FR3139915A1 - METHOD AND APPARATUS FOR ESTABLISHING AN INTERPRETATION MODEL FOR STRIPS-START FAULTS, AND INTERPRETATION METHOD AND APPARATUS FOR STRIPS-STRAP FAULTS - Google Patents

Abstract

The present invention relates to the technical field of seismic fault detection, and implements a method and apparatus for establishing an interpretation model for strike-slip faults, an interpretation method and apparatus for strike-slip faults. The method for establishing an interpretive model for strike-slip faults includes the steps of: receiving seismic data and logging information; determine initial strike-slip fault characterization images based on seismic data; adjust the initial strike-slip fault characterization images based on logging information and geological evolution laws to obtain spatial topological images of strike-slip faults; consider the seismic profile data in the seismic data as an input, consider the label information of the spatial topological images of strike-slip fault as an output, construct training samples based on the seismic profile data and images corresponding spatial topologies of strike-slip fault, and construct a neural network model based on the input and output; and training the neural network model using the training samples, and considering a learning-trained neural network model as an interpretation model for strike-slip faults. According to the embodiments of the present invention, the problems of low accuracy of detection and recognition of small faults and heavy workload of seismic interpreters are solved. Fig. 1

Description

METHOD AND APPARATUS FOR ESTABLISHING AN INTERPRETATION MODEL FOR STRIPS-START FAULTS, AND INTERPRETATION METHOD AND APPARATUS FOR STRIPS-STRAP FAULTS

The present invention relates to the technical field of seismic fault detection, and in particular to a method and apparatus for establishing an interpretation model for strike-slip faults, and an interpretation method and apparatus for strike-slip faults.

STATE OF PRIOR ART

This section is intended to provide background information or context for embodiments of the present invention set forth in the claims. The description herein shall not be considered prior art merely because it is included in this section.

As the most important stratum for exploration and development purposes, the carbonate formation is affected by regional compressive and torsional stresses, and a large number of strike-slip faults are developed in it. Different strike-slip fault segments with different characteristics or the same characteristics have obvious differences in terms of reservoir control, and they have a positive effect on carbonate reservoir reconstruction and oil and gas filling. But at the same time, there are also negative impacts of destroying the condensation preservation conditions, and communicating with the bottom water of the oil field, and then reducing the development efficiency of the oil field. Due to the fact that the strike-slip fault has the characteristics that the vertical displacement of the fault is small or even not recognizable, the formation structures on both sides of the fault are significantly changed, and the faults associated are complex, there is great uncertainty in the interpretation of strike-slip faulting in carbonate rocks.

Currently, the interpretation process for strike-slip faults mainly includes two types, the analysis and interpretation of strike-slip faults manually and by artificial intelligence, respectively. Based on seismic data processing or manual analysis of seismic attribute imagery, this process is very costly in terms of time and labor and there are the inevitable personal errors. Existing artificial intelligence fault interpretation methods only characterize large faults through seismic attributes, and cannot achieve precision detection and recognition for small faults, resulting in prone to high ambiguity and uncertainty of interpretation, and tending to miss or misdetect flaws. Additionally, many manual interpretations are required for intervention purposes to complete corrections and filtrations, and the manual workload remains high.

Therefore, it is urgent to implement an intelligent interpretation method for strike-slip faults to solve the problems of low detection and recognition accuracy of small faults and heavy characterization workload of interpreters. seismic.

Due to the low accuracy of detection and recognition of small faults and the heavy characterization workload of seismic interpreters, this solution is proposed to overcome or to at least partially resolve the above problems.

According to one aspect, the embodiments of the present invention aim to implement a method for establishing an interpretation model for strike-slip faults, comprising the steps consisting of:

receive seismic data and logging information;

determine initial strike-slip fault characterization images based on seismic data;

adjust the initial strike-slip fault characterization images based on logging information and geological evolution laws in order to obtain spatial topological images of strike-slip faults;

consider the seismic profile data in the seismic data as an input, consider the label information of the spatial topological images of strike-slip fault as an output, construct training samples based on the seismic profile data and images corresponding spatial topologies of strike-slip fault, and construct a neural network model based on the input and output; And

training the neural network model using the training samples, and considering a learning-trained neural network model as an interpretation model for strike-slip faults.

Furthermore, the step consisting of determining initial strike-slip fault characterization images based on the seismic data includes the steps consisting of:

determine seismic attributes and seismic reflection features based on the seismic data;

determining sensitive seismic attributes based on the seismic reflection features together with two seismic attribute analysis modes in terms of profile attributes and slice attributes; And

determine initial strike-slip fault characterization images based on sensitive seismic attributes and seismic reflection features.

Furthermore, the step of adjusting the initial strike-slip fault characterization images based on the logging information and the laws of geological evolution in order to obtain spatial topological images of strike-slip faults comprises the steps consisting of has :

obtaining corresponding petro-physical information based on the logging information;

obtain analysis results of geological structures that correspond with the corresponding petro-physical information based on the laws of geological evolution; And

adjust the initial strike-slip fault characterization images based on the analysis results of corresponding geological structures to obtain the spatial topological images of strike-slip faults.

Furthermore, the neural network model is constructed by a UNet++ neural network; and parameters in the neural network model are trained by learning and optimized using an adaptive type moment estimation optimizer.

