to implement and high in calculation efficiency, but the fitting relational expression reflects the statistical rule of a large amount of logging data, and large errors exist in practical application. The rock physical modeling method is mainly used for accurately calculating the elastic parameters of the rock by building a rock physical model and then calculating the shear wave velocity based on the relation between the elastic parameters and the shear wave velocity. The accuracy of predicting the shear wave velocity by using the rock physics modeling method is high, but the algorithm is complex, the needed parameters are more, and the calculation efficiency is low.
Further, 10 samples can be extracted for transverse wave velocity prediction, and the relative error between the prediction result of 10 wells in model data prediction and real data is shown in table 1. To analyze the prediction accuracy more clearly, the data of table 1 may be plotted as a line graph as shown in fig. 8. As can be seen from FIG. 8, the relative error predicted by the unsupervised GRU-CNN-based shear wave velocity prediction method is about 2.609%, which is less than that predicted by the conventional shear wave velocity prediction method (4.356%); the method for predicting shear velocity based on GRU alone (3.097%) and the method for predicting shear velocity based on CNN alone (3.207%).
TABLE 1
TABLE 2
Table 2 shows the correlation coefficient between the predicted result and the actual data of 10 wells in the model data test, and the corresponding line graph is shown in fig. 9. As can be seen from fig. 9, the average value of the correlation coefficient between the observed transverse wave velocity and the actual value predicted by the transverse wave velocity prediction method based on unsupervised GRU-CNN is 0.9828, which is greater than the conventional transverse wave velocity prediction method (0.9409), the transverse wave velocity prediction method based on the GRU alone (0.9691), and the transverse wave velocity prediction method based on the CNN alone (0.9697). Therefore, the test result based on the Marmousi model verifies the effectiveness and feasibility of prediction by using the target shear wave velocity prediction model.
Further, the transverse wave velocity of the D-zone actual data can be utilized to verify the effectiveness and the practicability. And (3) performing transverse wave velocity prediction by using well side channel seismic data and logging data of 30 wells in the D area, wherein each well comprises 3000 data points and 90000 data points in total. Firstly, noise suppression processing is carried out on the seismic data of the well side channel, and abnormal value processing is carried out on the logging data (longitudinal wave velocity, transverse wave velocity and stratum density). And then carrying out well seismic calibration and extracting seismic wavelets. And then, carrying out normalization processing on the data, and extracting 10 logs to carry out transverse wave velocity prediction. The relative error between the predicted result of 10 wells and the actual data in the actual data prediction is shown in table 3, and in order to analyze the prediction accuracy more clearly, the data in table 3 may be plotted as a line graph shown in fig. 10. As can be seen from FIG. 10, the predicted relative error of the shear wave velocity prediction method based on unsupervised GRU-CNN is the smallest, and the average is 2.854%.
TABLE 3
Table 4 shows the correlation coefficient between the predicted result of 10 wells in the actual data prediction and the real data, and the corresponding plotted line graph is shown in fig. 11. As can be seen from fig. 11, the transverse wave velocity prediction method based on unsupervised GRU-CNN predicts the largest correlation coefficient between the transverse wave velocity and the actual value, and the average value is 0.9802. FIG. 12 is a comparison graph of predicted and true values of shear velocity predicted by four wells through an unsupervised GRU-CNN based shear velocity prediction method. As can be seen from fig. 12, the predicted value of the shear wave velocity and the actual value of the shear wave velocity have the same curve trend, and have a good matching degree.
TABLE 4
Fig. 13 is a diagram of the absolute error between the predicted value of the shear velocity and the actual shear velocity, and it can be seen from fig. 13 that the absolute error of the training well is less than 110m/s, and the absolute error of the testing well is less than 150m/s, which indicates that the prediction using the target shear velocity prediction model has better practicability.
From the above description, it can be seen that the embodiments of the present application achieve the following technical effects: the data to be predicted can be obtained, wherein the data to be predicted comprises: seismic records, seismic wavelets, longitudinal wave velocities, and formation densities. The data to be predicted comprises seismic records and well logging data, and the data used for prediction has diversity and representativeness by considering the heaviest seismic records in seismic exploration, so that the accuracy of the transverse wave velocity prediction can be improved to a certain extent. Further, the seismic records, the seismic wavelets, the longitudinal wave velocity and the stratum density in the data to be predicted can be input into a target transverse wave velocity prediction model to obtain a predicted value of the transverse wave velocity; the target transverse wave prediction model is obtained by unsupervised pre-training by using a target neural network structure constructed by a convolutional neural network structure and a gated cyclic neural network structure. Because different neural network structures have different characteristics and the types of data suitable for processing are different, the target neural network structure constructed by the convolutional neural network structure and the gated cyclic neural network structure can be utilized, so that the constructed target neural network structure can be trained better according to the characteristics of data involved in shear wave velocity prediction, and the reliability of the trained target shear wave velocity prediction model is effectively improved. In addition, the prediction precision is not limited by the precision of training data under the condition that no marked, accurate and comprehensive training data is needed when the unsupervised pre-training mode is adopted for training, and the applicability and the prediction efficiency of predicting by using the target transverse wave prediction model are effectively improved.
