A reflection layer tracking method and device based on Markov decision process

Abstract

The present invention discloses a reflection layer tracking method and device based on a Markov decision process, including inputting an offset imaging section, preprocessing the imaging section and designing parameters for picking; calculating attribute information of the imaging section, including envelope attributes, phase attributes, inclination attributes, extreme point attributes and wavelet centroid attributes; determining the number and position of seed points; establishing a Markov decision process model, and using the Markov decision process to pick up reflection layer information; judging the validity of the picking result, and storing valid reflection layer information. The present invention introduces the Markov decision process into the event picking process, interacts with the environment by picking up targets, perceives the state of the environment, takes actions to obtain benefits, and affects the state of the environment, and can continuously learn from it to achieve the optimal picking of the event.

Description

Technical Field

[0001] The invention relates to the technical field of geophysical exploration, in particular to a reflection layer tracking method and device based on Markov decision process. Background Art

[0002] In seismic exploration, the reflection structure information on the imaging section plays an important role in both velocity inversion and seismic geological interpretation. In the process of velocity inversion, the rational application of reflection structure information can effectively improve the convergence of velocity inversion, constrain the inversion solution, and thus obtain a velocity model with geological significance. In seismic geological interpretation, the relationship between geological layers can be determined based on the clear reflection feature information on the imaging section, which improves the reliability and accuracy of seismic geological interpretation, and is conducive to establishing an accurate reservoir model and improving reservoir prediction accuracy. Therefore, it is a very important scientific issue to automatically and efficiently extract the desired reflection structure information from the complex and changeable imaging section. At present, there are mainly the following methods for automatically picking or tracking the phase axis on the indexed imaging section: 1. Mainly based on various attribute information on the seismic section, according to the lateral similarity of the phase axis, various algorithms (signal and image processing methods) are used to identify and extract the position of the phase axis. For example, using amplitude information (Howard, 1991), phase information (Zeng et al., 1998; Stark, 2003; Wu and Zhong, 2012), waveform information (Marfurt et al., 1998; Lacaze and Valding, 2009; Zou et al., 2013), dip / slope / structure information (Bakker, 2002; Lomask et al., 2006; Parks, 2010; Wu, 2017) or using multiple attribute information to jointly constrain (Lou and Zhang, 2018; Li et al., 2020; Z et al., 2021). hang et al., 2020); 2. Adopt intelligent learning algorithms, mainly including artificial neural network methods for supervised learning (Harrigan et al., 1992; Huang et al., 2005; Peters et al., 2019) and convolutional neural networks (Waldeland et al., 2018; Shi et al., 2019; Tschannen et al., 2020; Zhang et al., 2021); self-organizing network methods for unsupervised learning (Huang et al., 1990; Lu Wenkai and Mou Yongguang, 1998) and convolutional autoencoder network methods (Shi et al., 2020).

[0003] However, conventional picking methods based on attribute information are heavily dependent on attributes, and the quality of the attributes restricts the effectiveness of the methods. When the attribute changes (wavelet morphology characteristics) in the lateral direction are more drastic, it is difficult for such methods to track the correct event axis position. Using machine learning methods, with the help of big data analysis, it is possible to learn the structural information on the imaging profile based on the statistical characteristics of the data, which has broad application prospects. However, supervised network algorithms require a large amount of training data and labels, and the training process consumes a lot of time. Unsupervised network algorithms have problems such as no clear physical meaning and unreliable result evaluation. Therefore, it is necessary to explore more suitable and high-precision reflection structure picking methods.

[0004] If the automatic event-picking problem is regarded as an optimal decision-making process, computer algorithms can be used to replace manual picking by imitating human behavior. In machine learning algorithms, in addition to supervised learning algorithms and unsupervised learning algorithms, there is another type of learning algorithm. This type of learning algorithm learns through interaction with the environment, and makes decisions through continuous trial and error and benefits. This type of algorithm is called reinforcement learning, and its core algorithm relies on the Markov decision process. This method introduces the Markov decision process (MDP) into the event-picking process. By picking up targets and interacting with the environment, perceiving the state of the environment, taking actions to gain benefits, and influencing the state of the environment, it is possible to continuously learn from it and achieve the optimal picking of the event. Summary of the invention

[0005] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0006] In view of the above problems and / or the problems existing in the existing picking methods, the present invention is proposed.

