High-precision reverse-time migration imaging method based on intelligent seismic wave field separation

Abstract

The application discloses a high-precision reverse-time migration imaging method based on intelligent seismic wave field separation, and relates to the technical field of exploration geophysics. The method is based on Hilbert transform to decompose shot point wave field and receiver point wave field into up, down, left and right direction waves respectively. The seismic wave field is taken as neural network input data, and the decomposed four direction waves are taken as network label data. The corresponding neural network parameters are trained at the shot point position and the receiver point position respectively. Two neural networks automatically separate the full wave field of the shot point and the receiver point into up, down, left and right direction waves respectively. A new cross-correlation imaging condition depending on the stratum dip angle is proposed to realize zero delay cross-correlation imaging of the shot point down-going wave field and the receiver point up-going wave field along the normal direction of the stratum. The application has small additional calculation amount and storage space, and can effectively eliminate low-frequency noise generated by the cross-correlation of the same direction waves of the shot point and the receiver point in theory, and significantly improves the precision of the reverse-time migration imaging profile.

Description

Technical Field

[0001] This invention relates to the field of exploration geophysics, and in particular to a high-precision reverse time migration imaging method based on intelligent seismic wavefield separation. Background Technology

[0002] Compared to traditional methods such as one-way wave and ray migration, reverse-time migration (RTM) uses a two-way wave equation to extend the wave field, avoiding approximations of the wave equation. This method can image gyratory waves, prismatic waves, and other waves without imposing any stratigraphic dip constraints, achieving accurate descriptions of complex geological structures. RTM based on the two-way wave equation employs zero-delay cross-correlation imaging conditions to image the forward propagation source wavefield and the backward propagation receiver wavefield at the same time. While this imaging condition is easy to implement, convenient for parallelization, and has no stability issues, it does not restrict the wave propagation direction during wavefield simulation. All waves along the imaging path satisfy the imaging conditions, and direct cross-correlation imaging generates strong low-frequency noise, especially noticeable in shallow regions, which severely affects the accuracy of RTM imaging. Therefore, there is an urgent need to develop a method to suppress shallow low-frequency noise in RTM imaging results to achieve clearer and more accurate imaging to meet the current demand for high-precision imaging of complex structures.

[0003] To suppress low-frequency background noise with strong shallow energy in reverse-time migration imaging profiles, common solutions generally fall into two categories. The first category involves using filtering techniques within the imaging profile to remove the influence of low-frequency noise. The second category involves first decomposing the entire wavefield into traveling waves in different directions, and then suppressing low-frequency noise based on new imaging conditions. However, there is overlap between the effective signal and low-frequency noise, and the filter may lose some effective signal while suppressing low-frequency noise, reducing the accuracy of the imaging results. In addition, post-imaging low-frequency noise suppression also suffers from problems such as transverse amplitude non-preservation, phase distortion, and strong dependence on the incident angle. In the second category of methods, most wavefield decomposition methods are based on the Hilbert transform. However, this explicit wavefield decomposition method requires simultaneous calculation and storage of the real and imaginary parts of the seismic traveling waves, resulting in high computational costs, which significantly limits the widespread application of Hilbert transform-based wavefield separation methods in practical production. Summary of the Invention

[0004] To address the technical problem of generating a large amount of low-frequency, high-amplitude noise from the cross-correlation of traveling waves in the same propagation direction, which makes it difficult to achieve high-precision reverse-time migration imaging in complex strata, this invention discloses a high-precision reverse-time migration imaging method based on intelligent seismic wavefield separation. This method requires less additional computation and storage space, and theoretically can effectively eliminate the low-frequency noise generated by the cross-correlation of traveling waves in the same direction between the shot point and the receiver point, thereby improving the accuracy of the reverse-time migration imaging results while ensuring computational efficiency.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a high-precision reverse-time migration imaging method based on intelligent seismic wavefield separation, specifically including the following steps:

[0006] Step (1), obtain input data;

[0007] Step (2): Calculate the wave field at the shot point location using acoustic finite difference forward modeling technique;

[0008] Step (3): Calculate the wavefield at the receiver location using the recorded forward seismic records and the migration velocity model through back extrapolation.

[0009] Step (4): Based on the Hilbert transform, the wave fields at the shot point and receiver point are decomposed into upper, lower, left, and right traveling waves, respectively.

[0010] Step (5): For the problem of traveling wave decomposition in the direction of seismic wavefield, specific convolutional neural network frameworks are constructed for the wavefields of the shot point and receiver point respectively.

[0011] Step (6) uses the seismic wave field as the input to the neural network and the upper, lower, left, and right traveling waves decomposed in step (4) as the output of the neural network. The internal parameters of the two convolutional neural networks constructed in step (5) are iteratively trained.

[0012] Step (7) Apply the neural network trained in step (6) to the test dataset of the shot point and receiver point to separate the traveling waves in the four directions of up, down, left and right at the positions of the shot point and receiver point.

[0013] Step (8) yields a seismic dip-dependent cross-correlation imaging condition, enabling cross-correlation imaging of shot point downflow and receiver upflow along the stratum normal direction.