Based on the same inventive concept, the embodiment of the present application further provides a shear wave velocity prediction apparatus, as in the following embodiments. Because the principle of solving the problem of the shear wave velocity prediction device is similar to that of the shear wave velocity prediction method, the implementation of the shear wave velocity prediction device can refer to the implementation of the shear wave velocity prediction method, and repeated parts are not described again. As used hereinafter, the term "unit" or "module" may be a combination of software and/or hardware that implements a predetermined function. Although the means described in the embodiments below are preferably implemented in software, an implementation in hardware, or a combination of software and hardware is also possible and contemplated. Fig. 14 is a block diagram showing a configuration of a shear wave velocity prediction apparatus according to an embodiment of the present invention, and as shown in fig. 14, the apparatus may include: the configuration of the acquisition module 141 and the prediction module 142 will be described below.
The obtaining module 141 may be configured to obtain data to be predicted, where the data to be predicted includes: seismic records, seismic wavelets, longitudinal wave velocities, and formation densities.
The prediction module 142 may be configured to input the seismic record, the seismic wavelets, the longitudinal wave velocity, and the formation density in the data to be predicted into the target shear wave velocity prediction model to obtain a predicted value of the shear wave velocity; the target transverse wave prediction model is obtained by unsupervised pre-training of a target neural network structure constructed by utilizing a convolutional neural network structure and a gated cyclic neural network structure.
The embodiment of the present application further provides an electronic device, which may specifically refer to a schematic structural diagram of the electronic device based on the method for predicting the shear wave velocity provided in the embodiment of the present application shown in fig. 15, and the electronic device may specifically include an input device 51, a processor 52, and a memory 53. The input device 51 may be specifically configured to input data to be predicted. The processor 52 may be specifically configured to obtain data to be predicted, where the data to be predicted includes: seismic records, seismic wavelets, longitudinal wave velocities, and formation densities; inputting the seismic record, the seismic wavelets, the longitudinal wave velocity and the stratum density in the data to be predicted into a target transverse wave velocity prediction model to obtain a predicted value of the transverse wave velocity; the target transverse wave prediction model is obtained by unsupervised pre-training of a target neural network structure constructed by utilizing a convolutional neural network structure and a gated cyclic neural network structure. The memory 53 may be used to store parameters such as seismic records, seismic wavelets, longitudinal wave velocities, and formation densities.
In this embodiment, the input device may be one of the main apparatuses for information exchange between a user and a computer system. The input devices may include a keyboard, mouse, camera, scanner, light pen, handwriting input panel, voice input device, etc.; the input device is used to input raw data and a program for processing these numbers into the computer. The input device can also acquire and receive data transmitted by other modules, units and devices. The processor may be implemented in any suitable way. For example, the processor may take the form of, for example, a microprocessor or processor and a computer-readable medium that stores computer-readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, an embedded microcontroller, and so forth. The memory may in particular be a memory device used in modern information technology for storing information. The memory may include multiple levels, and in a digital system, it may be memory as long as it can hold binary data; in an integrated circuit, a circuit without a physical form and with a storage function is also called a memory, such as a RAM, a FIFO and the like; in the system, the storage device in physical form is also called a memory, such as a memory bank, a TF card and the like.
In this embodiment, the functions and effects specifically realized by the electronic device can be explained by comparing with other embodiments, and are not described herein again.
The present application further provides a computer storage medium based on a shear wave velocity prediction method, where the computer storage medium stores computer program instructions, and when the computer program instructions are executed, the computer storage medium can implement: acquiring data to be predicted, wherein the data to be predicted comprises: seismic records, seismic wavelets, longitudinal wave velocities, and formation densities; inputting the seismic record, the seismic wavelets, the longitudinal wave velocity and the stratum density in the data to be predicted into a target transverse wave velocity prediction model to obtain a predicted value of the transverse wave velocity; the target transverse wave prediction model is obtained by unsupervised pre-training of a target neural network structure constructed by utilizing a convolutional neural network structure and a gated cyclic neural network structure.