[0007] Therefore, the problem to be solved by the present invention is how to automatically and efficiently extract the desired reflection structure information from the complex and changeable imaging profiles.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0009] In a first aspect, an embodiment of the present invention provides a reflection layer tracking method based on a Markov decision process, which includes inputting a migration imaging section, preprocessing the imaging section and designing a picking parameter;

[0010] Calculating attribute information of the imaging section, including envelope attributes, phase attributes, inclination attributes, extreme point attributes and wavelet centroid attributes;

[0011] Determine the number and location of seed points;

[0012] Establish a Markov decision process model and use the Markov decision process to pick up reflection layer information;

[0013] Determine the validity of the picking results and store the valid reflection layer information.

[0014] As a preferred solution of the reflection layer tracking method based on Markov decision process of the present invention, wherein: the picking parameters include the maximum number N of event axis picking max , used to control the maximum number of picked-up events; event spacing parameter W axis , used to constrain the minimum interval between two in-phase axes; the in-phase axis termination threshold ST ε , used to determine the event axis stop criterion. When the tracking value function is lower than the threshold, tracking is stopped.

[0015] As a preferred solution of the reflection layer tracking method based on Markov decision process of the present invention, wherein: the number and position of the seed points are determined including:

[0016] According to the different attributes calculated, edge detection is used to calculate the edge points of different attributes;

[0017] Cluster the edge results with different attributes and determine the corresponding center point positions;

[0018] According to the event interval constraint and the event maximum number constraint, the initial point position of each event is extracted in turn.

[0019] As a preferred solution of the reflection layer tracking method based on Markov decision process described in the present invention, the Markov decision process model includes decision time, state space, action space, state transition probability and benefit function, wherein, in the two-dimensional imaging section, the longitudinal grid point is nz, and the transverse grid point is nx; the state is the longitudinal spatial position on the imaging section, the set of states is expressed as S = [1, ..., nz], and the action is to move from the transverse position x i To horizontal position x i+1 The movement is represented by a(s):x i →x i+1 .

[0020] As a preferred solution of the reflection layer tracking method based on the Markov decision process of the present invention, wherein: the decision time is the lateral spatial position on the imaging section, t∈[1,...,nx], wherein, in the two-dimensional imaging section, the longitudinal depth grid point is nz, and the lateral grid point is nx;

[0021] The width of the state space is Where dz1 and dz2 are the widths of the upper and lower interfaces respectively. is the middle position of the state space at different decision moments;

[0022] Under the guidance of the inclination attribute information, the action space adopts a random strategy to achieve the mapping from state to action. The formula is:

[0023]

[0024] Where p is the inclination angle at the current state s at time t, a0 is the calculated center position of the selected action, strategy μ represents the probability of executing action a in state s, and σ is the variance (width) of the Gaussian probability distribution;

[0025] The state transition probability is expressed as the state from time t-1 Take action a to transfer to the state at time t Probability Therefore, the system changes from the state Transfer to state The probability is:

[0026]

[0027] The profit function is determined according to the similarity of the multi-attribute parameters of adjacent channels. The profit function at the current decision moment is:

[0028]

[0029]

[0030] Where Δf=||f(s i )-f(s j )|| 2 represents the attribute distance between states, the F(·) operator is a measure of attribute difference, and R f is the maximum benefit value of attribute f, r f represents the benefit value of attribute f, α f Represents the weight when using attribute f.

[0031] As a preferred solution of the reflection layer tracking method based on Markov decision process described in the present invention, wherein: the use of Markov decision process to pick up reflection layer information includes:

[0032] When the decision moment reaches the end point (the boundary of the profile), or when the long-term cumulative benefit is less than the given threshold (there is a fault), it will automatically enter the termination state, stop tracking the event axis, and record the length of the event axis.