[0014] Furthermore, the input data includes the true velocity model, the migration velocity model, and the seismic wavelet.

[0015] Furthermore, the cross-correlation imaging conditions in step (8) are as follows:

[0016]

[0017] In the formula, I(x) represents the reverse time migration imaging result, S represents the wavefield at the shot point, R represents the wavefield at the receiver point, the subscripts u, d, l and r represent the upper, lower, left and right traveling wave directions, θ represents the dip angle of the subsurface reflecting layer, x represents the position of the imaging point, and t represents the position of the receiver point. max For recording time, β1 and β2 are angle factors related to the dip angle of the formation;

[0018] β1 and β2 are represented as follows:

[0019]

[0020]

[0021] This invention also discloses the application of a high-precision reverse time migration imaging method based on intelligent seismic wavefield separation in wavefield separation and reverse time migration imaging of the Sigsbee model.

[0022] The beneficial effects of this invention are as follows: First, seismic forward simulation and backward propagation are used to calculate the wavefields at the shot point and receiver. Then, the Hilbert transform method is used to decompose the wavefields at both locations into upper, lower, left, and right components. The entire wavefield is used as input to a neural network, and the four directional components are used as labeled data to train the neural network structure for both the shot point and receiver. Finally, a new cross-correlation imaging condition related to the formation reflection dip angle is introduced, combining the directional traveling waves predicted by the network. By achieving cross-correlation between the downward wavefield at the shot point and the upward wavefield at the receiver along the normal direction of the dipping strata, low-frequency noise is removed, improving the imaging quality of complex structures.

[0023] This invention only includes the network training and prediction stages. Although training the neural network at the shot point and receiver point takes several hours, once the neural network identifies the kinematic feature differences of traveling waves in different directions from the whole wave field during the training process, it only takes a few seconds to decompose the whole wave field in the new dataset into traveling waves in four directions, which improves the accuracy of the reverse time migration imaging results while ensuring computational efficiency. Attached Figure Description

[0024] Figure 1 This is a diagram of the Sigsbee real velocity model according to an embodiment of the present invention;

[0025] Figure 2 This is a diagram of the Sigsbee offset velocity model according to an embodiment of the present invention;

[0026] Figure 3 This is a cross-sectional view of the reverse-time migration imaging result when the true velocity model is used during the migration process in an embodiment of the present invention;

[0027] Figure 4 This is a diagram showing the shot point wavefield separation based on a neural network when using a real velocity model during the offset process in an embodiment of the present invention.

[0028] Figure 5 This is a diagram showing the wavefield separation at the detector point based on a neural network when using a real velocity model during the migration process, according to an embodiment of the present invention.

[0029] Figure 6 This is a cross-sectional view of the reverse time migration imaging results based on the intelligent wavefield separation method and new cross-correlation imaging conditions when using the real velocity model during the migration process in an embodiment of the present invention.

[0030] Figure 7 This is a cross-sectional view of the reverse-time migration imaging result when the true velocity model is used during the migration process in an embodiment of the present invention;

[0031] Figure 8 This is a diagram showing the shot point wavefield separation based on a neural network when using a real velocity model during the offset process in an embodiment of the present invention.

[0032] Figure 9 This is a cross-sectional view of the reverse time migration imaging results based on the intelligent wavefield separation method and new cross-correlation imaging conditions when using the true velocity model during the migration process in an embodiment of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to represent selected embodiments of the invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] The cross-correlation of traveling waves in the same propagation direction generates a large amount of low-frequency, high-amplitude noise, making it difficult to achieve high-precision reverse-time migration imaging of complex strata. To address the high computational cost of traditional Hilbert-based wavefield separation methods, this invention discloses a high-precision reverse-time migration imaging method and its application based on intelligent seismic wavefield separation. Specifically, it includes the following steps:

[0035] (1) Obtain input data, which includes the true velocity model, the smoothed migration velocity model, and the seismic wavelet;

[0036] (2) The wave field at the shot point location was calculated using the finite difference forward modeling technique for acoustic waves.

[0037] (3) The wavefield at the receiver location is calculated by back extrapolation using the recorded forward seismic records and the migration velocity model;

[0038] (4) Based on the Hilbert transform, the wave fields of the shot point and receiver point are decomposed into upper, lower, left and right traveling waves, respectively;

[0039] (5) For the problem of seismic wave field direction traveling wave decomposition, specific convolutional neural network frameworks are constructed for the wave fields of shot point and receiver point respectively.

[0040] (6) The seismic wave field is used as the input of the neural network, and the upper, lower, left and right traveling waves decomposed in (4) are used as the output of the neural network. The internal parameters of the two convolutional neural networks constructed in (5) are iteratively trained.

[0041] (7) The neural network trained in (6) is applied to the test dataset of the shot point and the receiver point to separate the traveling waves in the four directions of up, down, left and right of the shot point and receiver point.

[0042] (8) A cross-correlation imaging condition dependent on seismic dip angle is obtained, and cross-correlation imaging of shot point downflow and receiver upflow along the stratum normal direction is realized.