In this embodiment, the storage medium includes, but is not limited to, a Random Access Memory (RAM), a Read-Only Memory (ROM), a Cache (Cache), a Hard Disk Drive (HDD), or a Memory Card (Memory Card). The memory may be used to store computer program instructions. The network communication unit may be an interface for performing network connection communication, which is set in accordance with a standard prescribed by a communication protocol.
In this embodiment, the functions and effects specifically realized by the program instructions stored in the computer storage medium can be explained by comparing with other embodiments, and are not described herein again.
It will be apparent to those skilled in the art that the modules or steps of the embodiments of the present application described above may be implemented by a general purpose computing device, they may be centralized on a single computing device or distributed across a network of multiple computing devices, and alternatively, they may be implemented by program code executable by a computing device, such that they may be stored in a storage device and executed by a computing device, and in some cases, the steps shown or described may be performed in an order different from that described herein, or they may be separately fabricated into individual integrated circuit modules, or multiple ones of them may be fabricated into a single integrated circuit module. Thus, embodiments of the present application are not limited to any specific combination of hardware and software.
Although the present application provides method steps as in the above-described embodiments or flowcharts, additional or fewer steps may be included in the method, based on conventional or non-inventive efforts. In the case of steps where no necessary causal relationship exists logically, the order of execution of the steps is not limited to that provided by the embodiments of the present application. When implemented in an actual device or end product, the methods of (1) can be performed sequentially or in parallel according to the embodiments or methods shown in the figures (e.g., in the context of parallel processors or multi-threaded processing).
It is to be understood that the above description is intended to be illustrative, and not restrictive. Many embodiments and many applications other than the examples provided will be apparent to those of skill in the art upon reading the above description. The scope of the application should, therefore, be determined not with reference to the above description, but instead should be determined with reference to the pending claims along with the full scope of equivalents to which such claims are entitled.
The above description is only a preferred embodiment of the present application and is not intended to limit the present application, and it will be apparent to those skilled in the art that various modifications and variations can be made in the embodiment of the present application. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims (9)
1. A method for predicting shear wave velocity, comprising:
acquiring data to be predicted, wherein the data to be predicted comprises: seismic records, seismic wavelets, longitudinal wave velocities, and formation densities;
inputting the seismic record, the seismic wavelets, the longitudinal wave velocity and the stratum density in the data to be predicted into a target transverse wave velocity prediction model to obtain a predicted value of the transverse wave velocity; the target shear wave velocity prediction model is obtained by unsupervised pre-training of a target neural network structure constructed by utilizing a convolutional neural network structure and a gated cyclic neural network structure; the target shear wave velocity prediction model is obtained by unsupervised pre-training by utilizing a target neural network structure constructed by a convolutional neural network structure and a gated cyclic neural network structure according to the following modes: acquiring target training sample data in a plurality of groups of training sample data; inputting a first seismic record and seismic wavelets in the target training sample data into a convolution layer of a convolution neural network structure in a target shear wave velocity prediction model to obtain a first reflection coefficient; sequentially inputting the first reflection coefficient, the longitudinal wave velocity and the stratum density in the target training sample data into a convolution layer, a pooling layer and a gated cyclic neural network structure of the convolutional neural network structure to obtain a predicted value of the transverse wave velocity; determining whether preset training requirements are met or not according to the predicted value of the transverse wave velocity, the longitudinal wave velocity, the stratum density, the seismic wavelets and the first seismic record; and under the condition that the preset training requirement is met, ending the unsupervised pre-training of the target neural network structure to obtain the target shear wave velocity prediction model.
2. The method of claim 1, further comprising, prior to obtaining data to be predicted:
acquiring a plurality of groups of training sample data, wherein each group of training sample data comprises: seismic records, seismic wavelets, longitudinal wave velocities, and formation densities;
constructing the target neural network structure by using the convolutional neural network structure and the gated cyclic neural network structure;
carrying out unsupervised pre-training on the target neural network structure by using the multiple groups of training sample data to obtain a target shear wave velocity prediction model, wherein the target shear wave velocity prediction model comprises: and the corresponding objective function between the transverse wave velocity and the seismic record, the seismic wavelet, the longitudinal wave velocity and the stratum density.