[0033] As a preferred solution of the reflection layer tracking method based on Markov decision process described in the present invention, wherein: the judging validity of the picking result and storing the valid reflection layer information comprises:

[0034] If the tracked event is shorter than the given threshold length, the event will be discarded and the tracking process of the next seed point will be started again;

[0035] Otherwise, the reflection structure information of the current tracking is recorded and the tracking process of the next seed point is continued;

[0036] When all seed points are tracked, the final reflection structure information is output.

[0037] In a second aspect, an embodiment of the present invention provides a reflection layer tracking system based on a Markov decision process, which includes: a processing module for inputting a migration imaging section and preprocessing the imaging section;

[0038] A calculation module, used for designing the picking parameters and calculating the attribute information of the imaging section, and determining the position of the seed point according to the calculated different attributes;

[0039] A modeling module is used to establish a Markov decision process model and pick up reflection layer information using the Markov decision process;

[0040] The analysis module is used to determine the validity of the picking results and store the valid reflection layer information.

[0041] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: the processor implements any step of the above method when executing the computer program.

[0042] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: the computer program implements any step of the above method when executed by a processor.

[0043] The beneficial effects of the present invention are:

[0044] 1. Imaging profile results often cannot be directly mapped to the event axis. By mapping the seismic imaging profile to the attribute space, the differences in the event axis (vertical differences and lateral continuity) can be highlighted, so as to detect and extract the event axis. However, since the imaging profile contains coherent and incoherent noise, the imaging profile changes dramatically. Especially in the case of weak reflection, fault rupture zones, unconformity surfaces, etc., the calculated single attribute results are often incorrect and the correct event axis cannot be tracked. Multi-attributes can highlight different aspects of the properties of the imaging profile, and the shortcomings of a single attribute can be compensated by the joint constraints of multiple attributes.

[0045] 2. The reflection event (reflection structure) picking on the imaging section is regarded as an intelligent decision-making problem in a random environment. By picking the target to interact with the environment, perceiving the state of the environment, taking actions to gain benefits, and influencing the state of the environment, we can continuously learn from it and achieve the optimal picking of the event. By considering the optimal Markov optimal decision process, we can avoid the influence caused by waveform changes in a single channel or a local area, and robustly pick up the desired reflection structure information.

[0046] 3. The automated reflection layer picking results can be used not only in velocity modeling, but also in the interpretation of seismic geological layers, and have very wide application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0048] Figure 1 Flowchart of reflective structure tracing based on Markov decision process.

[0049] Figure 2 Schematic diagram of local migration imaging results.

[0050] Figure 3 It is the envelope property of the local section.

[0051] Figure 4 is the phase attribute of the local section.

[0052] Figure 5 is the inclination attribute of the local section.

[0053] Figure 6 A comparison chart of the picking results with the event axis picking results superimposed on the imaging results.

[0054] Figure 7Schematic diagram of the results of picking multiple reflection structures on the whale model.

[0055] Figure 8 This is the migration imaging result under the background velocity of the overthrust model.

[0056] Fig. 9 This is the reflection layer result picked on the overthrust model. DETAILED DESCRIPTION

[0057] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0058] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0059] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0060] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0061] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0062] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0063] Example 1

[0064] Reference Figure 1 , which is the first embodiment of the present invention, and provides a reflection layer tracking method based on a Markov decision process, comprising:

[0065] S100: Input the imaging section, perform some appropriate preprocessing on the imaging section (mainly including denoising, resolution enhancement, amplitude equalization and other operations), and design the picking parameters.

[0066] It should be noted that the picking parameters mainly include the maximum number of event axes picked up, N max , used to control the maximum number of picked-up events; event spacing parameter W axis , is the minimum interval between two in-phase axes; the in-phase axis termination threshold ST ε , when the tracking value function is lower than the threshold, stop tracking.

[0067] S200: Calculating attribute information of the imaging section, including envelope attributes, phase attributes, inclination attributes, extreme point attributes and wavelet centroid attributes, to jointly constrain the picking result.