Furthermore, the step of training the neural network model using the training samples comprises the steps of:

use the training samples and the neural network model to construct a loss function as in the following equation:

Or, designates the seismic profile data that has been entered into the model, denotes a probability matrix obtained by the processing of through the neural network model, denotes label information of a spatial topological image of strike-slip fault corresponding to the seismic profile data, denotes a summation of all elements in a matrix , designates a number of seismic profiles, denotes a total number of seismic profiles, and is a balancing coefficient; And

perform training of the neural network model based on the training samples and the loss function.

Furthermore, the step of training the neural network model by means of the training examples comprises the steps of:

divide the training samples into a training set and a validation set;

training the neural network model using the training set;

respectively inputting seismic profile data into the validation set into the neural network model trained by learning;

calculating an accuracy based on model output results and spatial topological images of strike-slip faults in the validation set; And

retraining the neural network model if the accuracy is lower than a set threshold, otherwise completing training of the neural network model.

According to another aspect, the embodiments of the present invention also implement an apparatus for establishing an interpretation model for strike-slip faults, comprising:

a receiving module configured to receive seismic data and logging information;

an initial strike-slip fault characterization image construction module configured to determine initial strike-slip fault characterization images based on the seismic data;

a strike-slip fault spatial topological image construction module configured to adjust the initial strike-slip fault characterization images based on logging information and geological evolution laws to obtain spatial topological images of strike-slip fault;

a neural network model building module configured to consider the seismic profile data in the seismic data as an input, consider the label information of the spatial topological images of strike-slip fault as an output, construct samples of learning based on the seismic profile data and corresponding spatial topological images of strike-slip fault, and constructing a neural network model based on the input and output; And

a neural network model training module configured to train the neural network model using the training samples, and consider a training-trained neural network model to be an interpretation model for the strike-slip faults.

Based on the same concept of the invention, the embodiments of the present invention also implement an interpretation method for strike-slip faults, comprising the steps consisting of:

receive seismic data from a well for analysis; And

inputting the seismic profile data into the seismic data into an interpretation model for strike-slip faults which is trained in the method according to any of the above embodiments, to obtain images of the strike-slip fault.

According to another aspect, the embodiments of the present invention also implement an interpretation apparatus for strike-slip faults, comprising:

a receiving module configured to receive seismic data from a well for analysis; And

an interpretation module configured to input the seismic profile data into the seismic data into an interpretation model for strike-slip faults which is trained by training in the method according to any of the above embodiments, in to obtain images of the strike-slip fault.

According to another aspect, the embodiments of the present invention further implement a computing device, comprising a memory, a processor, and a computer program stored in the memory, and when executed by the processor, the computer program implements instructions of the method described in any of the embodiments mentioned above.

According to another aspect, the embodiments of the present invention further implement a computer storage medium in which a computer program is stored, and when it is executed by a processor of a computing device, the computer program implements implementation of the instructions for the process mentioned above.

One or more technical solutions according to the embodiments of the present invention have at least the following technical effects.

In embodiments of the present invention, initial strike-slip fault characterization images are obtained based on the seismic data, and then spatial topological images of strike-slip fault are obtained by enhancing the initial strike-slip fault characterization images. strike-slip fault based on logging information and geological evolution laws, so as to obtain interpretation results for precision strike-slip faults, and the spatial topological images of strike-slip fault correspond to the profile data seismic in seismic data. Then, the seismic profile data in the seismic data is taken as an input, and the label information of the spatial topological images of strike-slip fault is taken as an output; training samples are constructed based on the seismic profile data and corresponding spatial topological images of strike-slip faults; a neural network model is built based on the input and output; After the neural network model has been trained using the training samples, an interpretation model for the strike-slip faults is eventually obtained which model automatically performs precision interpretation of the strike-slip fault by a small number of training samples, to thereby solve the problem of low detection and recognition precision of small faults and heavy characterization workload of seismic interpreters.

The above description is only an overview of the technical solutions of certain embodiments of the present invention. In order to more clearly understand the technical means of certain embodiments of the present invention, the implementation can be carried out according to the content of the present invention, and in order to render the above objectives, features and advantages and other embodiments of the present invention more obvious and more understandable, specific implementations of certain embodiments of the present invention are illustrated below.

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, a brief description of the drawings for the embodiments will be given below. Of course, the drawings in the following description merely illustrate certain embodiments of the present invention, and those skilled in the art can derive other drawings from them without any creative effort. In the drawings:

illustrates a flow diagram of a method for establishing an interpretive model for strike-slip faults according to certain embodiments of the present invention;

illustrates a schematic diagram of partial 3D seismic data of a work area according to certain embodiments of the present invention;

illustrates a flow diagram for determining initial strike-slip fault characterization images according to certain embodiments of the present invention;

illustrates a schematic diagram of original in-line seismic profiles 75 according to some embodiments of the present invention;

illustrates a schematic diagram of T74 horizon level coherent attribute slices according to certain embodiments of the present invention;

illustrates a schematic diagram of T76 horizon level coherent attribute slices according to certain embodiments of the present invention;

illustrates a schematic diagram of T80 horizon level coherent attribute slices according to certain embodiments of the present invention;

illustrates a schematic diagram of event features represented in a position marked by a box in the according to certain embodiments of the present invention;

illustrates another schematic diagram of event features represented in a position marked by a box in the according to certain embodiments of the present invention;

illustrates a schematic diagram of strike-slip fault structures and line profile fault distribution according to certain embodiments of the present invention;

