Seismic data compensation method and compensation system

Abstract

The invention discloses a seismic data compensation method and a seismic data compensation system. The seismic data compensation method comprises the following steps: acquiring seismic data based on a basic convolution model; adding an attenuation operator on the basis of the basic convolution model to form an attenuation convolution expression; synthesizing an attenuation seismic record based on the attenuation convolution expression; based on a preset neural network model, evaluating a difference condition between the observation attenuation seismic record and the synthesis attenuation seismic record, and based on the difference condition, obtaining a reflection coefficient of the seismic; and carrying out convolution on the reflection coefficient and an unattenuated seismic wavelet in the seismic data to construct and form a seismic attenuation compensation expression, and outputting an attenuation-compensated seismic record according to the seismic attenuation compensation expression. According to the technical scheme, the attenuation effect can be effectively compensated, the stratum interface is clearer, the recognition precision of the underground structure form and the prediction accuracy of reservoir distribution are improved, and smooth proceeding of oil-gas exploration work is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of earthquake data processing technology, and more specifically to earthquake data compensation methods. Background Technology

[0002] In oil and gas exploration, accurate capture of seismic wave propagation characteristics is a prerequisite for achieving detailed imaging of underground geological structures and accurate prediction of reservoir features. However, the rocks in actual underground media generally exhibit non-ideal elastic properties due to factors such as their composition, pore structure, and contained fluids. This inevitably leads to continuous inelastic absorption and attenuation effects on seismic waves as they propagate from the source underground and are reflected back to the surface receiving equipment. Specifically, this attenuation effect manifests as follows: the amplitude energy of seismic waves decreases significantly with propagation distance and depth, resulting in a substantial reduction in the intensity of reflected signals from deep strata, sometimes making them indistinguishable from noise; simultaneously, high-frequency components are rapidly lost due to the strong absorption characteristics of the medium, compressing the frequency bandwidth of the seismic record and narrowing the frequency band, directly causing a decrease in the vertical resolution of the seismic data; and the superposition of amplitude attenuation and high-frequency loss further blurs the reflection characteristics of different strata interfaces, significantly increasing the difficulty of identifying these interfaces. Previously clearly discernible strata boundaries become blurred, severely reducing the accuracy of identifying underground structural morphology and predicting reservoir distribution, posing a significant obstacle to subsequent oil and gas exploration. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a seismic data compensation method that can effectively compensate for attenuation effects, making stratigraphic interfaces clearer, improving the accuracy of identifying underground structural morphology and predicting reservoir distribution, and ensuring the smooth progress of oil and gas exploration work.

[0004] This application provides a method for compensating seismic data, the method comprising:

[0005] Seismic data is acquired based on a fundamental convolutional model;

[0006] Based on the aforementioned basic convolution model, a decay operator is added to form a decay convolution expression;

[0007] Attenuated seismic records are synthesized based on the aforementioned attenuation convolution expression;

[0008] The difference between the observed attenuated seismic record and the synthetic attenuated seismic record is evaluated based on a pre-set neural network model, and the reflection coefficient of the earthquake is obtained based on the difference.

[0009] The reflection coefficient is convolved with the unattenuated seismic wavelet in the seismic data to construct a seismic attenuation compensation expression, and the attenuated compensated seismic record is output according to the seismic attenuation compensation expression.

[0010] In one aspect, the step of adding a decay operator to the basic convolution model to form a decay convolution expression includes:

[0011] The expression for the unattenuated seismic wave is obtained based on the aforementioned basic convolution model;

[0012] Convert the seismic wave expression into a frequency domain expression form;

[0013] Add an attenuation operator to the frequency domain expression to transform it into an attenuation convolution expression.

[0014] In one aspect, the expression for an unattenuated seismic wave is s(t) = W(t) * r(t), where s(t) is the seismic record, W(t) is the seismic wavelet matrix, r(t) represents the reflection coefficient, * represents the convolution operation, and t is the total propagation and reflection time of the seismic wave.

[0015] The frequency domain expression is: s(ω) is the frequency domain representation of the seismic record amplitude, W(ω) is the frequency domain representation of the seismic wavelet matrix, A(ω,t) is the frequency domain representation of the attenuation operator, and ω is the frequency;

[0016] The decay convolution expression is: W a (t) Attenuated seismic wavelet, W a =W×A, where W is the original seismic wavelet and A is the attenuation operator.