[0068] (1) Phase attribute

[0069] It should be noted that the phase attribute is obtained by applying Hilbert transform to the seismic section to generate a complex signal. For the seismic section, apply Hilbert transform:

[0070] h(z,x)=h(z,x)+ih'(z,x) (1)

[0071] where h'(z,x) is the imaginary part of h(z,x). h(z,x) is the complex seismic trace.

[0072]

[0073]

[0074] Where env(z,x) is the envelope and θ(z,x) is the instantaneous phase.

[0075] However, the range of the instantaneous phase solved by the above formula is (-180, 180), which is discontinuous. Therefore, the following is used to calculate the cosine value of the instantaneous phase:

[0076]

[0077]

[0078] Furthermore, considering that when env(z,x) is zero, the division will cause instability, this method uses the least squares inversion division method to optimize the calculation of the instantaneous phase cosine value:

[0079] cos(θ(z,x))=argmin||h(z,x)-cos(θ(z,x))env(z,x)|| 2 (6)

[0080] Finally, the instantaneous phase cosine is calculated as:

[0081]

[0082] Among them, eps is a given minimum threshold to ensure the stability of the calculation.

[0083] (2) Tilt attribute

[0084] It should be noted that the dip attribute is obtained by the structure tensor method. Considering the oscillation of the wavelet and the existence of noise, the imaging section is first smoothed and filtered, and then the envelope is obtained by Hilbert transform, and the structure tensor is obtained by using the envelope value.

[0085] Given an imaging profile envelope A, its structure tensor matrix can be expressed as:

[0086]

[0087] in, Represents the gradient of the wavefield envelope, env x ,env z are the directional derivatives in the x-direction and z-direction respectively. Perform eigenvalue decomposition on the structure tensor matrix to obtain the eigenvalues ​​and eigenvectors of the structure tensor matrix:

[0088]

[0089]

[0090] where λ e is the largest eigenvalue, representing the tensor energy in the direction of the first eigenvector e. The direction of the eigenvector e is consistent with the direction of the gradient; vis the smallest eigenvalue, representing the tensor energy in the direction of the smallest eigenvector v, where the eigenvector v represents the direction parallel to the gradient. Therefore, the inclination angle on the imaging section can be calculated by the following formula:

[0091]

[0092] (3) Extreme point attributes

[0093] It should be noted that the extraction of extreme points mainly includes maximum and minimum. The maximum and minimum can be extracted from the waveform, and the extreme points can also be extracted from other attribute results (such as envelope, phase). There are many ways to obtain extreme points, such as through first-order derivatives, second-order derivatives, etc. This method extracts extreme points by comparing the size in the local window.

[0094] The maximum and minimum values ​​are calculated using the following strategies:

[0095]

[0096]

[0097] Among them, f(x) is the selected attribute value. Through multiple weighted superposition of different local windows, the existence of outliers can be effectively avoided. In addition, according to the actual situation, the picking of isotropic axes is required to be sparse, and the adjacent extreme value points are processed by the maximum value suppression method.

[0098] (4) Wavelet centroid properties

[0099] It should be noted that the centroid of a wavelet (waveform) can be expressed as an arithmetic mean of all points in the wavelet:

[0100]

[0101] This method depends on the size of the local window and needs to estimate the approximate range of the sub-wave. Therefore, this property is applied to the state transfer and optimal decision process in the second stage of the picking process, and the size of the local window is determined by the constraints of the first step.

[0102] S300: Establish a coherent axis picking model for Markov decision process, including decision time, state space, action space, state transition probability and benefit function.

[0103] (1) Decision-making moment

[0104] It should be noted that the decision time is defined as the transverse spatial position on the imaging section. t∈[1,...,nx], (in a two-dimensional imaging section, the longitudinal depth grid point is nz and the transverse grid point is nx).