illustrates a schematic diagram of strike-slip fault structures and cross-line profile fault distribution according to certain embodiments of the present invention;

illustrates a schematic diagram of a process of determining seismic sensitive attributes based on seismic reflection features according to certain embodiments of the present invention;

illustrates a flow diagram of a process of determining spatial topological images of strike-slip faulting according to certain embodiments of the present invention;

illustrates a schematic diagram of spatial topological images of strike-slip fault and original seismic amplitude profile data according to certain embodiments of the present invention;

illustrates a schematic structure diagram of a UNet++ neural network according to certain embodiments of the present invention;

illustrates a learning flow diagram of a neural network model according to some embodiments of the present invention;

illustrates another learning flow diagram of a neural network model according to some embodiments of the present invention;

illustrates a flow diagram of an interpretation method for strike-slip faults according to certain embodiments of the present invention;

illustrates a structural block diagram of an apparatus for establishing an interpretive model for strike-slip faults according to certain embodiments of the present invention;

illustrates a structural block diagram of an interpretation apparatus for strike-slip faults according to some embodiments of the present invention; And

illustrates a structural block diagram of a computing device according to some embodiments of the present invention.

DETAILED DESCRIPTION

Reference numbers

151: reception module;

152: module for constructing initial images for characterization of strike-slip faults;

153: module for constructing spatial topological images of strike-slip faults;

154: neural network model construction module;

155: neural network model learning module;

161: reception module;

162: interpretation module;

1702: computing device;

1704: processor;

1706: memory;

1708: drive mechanism;

1710: input/output interface;

1712: input device;

1714: exit device;

1716: presentation device;

1718: graphical user interface;

1720: network interface;

1722: communications link;

1724: communication bus.

In order that those skilled in the art may better understand the technical solutions of the present invention, the technical solutions in certain embodiments of the present invention will be described clearly and fully below with reference to the drawings in the embodiments of the present invention. Of course, those described are only part, rather than all, of the embodiments of the present invention. Based on certain embodiments of the present invention, any other embodiments obtained by those of ordinary skill in the art without any creative effort must be within the scope of protection of the present invention.

It should be noted that the terms "first", "second" and similar terms in the description and claims of the present invention and the aforesaid drawings are used to distinguish similar objects and not necessarily to describe any particular order or precedence. particular. It is to be understood that data so used may be exchanged under appropriate circumstances such that the embodiments of the present invention described herein may be implemented in an order other than those illustrated or described herein. In addition, the terms “include”, “include”, “have” and any variations thereof are intended to cover all non-exclusive inclusions. For example, a process, process, apparatus, product or device, which includes a series of steps or units, is not necessarily limited to those listed explicitly, but may include other steps or units which are not expressly listed or which are not inherent in the process, process, apparatus, product or device.

Existing interpretation methods for strike-slip faults only characterize large faults through seismic attributes, and cannot achieve precision detection and recognition for small faults, thereby resulting in with high ambiguity and uncertainty of interpretation, and tending to miss or poorly detect flaws. Additionally, many manual interpretations are required for intervention purposes to complete corrections and filtrations, and the manual workload remains high. Due to the low accuracy of detection and recognition of small faults and the heavy workload of seismic interpreters, this solution is proposed to overcome or to at least partially resolve the above problems.

In reference to the , some embodiments of the present invention provide a method for establishing an interpretive model for strike-slip faults. The present invention provides the steps for executing the method as described in the embodiments or in the flow diagrams, but a greater or smaller number of execution steps may be included depending on the work. routine or non-creative. Any order of execution of the steps listed in the embodiments is only one of numerous orders of execution of the steps and does not represent a single order of execution. In the case of a real system or real device product, the steps may be performed in sequence or simultaneously depending on the method illustrated in the embodiments or drawings. Specifically, as illustrated in the , the method can comprise the steps consisting of:

S100: receive seismic data and logging information;

S200: determine initial strike-slip fault characterization images based on seismic data;

S300: adjust the initial strike-slip fault characterization images based on logging information and geological evolution laws to obtain spatial topological images of strike-slip faults;

S400: consider the seismic profile data in the seismic data as an input, consider the label information of the spatial topological images of strike-slip fault as an output, construct training samples according to the seismic profile data, and corresponding spatial topological images of strike-slip faults, and constructing a neural network model based on the input and output;

S500: performing training of the neural network model using the training samples, and considering a neural network model trained by training as an interpretation model for strike-slip faults.

In embodiments of the present invention, initial strike-slip fault characterization images are obtained based on the seismic data, and then spatial topological images of strike-slip fault are obtained by enhancing the initial strike-slip fault characterization images. strike-slip fault based on logging information and geological evolution laws, so as to obtain interpretation results for precision strike-slip faults, and the spatial topological images of strike-slip fault correspond to the profile data seismic in seismic data. Then, the seismic profile data in the seismic data is taken as an input, and the label information of the spatial topological images of strike-slip fault is taken as an output; training samples are constructed based on the seismic profile data and corresponding spatial topological images of strike-slip faults; a neural network model is built based on the input and output; After the neural network model has been trained using the training samples, an interpretation model for the strike-slip faults is eventually obtained which model automatically performs precision interpretation of the strike-slip fault by a small number of training samples, to thereby solve the problem of low precision of detection and recognition of small faults and the heavy workload of seismic interpreters.