3. The method of claim 1, wherein determining whether a predetermined training requirement is met based on the predicted value of shear velocity, the compressional velocity, the formation density, the seismic wavelet, and the first seismic record comprises:
obtaining a second reflection coefficient by utilizing a Zoepper equation according to the predicted value of the transverse wave velocity, the longitudinal wave velocity and the formation density;
obtaining a second seismic record based on a seismic convolution model according to the second reflection coefficient and the seismic wavelets;
determining a Euclidean norm between the first seismic record and the second seismic record according to the first seismic record and the second seismic record;
and determining whether the maximum value of the Euclidean norm reaches a preset training requirement.
4. The method of claim 3, wherein after determining whether a predetermined training requirement is met based on the predicted value of shear velocity, the compressional velocity, the formation density, the seismic wavelets, and the first seismic record, further comprising:
and under the condition that the preset training requirement is not met, reversely updating the parameter matrix and the paranoim vector layer by layer from an output layer in the target neural network structure according to the Euclidean norm between the first seismic record and the second seismic record.
5. The method of claim 1, wherein the data to be predicted input into the target shear wave velocity prediction model are seismic records of twice the wavelength, seismic wavelets, compressional velocity and formation density.
6. The method of claim 1, wherein obtaining multiple sets of training sample data comprises:
obtaining a plurality of sets of training data from the geophysical model, wherein each set of training data comprises: compressional velocity, shear velocity, and formation density;
according to the longitudinal wave velocity, the transverse wave velocity and the stratum density in each group of training data, calculating by utilizing a Zoeppritz equation and a seismic convolution model to obtain seismic records corresponding to each group of training data;
and respectively carrying out normalization processing on each group of training data and the corresponding seismic records to obtain a plurality of groups of training sample data comprising the seismic records, the seismic wavelets, the longitudinal wave velocity and the stratum density.
7. The method of claim 1, wherein obtaining multiple sets of training sample data comprises:
acquiring a well side channel seismic record and logging data;
denoising the well side channel seismic record to obtain a denoised seismic record;
processing abnormal values of the logging data to obtain target logging data;
extracting target seismic wavelets through well seismic calibration according to the target logging data and the denoised seismic records;
normalizing the target logging data, the denoised seismic records and the target seismic wavelets to obtain normalized logging data, seismic records and seismic wavelets;
and generating training sample data comprising the seismic record, the seismic wavelet, the longitudinal wave velocity and the stratum density according to the normalized logging data, the seismic record and the seismic wavelet.
8. A shear wave velocity prediction apparatus comprising:
the device comprises an acquisition module, a prediction module and a prediction module, wherein the acquisition module is used for acquiring data to be predicted, and the data to be predicted comprises: seismic records, seismic wavelets, longitudinal wave velocities, and formation densities;
the prediction module is used for inputting the seismic record, the seismic wavelets, the longitudinal wave velocity and the stratum density in the data to be predicted into a target transverse wave velocity prediction model to obtain a predicted value of the transverse wave velocity; the target shear wave velocity prediction model is obtained by unsupervised pre-training of a target neural network structure constructed by utilizing a convolutional neural network structure and a gated cyclic neural network structure; the target shear wave velocity prediction model is obtained by unsupervised pre-training by utilizing a target neural network structure constructed by a convolutional neural network structure and a gated cyclic neural network structure according to the following modes: acquiring target training sample data in a plurality of groups of training sample data; inputting a first seismic record and seismic wavelets in the target training sample data into a convolution layer of a convolution neural network structure in a target shear wave velocity prediction model to obtain a first reflection coefficient; sequentially inputting the first reflection coefficient, the longitudinal wave velocity and the stratum density in the target training sample data into a convolution layer, a pooling layer and a gated cyclic neural network structure of the convolutional neural network structure to obtain a predicted value of the transverse wave velocity; determining whether preset training requirements are met or not according to the predicted value of the transverse wave velocity, the longitudinal wave velocity, the stratum density, the seismic wavelets and the first seismic record; and under the condition that the preset training requirement is met, ending the unsupervised pre-training of the target neural network structure to obtain the target shear wave velocity prediction model.