[0105] (2) State space

[0106] It should be noted that the longitudinal spatial position on the imaging section is defined as a state, and the set of states is represented by S = [1, ..., nz]. Considering that there are multiple event axes in the section, when picking a certain event axis, the position farther away in the longitudinal direction has less influence on the picking of the current event axis. This method reduces the state space to a width within the specified range. Where dz1 and dz2 are the width of the upper interface and the width of the lower interface respectively. is the middle position of the state space at different decision moments. is the position of the seed point, and at other times, the position of this point is determined by picking the event using the conventional multi-attribute method. Reducing the size of the state space can not only reduce the impact of events at other locations on the current picking, but also improve the computational efficiency.

[0107] (3) Action Space

[0108] It should be noted that the action is defined as moving from the lateral position x i To horizontal position x i+1 Movement, a(s):x i →x i+1 The range of action movement is from the current state to the next state range, so the set of actions is expressed as A = [a1, ..., a M ]. Under the guidance of the inclination attribute information, this method uses a random strategy to achieve the mapping from state to action:

[0109]

[0110] Among them, p is the inclination angle at the current state s at time t, a0 is the calculated center position of the selected action, the strategy μ represents the probability of executing action a in state s, and σ is the variance (width) of the Gaussian probability distribution. Along the direction of the inclination angle, the probability of selecting the action is large, and away from the direction of the inclination angle, the probability of selecting the action is small. Depending on the state position, the probability of executing the action is also different.

[0111] (4) State transition probability

[0112] It should be noted that the state transition probability is expressed as the state from time t-1 Take action a to transfer to the state at time t Probability Given the state-action at the decision moment, the occurrence of the state at the next moment is relatively independent. Transfer to state The probability is:

[0113]

[0114] (5) Profit function

[0115] It should be noted that this method defines the profit function in a high-dimensional multi-attribute framework. The state profit function value at the current decision moment is determined according to the similarity of the multi-attribute parameters of adjacent channels. Therefore, the profit function at the current decision moment can be defined as:

[0116]

[0117]

[0118] Where Δf=||f(s i )-f(s j )|| 2 represents the attribute distance between states, the F(·) operator is a measure of attribute difference, and R f is the maximum benefit value of attribute f, r f represents the benefit value of attribute f, α f Indicates the weight when attribute f is used. The weight coefficient of the state attribute is determined by the fuzzy clustering analysis method and the cross-validation method. First, the conventional event picking method is used to track different attributes, and the size of the weight coefficient is determined according to the focusing degree of the tracking results of different attributes.

[0119] S400: Determine the position of the initial seed point, the size of the state space and the discount factor, wherein the initial seed point corresponds to the initial state in the state space.

[0120] The location of the seed point can be determined by:

[0121] According to the different attributes calculated, edge detection is used to calculate the edge points of different attributes;

[0122] Cluster the edge results with different attributes and determine the corresponding center point positions;

[0123] According to the event interval constraint and the event maximum number constraint, the initial point position of each event is extracted in turn.

[0124] According to the seed point position at the starting time The conventional attribute tracking method is used to preliminarily pick the event axis. This method uses waveform similarity, phase similarity and extreme point similarity for tracking. The tracking results are CF waveform CF phase CF extre , and then superimpose them to form the initial state space:

[0125]

[0126] If the lateral fluctuation is not drastic, the initial state space size can also be defined as follows:

[0127]

[0128] Only the impact of the current decision-making state on the subsequent finite steps is calculated, and the discount factor is defined as:

[0129]

[0130] Among them, N t The maximum impact range determined in advance.

[0131] S500: Reflection event axis tracking process and outputting reflection structure information.

[0132] It should be noted that after defining the event picking process as a high-dimensional multi-attribute Markov state transfer model, it can be tracked and solved based on the changes in state and benefits.

[0133] When the decision moment reaches the end point (the boundary of the profile), or when the long-term cumulative benefit is less than the given threshold (there is a fault), it automatically enters the termination state, stops tracking the sync axis, and records the length of the sync axis.

[0134] It should be noted that if the tracked event is less than the given threshold length, the event will be discarded and the tracking process of the next seed point will be started and re-tracked. Otherwise, the reflection structure information currently tracked is recorded and re-tracked. When all seed points are tracked, the final reflection structure information is output.