In some embodiments, the received seismic data may be from platform institutions, such as seismic information centers in relevant regions or countries, and the logging information may be information, such as geological structures and hydrological information, which can be obtained and used for geological interpretation purposes. In embodiments of the present invention, the illustrates partial 3D seismic data from a work area with an original area of approximately 200 square kilometers. After 3D migration seismic data that matches the strike-slip fault is intercepted, a corresponding size is 400 Inlines × 400 Crosslines × 448 time sampling points, a time interval is from 4502 ms to 5400 ms , and a time sampling interval is 2 ms. As illustrated on the , a high north "X" type shear system is developed in the partial 3D seismic data area of the work area, where the formation is generally flat, the development process is dominated by marine sediment, and faults of strike-slip are mainly developed, as illustrated by the arrow on the . And, strike-slip faults include faults of multiple levels and the structure of the system is complex. The embodiments of the present invention mainly focus on the fault precision interpretation of Fault No. 5 (i.e. the fault on the ) among strike-slip faults. the label information of the spatial strike-slip fault topological images may, for example, include part or all of the spatial strike-slip fault topological images.

In reference to the , in some embodiments, the step of determining initial strike-slip fault characterization images based on the seismic data may comprise the steps of:

S210: determine seismic attributes and seismic reflection features based on seismic data;

S220: determining sensitive seismic attributes based on the seismic reflection features together with two seismic attribute analysis modes in terms of profile attributes and slice attributes;

S230: determine initial strike-slip fault characterization images based on sensitive seismic attributes and seismic reflection features.

Specifically, firstly, the received seismic data are preprocessed, and spectral decomposition is performed by sustained wavelet transform and time-frequency sustained wavelet transform to obtain the seismic data in different frequency domains, between -time, seismic data in an angular domain are divided by angular decomposition. Various seismic attributes, such as coherence and curvature, are extracted based on the preprocessed seismic data. The seismic attribute is a measure of the geometric, kinematic, dynamic and statistical characteristics of seismic data. The seismic attribute can also be understood as a characteristic used to describe and quantify seismic data, and it is a subset of all the information found in the original seismic data.

Currently, seismic attributes are mainly classified into amplitude attributes, frequency attributes, time attributes and geometry attributes. Amplitude attributes include root mean square amplitude, average absolute amplitude, maximum peak amplitude, average peak amplitude, maximum trough amplitude, average trough amplitude, and average energy. The frequency attributes mainly include average instantaneous frequency, root mean square instantaneous frequency, reflection width, etc. The time attributes mainly include coherence, inclination angle, azimuth angle, curvature, etc.

Amplitude attributes are typically used for seismic lithology interpretation and reservoir prediction. Frequency attributes are mainly found in the propagation of seismic waves, the frequency spectrum changes due to attenuation, absorption and geometric diffusion of the formation. Frequency attributes are important reference factors for reservoir thickness, and seismic wave scattering, absorption and attenuation, and time attributes are important reference factors for the purpose of structural analysis.

There are mature algorithms used to extract seismic attributes. For example, the third generation consistent body algorithm can be used to extract consistent attributes. The third generation coherent body algorithm is based on an eigenvalue structure algorithm that needs to calculate an eigenvalue of a covariance matrix, and has the advantages of high resolution and strong noise elimination ability. .

As illustrated on the , the original online seismic profile 75 is selected, it includes three horizons T74, T76 and T80 having different depths. The results illustrated in Figures 5a, 5b and 5c can be obtained once the slices with consistent attributes are extracted from the three horizons T74, T76 and T80 having different depths. It can be seen that the attribute-consistent slices from different horizons have rather different sharpness because the faults are influenced by the surrounding sedimentary environment. The T74 horizon is an intense event where distortion and dislocation are evident, and the consistent attributes exhibit very clear fault morphology, i.e., the boundary characterization of large-scale faults is consistent and complete, there is no has no excessive background information interference around the faults, the related secondary faults are relatively fractured, and the interior detection effect is slightly poor. Reflections of events at fault positions in horizons T76 and T80 are disordered, distortions are anomalous and inconsistent, background information in coherent attributes is confused, boundaries of large faults are unclear , the interior information is interrupted and confused, and the detection is vague. Similarly, corresponding seismic attribute extraction results can be obtained based on other seismic attribute extraction algorithms.

In some embodiments, the seismic reflection features may exhibit changing characteristics of seismic attributes such as seismic event, waveform, and amplitude at different frequencies. There illustrates particularities of events represented in a position marked by a frame in the at low frequency, and the illustrates features of events represented in a position marked by a high-frequency frame. By comparing the and the , it can be seen that the event signal is stronger at low frequency. There are three event dislocations evident in the lower part of the , and the boundaries of large faults on both sides are clear. However, as illustrated in the , at high frequency, the number of events at the same position increases obviously, bifurcations and mergers appear, and the intermediate parts of the large faults on both sides can be determined as small faults . The seismic reflection features of seismic profiles at various frequencies are varied, and the fault distribution morphologies are different. Therefore, the fault distribution situations should be obtained by comprehensive analysis of the seismic reflection features at different frequencies.

Furthermore, since there are currently hundreds of seismic attributes, too many attributes will increase the calculation load, and there is overlap and redundancy of information among seismic attributes, affecting the prediction result of flaws. Therefore, it is necessary to determine the reflection-sensitive seismic attributes of the strike-slip fault to improve the prediction accuracy of strike-slip faults.