9. A shear wave velocity prediction device comprising a processor and a memory for storing processor-executable instructions which, when executed by the processor, implement the steps of the method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010321843.7A CN111487679B (en) | 2020-04-22 | 2020-04-22 | Transverse wave velocity prediction method, device and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010321843.7A CN111487679B (en) | 2020-04-22 | 2020-04-22 | Transverse wave velocity prediction method, device and equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111487679A CN111487679A (en) | 2020-08-04 |
| CN111487679B true CN111487679B (en) | 2023-04-07 |
Family
ID=71798696
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010321843.7A Active CN111487679B (en) | 2020-04-22 | 2020-04-22 | Transverse wave velocity prediction method, device and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111487679B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108182450A (en) * | 2017-12-25 | 2018-06-19 | 电子科技大学 | A kind of airborne Ground Penetrating Radar target identification method based on depth convolutional network |
| WO2019136946A1 (en) * | 2018-01-15 | 2019-07-18 | 中山大学 | Deep learning-based weakly supervised salient object detection method and system |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101609162A (en) * | 2009-07-17 | 2009-12-23 | 中国石化集团胜利石油管理局 | Pre-stack seismic non-linear three-parameter association inversion method |
| CN106443768B (en) * | 2016-12-14 | 2018-07-13 | 成都理工大学 | A kind of prestack depth domain synthetic seismogram production method |
| CN107607992B (en) * | 2017-08-24 | 2020-08-18 | 电子科技大学 | Multi-wave matching method based on convolutional neural network |
| US20210165938A1 (en) * | 2017-12-14 | 2021-06-03 | Schlumberger Technology Corporation | System and Method for Simulating Reservoir Models |
| CA3112197C (en) * | 2018-09-14 | 2023-08-29 | Bp Corporation North America Inc. | Machine learning-based analysis of seismic attributes |
| CN110967757A (en) * | 2018-09-30 | 2020-04-07 | 中国石油化工股份有限公司 | Real-time pre-stack seismic inversion method and system while drilling |
| CN109799533B (en) * | 2018-12-28 | 2021-07-27 | 中国石油化工股份有限公司 | Reservoir prediction method based on bidirectional circulation neural network |
-
2020
- 2020-04-22 CN CN202010321843.7A patent/CN111487679B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108182450A (en) * | 2017-12-25 | 2018-06-19 | 电子科技大学 | A kind of airborne Ground Penetrating Radar target identification method based on depth convolutional network |
| WO2019136946A1 (en) * | 2018-01-15 | 2019-07-18 | 中山大学 | Deep learning-based weakly supervised salient object detection method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111487679A (en) | 2020-08-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109709603B (en) | Seismic horizon identification and tracking method and system | |
| Ross et al. | P wave arrival picking and first‐motion polarity determination with deep learning | |
| CN112703429A (en) | Machine learning based seismic attribute analysis | |
| US7433851B2 (en) | System and method for inferring geological classes | |
| US11169287B2 (en) | Method and system for automated velocity model updating using machine learning | |
| CN111596978A (en) | Web page display method, module and system for lithofacies classification by artificial intelligence | |
| Wang et al. | Data-driven S-wave velocity prediction method via a deep-learning-based deep convolutional gated recurrent unit fusion network | |
| CN110954950B (en) | Underground transverse wave velocity inversion method, device, computing equipment and storage medium | |
| CN110895348A (en) | Method, system and storage medium for extracting low-frequency information of seismic elastic impedance | |
| CN116047583A (en) | Adaptive wave impedance inversion method and system based on depth convolution neural network | |
| CN111894563A (en) | Method and system for determining classification of fractured reservoir in buried hill section | |
| Yang et al. | Porosity prediction from pre-stack seismic data via a data-driven approach | |
| CN112882123B (en) | CNN well-seismic joint inversion method, system and application based on two-step method | |
| Li et al. | Pertinent multigate mixture-of-experts-based prestack three-parameter seismic inversion | |
| CN111487679B (en) | Transverse wave velocity prediction method, device and equipment | |
| US20230140656A1 (en) | Method and system for determining seismic processing parameters using machine learning | |
| Shi et al. | Acoustic impedance inversion in coal strata using the priori constraint-based TCN-BiGRU method | |
| CN110646846B (en) | Method, device and equipment for determining anisotropic parameters of VTI medium | |
| CN112147679B (en) | Lithology prediction method and device based on elastic parameters under fuzzy logic framework | |
| CN112987091A (en) | Reservoir detection method and device, electronic equipment and storage medium | |
| US20230408718A1 (en) | Method and system for generalizable deep learning framework for seismic velocity estimation robust to survey configuration | |
| US11846740B2 (en) | First break picking for full wavefield travel-time inversion | |
| CN117421642B (en) | Deep learning-based intelligent detector data storage method and related equipment | |
| US20240045094A1 (en) | A method of and apparatus for determining a multiple well seismic-to-well tie | |
| Campos et al. | An Empirical Analysis of the Influence of Seismic Data Modeling for Estimating Velocity Models with Fully Convolutional Networks |
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 |