[0135] Furthermore, this embodiment also provides a reflection layer tracking system based on a Markov decision process, comprising:

[0136] A processing module, used for inputting a migration imaging section and preprocessing the imaging section;

[0137] A calculation module, used for designing the picking parameters and calculating the attribute information of the imaging section, and determining the position of the seed point according to the calculated different attributes;

[0138] A modeling module is used to establish a Markov decision process model and pick up reflection layer information using the Markov decision process;

[0139] The analysis module is used to determine the validity of the picking results and store the valid reflection layer information.

[0140] This embodiment also provides a computer device, which is applicable to the case of the reflection layer tracking method based on the Markov decision process, including:

[0141] Memory and processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the reflection layer tracking method based on Markov decision process proposed in the above embodiment.

[0142] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0143] This embodiment further provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for tracing a reflection layer based on a Markov decision process as proposed in the above embodiment is implemented.

[0144] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be found in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0145] Example 2

[0146] Reference Figures 2 to 9 , which is the second embodiment of the present invention, and this embodiment provides a reflection layer tracking method based on a Markov decision process. In order to verify the beneficial effects of the present invention, a scientific demonstration is carried out through comparative experiments.

[0147] First, the application effect of using this method to pick up a single event axis is tested.

[0148] Figure 2 It is the local section of the migration imaging of a theoretical model data when the background velocity is inaccurate.

[0149] It can be seen that due to the inaccurate migration speed, the imaging result is not focused, a large number of migration arc artifacts appear in the section, and the sub-waveform in the lateral direction changes greatly.

[0150] Figures 3 to 5The envelope, phase and dip properties of the local profile are shown respectively.

[0151] It can be seen that different attributes can highlight the characteristics of a certain part of the imaging section, such as Figure 3 As shown in the black box, when the envelope attribute is discontinuous in the horizontal direction, Figure 4 The phase attribute shown in the black box has good continuity, so the use of multiple attributes can complement the defects of a single attribute.

[0152] But there are also cases where all attributes are bad, such as Figures 3 to 5 As shown in the black circle, if we track based on the conventional single-channel similarity, we cannot track the accurate reflection structure and obtain effective information of the area. Although all attributes are not good in the local area, there is a certain similarity in the lateral local neighborhood of the area. Therefore, we can use the attributes of the front and back neighborhoods and adopt a multi-step prediction method to determine the reflection structure of this part of the area.

[0153] Figure 6 The results of conventional single-channel cross-correlation picking and the combination of multi-attribute and Markov decision process proposed by this method are shown. It can be seen that the drastic changes in the lateral waveform and the attribute differences in the region make it impossible for the conventional single-channel method to correctly pursue the position of the event axis. The Markov decision process method based on high-dimensional multi-attribute and multi-channel benefits can effectively avoid local noise interference and successfully track the correct reflection layer.

[0154] Next, the migration imaging results of the whale model are used to test the picking of multiple reflection structures.

[0155] In this embodiment, the maximum number of seed points is set to 5, the minimum interval of the phase axis is 0.1km, and the minimum length of the phase axis is 1 / 10 of the number of horizontal points. Figure 7 On this basis, after the state space is determined by conventional multi-attribute constraints, the Markov decision method is used to automatically track the event axis. The result is as follows Figure 7 As shown in Figure 2, it can be seen that the reflection structure extracted by this method is consistent with the real structure, which shows the effectiveness of this method.

[0156] Finally, the effectiveness of this method is further tested using 2D sections in a more complex overthrust model. Figure 8 To use the background velocity model to migrate the imaging results, Fig. 9 This is the reflection layer result picked by this method.

[0157] In this case, only 8 reflection layers were extracted. Due to inaccurate background velocity, the reflection layers in the offset imaging section are unfocused and discontinuous, making it difficult to track the correct reflection layer information using conventional methods. However, the Markov decision process proposed by this method can track the correct reflection layer information, so that the desired reflection layer can be selectively screened out and applied to velocity modeling and seismic interpretation.