In some embodiments, seismic attributes can be analyzed from two perspectives, namely, slices and profiles. For example, in Figures 5a, 5b and 5c, slices with consistent attributes are extracted from three horizons T74, T76 and T80 having different depths respectively, so as to analyze the seismic attributes from the perspective of the slices. Figures 7a and 7b illustrate structural configurations and distributions of strike-slip faults on in-line and transverse line profiles, respectively, in which breakpoints are corresponding salient positions on the fault profiles extracted from the slices at seismic attributes, and marked frame positions indicate fault distribution ranges. It can be seen that strike-slip fault structures on the line profile are obvious with rich features and fixed and concentrated positions, and strike-slip fault structures on the line profile are distributed disorderly and vaguely. In order to reduce the error of the prediction result, the cross-line profile is selected as the main profile for prediction purposes in the embodiment of the present invention, and meanwhile, the construction of the spatial topological structure of strike-slip faults on the line profile is subject to breakpoint connection line constraints using the cross-line breakpoints, so as to further ensure the accuracy of characterization of strike-slip faults.

Furthermore, the sensitive seismic attributes can be determined by filtering the obtained seismic attributes based on the seismic reflection features using the two modes of seismic attribute analysis in terms of profile attributes and slice attributes. As illustrated on the , in the process of determining the seismic sensitive attributes based on the seismic reflection features, the seismic sensitive attributes of faults of different scales can be determined by analyzing the instantaneous amplitude, maximum positive curvature, and slices at consistent attributes of the T74 horizon in the Inline 180 profile in conjunction with the seismic reflection features that correspond to the profile, so as to better recognize small faults.

Once sensitive seismic attributes are determined, it is possible to observe differences in imaging of large- and medium-scale strike-slip faults, particularly small-scale ones at different frequencies based on the combination of profile, seismic slice and cube. Meanwhile, the distribution characteristics of the three-dimensional spatial geological structure are monitored to complete the characterizations of the main faults and their marginal and internal secondary fault zones, crush zones and other micro-geological structures, and further optimize the characterization results based on different amplitudes of strike-slip faults and the distortion and disturbance of the waveform with the change of fault morphology, etc. as indicated by the particularities of seismic reflection, thereby obtaining the initial images of strike-slip fault characterization.

In reference to the , in some embodiments, the step of adjusting the initial strike-slip fault characterization images based on the logging information and geological evolution laws to obtain spatial topological images of strike-slip faults may include the steps of:

S310:obtaining corresponding petro-physical information based on the logging information;

S320: obtain analysis results of geological structures that correspond with the corresponding petro-physical information based on the laws of geological evolution;

S330: adjust the initial strike-slip fault characterization images based on the analysis results of corresponding geological structures to obtain the spatial topological images of strike-slip faults.

Specifically, logging information may be information, such as geological structures and hydrological information, that can be obtained and used for geological interpretation purposes. The petro-physical information relating to the strike-slip fault zone can be obtained from the logging information. Based on the laws of geological evolution (including, but not limited to, dynamic, geometric, kinematic and geological mechanisms formed in strike-slip faults), the corresponding results of geological structure analysis can be obtained by the corresponding petro-physical information. Geological structure refers to the morphology left by the deformation or movement of rock strata or rock masses under the effect of internal and external stresses of the earth. Using the corresponding results of geological structure analysis, the initial strike-slip fault characterization images obtained previously can be optimized and adjusted to obtain the spatial topological images of strike-slip fault. As illustrated on the , the spatial topological images of strike-slip fault having different branching types end up being obtained by corresponding interpretation processes of Figures 3 and 9 for the original seismic amplitude profile data.

In some embodiments, the neural network model is constructed by a UNet++ neural network; and parameters in the neural network model are trained by learning and optimized using an adaptive type moment estimation optimizer.

Specifically, the structure diagram of the UNet++ neural network is illustrated in the , comprising an oversampling layer, a subsampling layer, and a skip connection layer making it possible to perform a superposition of channel dimensions. The adaptive type moment estimation optimizer is an algorithm for optimizing a random objective type function based on a first-order gradient, and is used for the purpose of iteratively updating the weight of the network of neurons based on the training data. The adaptive type moment estimation optimizer has the advantages of simple implementation, high calculation efficiency and low storage capacity requirements, and is suitable for optimization purposes regarding large-scale parameters and data.

In reference to the , in some embodiments, the step of training the neural network model using the training samples comprises the steps of:

S510: Use the training samples and the neural network model to construct a loss function as in the following equation:

Or, designates the seismic profile data that has been entered into the model, denotes a probability matrix obtained by the processing of through the neural network model, denotes label information of a spatial topological image of strike-slip fault corresponding to the seismic profile data, denotes a summation of all elements in a matrix , designates a number of seismic profiles, denotes a total number of seismic profiles, and is a balancing coefficient.

S520: Perform training of the neural network model based on the training samples and loss function.

Specifically, the training samples are the seismic profile data in the seismic data and the spatial topological images of strike-slip fault. Traditional fault detection often adopts a cross-entropy loss function, which has some limitations due to the extremely small proportion of faults in the data. In order to alleviate the unbalanced proportions of positive and negative samples in the process of fault detection, the embodiments of the present invention adopt the balanced cross-entropy for the loss function of the model, where is a balancing coefficient making it possible to control the weights of positive and negative samples in the total loss. In the embodiments of the present invention, since the positive samples (flaws) represent a small proportion in the overall samples, and in combination with the experimental tests, ends up being set to 0.75. The neural network model can be trained by learning based on the training samples and the loss function.