[0043] The conditions for cross-correlation imaging are:

[0044]

[0045] In the formula, I(x) represents the reverse time migration imaging result, S represents the wavefield at the shot point, R represents the wavefield at the receiver point, the subscripts u, d, l and r represent the upper, lower, left and right traveling wave directions, θ represents the dip angle of the subsurface reflecting layer, x represents the position of the imaging point, and t represents the position of the receiver point. max For recording time, β1 and β2 are angle factors related to the dip angle of the formation;

[0046] β1 and β2 are represented as follows:

[0047]

[0048]

[0049] This invention also discloses the application of a high-precision reverse time migration imaging method based on intelligent seismic wavefield separation in wavefield separation and reverse time migration imaging of the Sigsbee model.

[0050] This method was applied to wavefield separation and reverse time migration imaging of the Sigsbee model, and high-precision imaging results were obtained.

[0051] Specifically, Figures 1-3 As can be seen, there is significant low-frequency noise in the imaging profile, which significantly reduces the imaging quality.

[0052] Figure 4 It can be seen that despite the complexity of wave field types and the presence of many overlapping wave fields, a well-trained neural network can still distinguish traveling waves propagating in different directions within a complex shot point wave field and produce accurate decomposition results.

[0053] Figure 5It can be seen that the neural network successfully decomposed the wave field at the detector point, and the predicted directional traveling wave has the same kinematic properties as the traveling wave decomposed by the Hilbert transform method.

[0054] Figure 6 It can be seen that the proposed method can effectively remove low-frequency noise and provide accurate offset results.

[0055] Figure 7 It can be seen that although the bottom of the salt body is well imaged, the shallow structure is severely masked by high-amplitude low-frequency noise.

[0056] Figure 8 It can be seen that the directional traveling wave predicted by the neural network is kinematically consistent with the directional traveling wave separated by the Hilbert transform method.

[0057] Figure 9 As can be seen, this invention removes low-frequency noise that severely reduces the accuracy of imaging results and produces a better offset imaging profile than the traditional reverse time migration method at the steeply tilted salt wing structure.

[0058] This invention utilizes seismic forward modeling and backward extrapolation to calculate the seismic wavefields at the shot and receiver locations. Then, based on the traditional Hilbert transform, the shot and receiver wavefields are decomposed into ascending, descending, left, and right traveling waves, respectively. The seismic wavefields are then used as input data for a neural network, and the four directional traveling waves are used as network label data. Corresponding neural network parameters are trained at the shot and receiver locations, respectively. Based on the differences in kinematic characteristics such as travel time and wavefront curvature of the traveling waves in different directions, the two neural networks automatically separate the entire wavefields at the shot and receiver locations into ascending, descending, left, and right traveling waves. Finally, this invention proposes a new cross-correlation imaging condition dependent on the formation dip angle, achieving zero-delay cross-correlation imaging of the descending wavefield at the shot and the ascending wavefield at the receiver along the formation normal direction.

[0059] Compared with traditional wavefield separation methods, this invention requires less additional computation and storage space. Theoretically, it can effectively eliminate low-frequency noise generated by the cross-correlation of traveling waves in the same direction between the shot point and the receiver point, and significantly improve the accuracy of the reverse time migration imaging profile. It has important significance and application value in actual seismic data imaging and other processing.

[0060] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A high-precision reverse-time migration imaging method based on intelligent seismic wavefield separation, characterized in that, Specifically, the following steps are included: Step (1), obtain input data; Step (2): Calculate the wave field at the shot point location using acoustic finite difference forward modeling technique; Step (3): Calculate the wavefield at the receiver location using the recorded forward seismic records and the migration velocity model through back extrapolation. Step (4): Based on the Hilbert transform, the wave fields at the shot point and receiver point are decomposed into upper, lower, left, and right traveling waves, respectively. Step (5): For the problem of traveling wave decomposition in the direction of seismic wavefield, specific convolutional neural network frameworks are constructed for the wavefields of the shot point and receiver point respectively; Step (6): The seismic wave field is used as the input of the neural network, and the upper, lower, left and right traveling waves decomposed in step (4) are used as the output of the neural network. The internal parameters of the two convolutional neural networks constructed in step (5) are iteratively trained. Step (7): Apply the neural network trained in step (6) to the test dataset of the shot point and receiver point to separate the traveling waves in the four directions of up, down, left and right at the positions of the shot point and receiver point; Step (8) yields a cross-correlation imaging condition dependent on seismic dip angle, enabling cross-correlation imaging of shot point downflow and receiver upflow along the stratum normal direction; Input data includes the true velocity model, the migration velocity model, and the seismic wavelet; The cross-correlation imaging conditions in step (8) are as follows: ; In the formula, This is the result of reverse time migration imaging. For the wave field of the shot point, The wave field at the receiver point is represented by the subscripts u, d, l, and r, which represent the upward, downward, leftward, and rightward traveling wave directions, respectively. The dip angle of the underground reflector layer. The location of the imaging point. To record time, and An angle factor related to the dip angle of the strata; and They are represented as follows: ; 。 2. The application of the high-precision reverse time migration imaging method based on intelligent seismic wavefield separation as described in claim 1 in wavefield separation and reverse time migration imaging of the Sigsbee model.

Images