[0158] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A reflection layer tracking method based on Markov decision process, characterized in that: include, Inputting a migration imaging section, preprocessing the imaging section and designing picking parameters; Calculating attribute information of the imaging section, including envelope attributes, phase attributes, inclination attributes, extreme point attributes and wavelet centroid attributes; Determine the number and location of seed points; Establish a Markov decision process model and use the Markov decision process to pick up reflection layer information; Determine the validity of the picking results and store the valid reflection layer information; The Markov decision process model includes decision time, state space, action space, state transition probability and benefit function, wherein in a two-dimensional imaging section, the longitudinal grid point is nz, the transverse grid point is nx, the state is the longitudinal spatial position on the imaging section, the set of states is represented by S = [1, ..., nz], and the action is to move from the transverse position x i To horizontal position x i+1 The movement is represented by a(s):x i →x i+1 ; The decision time is the lateral spatial position on the imaging section, t∈[1,...,nx]; The width of the state space is Where dz1 and dz2 are the widths of the upper and lower interfaces respectively. is the middle position of the state space at different decision moments; Under the guidance of the inclination attribute information, the action space adopts a random strategy to achieve the mapping from state to action. The formula is: Where p is the inclination angle at the current state s at time t, a0 is the calculated center position of the selected action, strategy μ represents the probability of executing action a in state s, and σ is the variance of the Gaussian probability distribution; The state transition probability is expressed as the state from time t-1 Take action a to transfer to the state at time t Probability Therefore, the system changes from the state Transfer to state The probability is: The profit function is determined according to the similarity of the multi-attribute parameters of adjacent channels. The profit function at the current decision moment is: Where Δf=||f(s i )-f(s j )|| 2 represents the attribute distance between states, the F(·) operator is a measure of attribute difference, and R f is the maximum benefit value of attribute f, r f represents the benefit value of attribute f, α f Represents the weight when using attribute f.

2. The method for tracking reflection layers based on a Markov decision process as claimed in claim 1, characterized in that: The picking parameters include the maximum number of event axes picked up, N max , used to control the maximum number of picked-up events; event spacing parameter W axis , used to constrain the minimum interval between two in-phase axes; the in-phase axis termination threshold ST ε , used to determine the event axis stop criterion. When the tracking value function is lower than the threshold, tracking is stopped.

3. The method for tracking reflection layers based on Markov decision process as claimed in claim 2, characterized in that: Determining the number and position of seed points includes: According to the different attributes calculated, edge detection is used to calculate the edge points of different attributes; Cluster the edge results with different attributes and determine the corresponding center point positions; According to the event interval constraint and the event maximum number constraint, the initial point position of each event is extracted in turn.

4. The method for tracking reflection layers based on Markov decision process as claimed in claim 3, characterized in that: The method of picking up reflection layer information by using Markov decision process includes: When the decision moment reaches the end point, or when the long-term cumulative benefit is less than a given threshold, it will automatically enter the termination state, stop tracking the event axis, and record the length of the event axis.

5. The method for tracking reflection layers based on Markov decision process according to claim 1, characterized in that: The determination of the validity of the picking result and the storage of the valid reflection layer information include: If the tracked event is shorter than the given threshold length, the event will be discarded and the tracking process of the next seed point will be started again; Otherwise, the reflection structure information of the current tracking is recorded and the tracking process of the next seed point is continued; When all seed points are tracked, the final reflection structure information is output.

6. A reflection layer tracking system based on a Markov decision process, based on the reflection layer tracking method based on a Markov decision process according to any one of claims 1 to 5, characterized in that: include, A processing module, used for inputting a migration imaging section and preprocessing the imaging section; A calculation module, used for designing the picking parameters and calculating the attribute information of the imaging section, and determining the position of the seed point according to the calculated different attributes; A modeling module is used to establish a Markov decision process model and pick up reflection layer information using the Markov decision process; The analysis module is used to determine the validity of the picking results and store the valid reflection layer information.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of any one of the methods of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of any one of the methods of claims 1 to 5 are implemented.

Images