In reference to the , in some embodiments, the method for establishing an interpretation model for strike-slip faults may include the steps of:

S610: divide the training samples into a training set and a validation set;

S620: training the neural network model using the training set;

S630: respectively inputting seismic profile data into the validation set into a neural network model trained by learning;

S640: calculate accuracy based on model output results and spatial topological images of strike-slip faults in the validation set;

S650: Retrain the neural network model if the accuracy is lower than a set threshold, otherwise complete the training of the neural network model.

Specifically, the training samples include the seismic profile data in the seismic data and the spatial topological images of strike-slip fault. The training samples are divided into a training set and a validation set. Given the time cost of manually producing profile labels and traditional deep learning reporting, a ratio of training set to verification set is 10:1 in embodiments of the present invention. After the neural network model has been trained by training with the training set, the seismic profile data in the validation set is input to the training-trained neural network model, respectively, and the accuracy is calculated based on the results output from the model and the spatial topological images of strike-slip faults in the validation set.

Since the final prediction results, i.e., the results output from the model, constitute a probability graph, which shows that the probability value of each point in the graph of the final prediction results is between 0 and 1. As and as the probability value approaches 1, the possibility that the corresponding point is a fault increases, and as the probability value approaches 0, the possibility that the corresponding point is a fault decreases. The probability graph can be split with a threshold of 0.5 to obtain a binary graph containing only 0 and 1, such that the accuracy can be calculated with the spatial topological images of strike-slip faulting.

The formula for precision is:

where Acc denotes the accuracy, TP indicates that the spatial topological image of strike-slip fault is a positive example, and the model output result is also a positive example; TN indicates that the spatial topological image of strike-slip fault is a negative example, and the model output result is also a negative example; FP indicates that the spatial topological image of strike-slip fault is a negative example, and the model output result is a positive example; FN indicates that the spatial topological image of strike-slip fault is a positive example, and the model output result is a negative example; the positive example indicates that the value is 1, and the negative example indicates that the value is 0.

TP , TN , FP and FN in precision can be calculated from Table 1:

Table 1: Calculation tableTP,TN,FPAndF.N. Predicted value = 1
(Model output) Predicted value = 0
(Model output) True value = 1
(Spatial topological image of strike-slip fault) TP F.N. True value = 0
(Spatial topological image of strike-slip fault) FP TN

The neural network model is retrained by learning if the accuracy is below a set threshold, otherwise training of the neural network model is completed. In the embodiments of the present invention, the threshold is set in the range of 90% ±5% in view of preventing the possibility that the accuracy of the prediction result decreases to a certain extent due to the existence of individual profiles with large differences in the seismic data.

Based on the same concept of the invention, with reference to the , in certain embodiments, the interpretation method for strike-slip faults may comprise the steps consisting of:

S710: receive seismic data from a well for analysis;

S720: inputting the seismic profile data into the seismic data into an interpretation model for strike-slip faults which is trained by the method described in any of the embodiments mentioned above, in order to obtain images of the strike-slip fault.

Corresponding to the above method for establishing an interpretation model for strike-slip faults, certain embodiments of the present invention further provide apparatus for establishing an interpretation model for strike-slip faults. . In reference to the , in some embodiments, the apparatus for establishing an interpretation model for strike-slip faults may include:

a receiving module 151 configured to receive seismic data and logging information;

an initial strike-slip fault characterization image construction module 152 configured to determine initial strike-slip fault characterization images based on the seismic data;

a strike-slip fault spatial topological image construction module 153 configured to adjust the initial strike-slip fault characterization images based on logging information and geological evolution laws in order to obtain spatial topological images strike-slip fault;

a neural network model building module 154 configured to consider the seismic profile data in the seismic data as an input, consider the label information of the spatial topological images of strike-slip fault as an output, construct samples learning according to the seismic profile data and the corresponding spatial topological images of strike-slip fault, and constructing a neural network model based on the input and the output; And

a neural network model learning module 155 configured to train the neural network model using the training samples, and consider a trained neural network model to be an interpretation model for strike-slip faults.

Based on the same concept of the invention, in correspondence with the above method used to establish an interpretation model for strike-slip faults, certain embodiments of the present invention also implement a method of interpretation for strike-slip faults. In reference to the , in certain embodiments, the interpretation apparatus for strike-slip faults may include:

a reception module 161 configured to receive seismic data from a well for analysis purposes;

an interpretation module 162 configured to input the seismic profile data into the seismic data into an interpretation model for strike-slip faults which is trained by training in the method described in any of the embodiments mentioned above above, in order to obtain images of the strike-slip fault.

For ease of description, the above device is described as being divided into different units based on its functions and described respectively. Of course, the functions of the different units can be carried out in the same one or more software and/or hardware when the present invention is implemented.

Please note that user information (including, but not limited to, information relating to the user's device, user's personal information, etc.) and data (including, but not limited to, non-limiting data for analysis purposes, stored data, displayed data, etc.) involved in the embodiments of the present invention are all information and data admitted for use by the user and fully authorized by all parties.

The embodiments of the present invention further implement a computing device. As illustrated on the , in some embodiments of the present invention, the computing device 1702 may include one or more processors 1704, such as one or more central processing units (CPUs) or graphics processors (PGs), and each processing unit may implement one or more hardware threads. The computing device 1702 may further include any memory 1706 configured to store any type of information such as codes, settings, data, etc. In a specific embodiment, a computer program is stored in memory 1706 and is capable of being executed on the processor 1704, and when executed by the processor 1704, the computer program can implement instructions to a method according to any of the above embodiments. Without limitation, for example, memory 1706 may include any or combinations of any type of RAM (random access memory), any type of ROM (read only memory), a flash memory device , hard disk, optical disk, etc. More generally, any memory can use any technology used to store information. Furthermore, any memory can provide volatile or non-volatile retention of information. Furthermore, any memory can represent a fixed or removable component of the computing device 1702. In a scenario where the processor 1704 executes associated instructions stored in any memory or combinations of memories, the computing device 1702 can perform any operation of the associated instructions. The computing device 1702 further includes one or more drive mechanisms 1708 (e.g., a hard disk drive, an optical disk drive, etc.) interacting with any memory.

The computing device 1702 may further include an input/output (I/O) interface 1710 configured to receive different inputs (through an input device 1712) and to provide different outputs (through an output device 1714). A specific output mechanism may include a presentation device 1716 and an associated graphical user interface (GUI) 1718. In other embodiments, the input/output (I/O) interface 1710, the presentation device input 1712, and output device 1714 may not be included, and the device is only used as a computing device in the network. The computing device 1702 may further include one or more network interfaces 1720 configured to exchange data with other devices through one or more communications links 1722. One or more communications buses 1724 couple the components together described above.

The communications link 1722 may be implemented in any manner, for example, over a local area network, over a wide area network (e.g., the Internet), a point-to-point connection, etc., or through any combination of these. The communications link 1722 may include any combination of a wired link, a wireless link, a router, a gateway function, a name server, etc. governed by any protocol or combination of protocols.

The present invention is described with reference to a flow diagram and/or block diagram of the method, apparatus (system), computer readable storage medium and computer program product according to embodiments of the present invention. invention. It should be appreciated that each flow and/or each block in the flow diagram and/or the block diagram, and the combinations of the flows and/or blocks in the flow diagram and/or the block diagram can be implemented implemented through computer program instructions. The instructions of the computer program may be implemented in a general purpose computer, a specialized computer, an integrated processor, or a processor of another programmable data processing device, to form a machine such that the instructions, which are executed by the computer or by the processor of another programmable data processing device, generate means for carrying out the functions specified in one or more flows in the flow diagram and one or more blocks in the block diagram.

The computer program instructions may also be stored in computer-readable memory capable of causing the computer or other programmable data processing devices to operate in a specific manner, such that the instructions stored in the computer readable memory generates an article of manufacture including an instruction device that implements the designated function(s) in a flow or a plurality of flows in the flow diagram and/or in a block or a plurality of blocks in the block diagram.

The instructions of this computer program may also be loaded onto a computer or other programmable data processing devices, such that a series of execution steps are executed on the computer or other devices programmable to generate the processing performed by the computer, therefore the instructions executed on the computer or other programmable devices provide the steps for implementing the designated function in a flow or a plurality of flows in the diagram flow and/or in a block or a plurality of blocks in the block diagram.

In one possible configuration, the computing device includes one or more processors (UCT), an input/output interface, a network interface and a memory.

The memory may include the form of volatile memory, random access memory (RAM) and/or non-volatile memory such as read only memory (ROM) or flash RAM, etc. among computer-readable media. Memory is an example of computer-readable media.

Computer readable media includes permanent and non-permanent, removable and non-removable media that can perform information storage in any process or technique. The information may be computer-readable instructions, data structures, program modules, or other data. An example of the computer storage medium includes, but is not limited to, phase change memory (PRAM memory), static random access memory (SRAM memory), dynamic random access memory (DRAM memory), other types of memory random access memory (RAM memory), read only memory (ROM memory), electrically erasable and programmable read only memory (EEPROM memory), flash memory or other memory techniques, compact disk with read only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic tape cassettes, magnetic disks or other magnetic storage device or other non-transmission media, which may be used for the purpose of storing information accessible to a computing device. As defined herein, computer readable media does not include computer readable media of the temporary type (transient media), of the modulated digital signal and carrier wave type.

Those skilled in the art will appreciate that any embodiment of the present invention may be implemented as a method, system or computer program product. Therefore, the present invention may take the form of an all-hardware embodiment, an all-software embodiment, or a combined software and hardware embodiment. Additionally, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, magnetic disk memory, CD-ROM, optical storage , etc.) containing program codes usable on a computer.

Embodiments of the present invention can be described in the general context of computer-executable instructions that are executed by the computer, e.g., the program module. Generally, the program module includes a routine, a program, an object, a component, a data structure, etc. performing a particular task or realizing a particular abstract data type. Embodiments of the present invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communications network. In distributed computing environments, program modules may be located in local and remote computer storage media comprising the storage device.

It should also be understood that, in embodiments of the present invention, the term "and/or" is only an association relationship that describes the associated objects, indicating that there may be three relationships. For example, A and/or B can mean that A exists alone, A and B both exist, and B exists alone. Additionally, the character "/" in the present invention generally indicates that the contextually associated objects are in an "or" relationship.

The embodiments herein are all described step by step, and the same or similar portions of the embodiments may refer to one another. Each embodiment emphasizes its distinctions from other embodiments. In particular, the embodiment of the system is described in a simple manner since it is substantially similar to the embodiment of the method, and it is sufficient to refer to the descriptions of the embodiment of the method for the relevant part.

In the description of the present invention, the description of the reference terms "an embodiment", "certain embodiments", "an example", "a specific example" or "examples" and the like means that the features, the specific structures, materials, or features described in conjunction with the embodiment(s) or example(s) are included in at least one embodiment or example of the present invention. In the present invention, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples as appropriate. Furthermore, those skilled in the art may combine different embodiments or examples described in the present invention and particularities thereof if there is no contradiction between them.

These modes described above are merely embodiments of the present invention, rather than limitations of the present invention. For those in the art, the present invention is intended to cover any modification or variation. Any modification, equivalent substitution, improvement, etc. which are made in the spirit and principle of the present invention must be within the scope of the claims of the present invention.

Claims (10)

A method for establishing an interpretation model for strike-slip faults, comprising the steps of:
receive seismic data and logging information;
determine initial strike-slip fault characterization images based on seismic data;
adjusting the initial strike-slip fault characterization images based on the logging information and geological evolution laws to obtain spatial topological images of strike-slip faults;
consider the seismic profile data in the seismic data as an input, consider the label information of the spatial topological images of strike-slip fault as an output, construct training samples based on the seismic profile data and images corresponding spatial topologies of strike-slip fault, and construct a neural network model based on the input and output; And
training the neural network model using the training samples, and considering a training-trained neural network model as an interpretation model for strike-slip faults. A method for establishing an interpretation model for strike-slip faults according to claim 1, wherein the step of determining initial strike-slip fault characterization images based on the seismic data comprises the steps of:
determine seismic attributes and seismic reflection features based on the seismic data;
determining sensitive seismic attributes based on the seismic reflection features together with two seismic attribute analysis modes in terms of profile attributes and slice attributes; And
determine initial strike-slip fault characterization images based on sensitive seismic attributes and seismic reflection features. A method for establishing an interpretation model for strike-slip faults according to claim 1, wherein the step of adjusting the initial strike-slip fault characterization images based on the logging information and the evolution laws geological analysis with a view to obtaining spatial topological images of strike-slip faults comprises the steps consisting of:
obtaining corresponding petro-physical information based on the logging information;
obtaining analysis results of geological structures which correspond by the corresponding petro-physical information based on the laws of geological evolution; And
adjusting the initial strike-slip fault characterization images based on the analysis results of corresponding geological structures to obtain the spatial topological images of the strike-slip fault. A method for establishing an interpretation model for strike-slip faults according to claim 1, wherein the neural network model is constructed by a UNet++ neural network; and parameters in the neural network model are trained by learning and optimized using an adaptive type moment estimation optimizer. A method for establishing an interpretation model for strike-slip faults according to claim 1, wherein the step of training the neural network model using the training samples comprises the steps of:
use the training samples and the neural network model to construct a loss function as in the following equation:

Or, designates the seismic profile data that has been entered into the model, denotes a probability matrix obtained by the processing of through the neural network model, denotes label information of a spatial topological image of strike-slip fault corresponding to the seismic profile data, denotes a summation of all elements in a matrix , designates a number of seismic profiles, denotes a total number of seismic profiles, and is a balancing coefficient; And
perform training of the neural network model based on the training samples and the loss function. A method for establishing an interpretation model for strike-slip faults according to claim 1, wherein the step of training the neural network model using the training examples comprises the steps of:
dividing the training samples into a training set and a validation set;
training the neural network model using the training set;
respectively inputting seismic profile data into the validation set into a neural network model trained by learning;
calculating an accuracy based on model output results and spatial topological images of strike-slip faults in the validation set; And
retraining the neural network model if the accuracy is less than a set threshold, otherwise completing training of the neural network model. Interpretation method for strike-slip faults, comprising the steps consisting of:
receive seismic data from a well for analysis; And
inputting the seismic profile data into the seismic data in an interpretation model for strike-slip faults which is trained by the method according to any one of claims 1 to 6, in order to obtain images of the fault of stall. Apparatus for establishing an interpretation model for strike-slip faults, comprising:
a receiving module configured to receive seismic data and logging information;
an initial strike-slip fault characterization image construction module configured to determine initial strike-slip fault characterization images based on the seismic data;
a strike-slip fault spatial topological image construction module configured to adjust the initial strike-slip fault characterization images based on logging information and geological evolution laws to obtain spatial topological images of strike-slip fault;
a neural network model building module configured to consider the seismic profile data in the seismic data as an input, consider the label information of the spatial topological images of strike-slip fault as an output, construct samples of learning based on the seismic profile data and corresponding spatial topological images of strike-slip fault, and constructing a neural network model based on the input and output; And
a neural network model training module configured to train the neural network model using the training samples, and consider a training-trained neural network model to be an interpretation model for the strike-slip faults. Interpretation apparatus for strike-slip faults, comprising:
a receiving module configured to receive seismic data from a well for analysis; And
an interpretation module configured to input the seismic profile data into the seismic data into an interpretation model for strike-slip faults which is trained by learning in the method according to any one of claims 1 to 6, with a view to 'obtain images of the strike-slip fault. Computer device, comprising a memory, a processor, and a computer program stored in the memory, wherein, when executed by the processor, the computer program implements instructions of the method according to any one of claims 1 to 7 .