[0017] In one aspect, the frequency domain expression of the attenuation operator is: Where γ = (1 / πQ), γ is the frequency dependence exponent. e is the amplitude attenuation factor. iωt The phase rotation factor, For frequency-dependent compensation factors, ω h Here, Q is the reference angular frequency, and Q is the quality factor.

[0018] In one aspect, the difference includes residuals;

[0019] The steps of evaluating the differences between observed attenuated seismic records and the synthetic attenuated seismic records based on a pre-set neural network model include:

[0020] The initial reflection coefficients are output based on a pre-set neural network model;

[0021] Attenuated seismic records are synthesized by convolution based on the initial reflection coefficient and the attenuated wavelet in the attenuation convolution expression;

[0022] Determine the residual between the observed attenuated seismic record and the synthesized attenuated record, and minimize the residual to determine the reflection coefficient.

[0023] In one aspect, the residual is represented as a loss function;

[0024] The steps to minimize the residuals to determine the reflection coefficient include:

[0025] The loss function between the observed attenuated seismic record and the synthetic attenuated seismic record is calculated, and the neural network parameters are updated via backpropagation to update the reflection coefficient.

[0026] In one aspect, the loss function is J(r). The actual observed attenuation records satisfy the following:

[0027]

[0028] In one aspect, the step of backpropagating to update the neural network parameters to update the reflection coefficients includes:

[0029] The loss function is differentiated according to a pre-defined chain rule, and the network parameters of the neural network are updated accordingly.

[0030] Based on the updated network parameters, a nonlinear mapping between the observed attenuated seismic records and the reflection coefficients is established, so as to predict and update the reflection coefficients based on the neural network.

[0031] The derivative of the loss function satisfies the following:

[0032] w new Updated neural network parameters, w old The neural network parameters before the update, where 'a' is the learning rate.

[0033] In one aspect, the seismic attenuation compensation expression is: Updated reflectance coefficient;

[0034] Net() represents a neural network, and w represents the parameters of the neural network.

[0035] Furthermore, to address the aforementioned problems, this application also provides a seismic data compensation system, the compensation system comprising:

[0036] The acquisition module is used to acquire seismic data based on the basic convolutional model;

[0037] Add a module to add a decay operator based on the basic convolution model to form a decay convolution expression;

[0038] The synthesis module is used to synthesize attenuated seismic records based on the attenuation convolution expression;

[0039] The evaluation module is used to evaluate the differences between the observed attenuated seismic records and the synthetic attenuated seismic records based on a pre-set neural network model, and to obtain the reflection coefficient of the earthquake based on the differences.

[0040] The compensation module is used to convolve the reflection coefficient with the unattenuated seismic wavelet in the seismic data to construct a seismic attenuation compensation expression, and output attenuation-compensated seismic records based on the seismic attenuation compensation expression.

[0041] The beneficial effects of this invention are as follows: After acquiring seismic data based on the basic convolution model, an attenuation convolution expression is formed by adding an attenuation operator, thereby synthesizing attenuated seismic records. Then, a pre-set neural network model is used to evaluate the difference between the observed attenuated seismic records and the synthesized attenuated seismic records to obtain reflection coefficients. Finally, these reflection coefficients are convolved with unattenuated seismic wavelets to construct a compensation expression, outputting the attenuated-compensated seismic records. By introducing attenuation operators to accurately characterize effects such as amplitude energy attenuation, high-frequency component loss, and bandwidth narrowing, the reflection coefficients obtained based on the differences are closer to the actual stratigraphic characteristics. Combined with the compensation operation of the unattenuated wavelet, the attenuated energy and high-frequency information can be effectively recovered, improving the resolution of the seismic records, clearly displaying the reflection characteristics of stratigraphic interfaces, and avoiding the submersion of thin reservoir signals. This improves the accuracy of subsurface structure identification and reservoir prediction, providing reliable data support for key aspects of subsequent oil and gas exploration and ensuring the smooth progress of oil and gas exploration work. Attached Figure Description

[0042] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0043] Figure 1 This is a schematic diagram illustrating the steps of the earthquake data compensation method in this application;

[0044] Figure 2 This is a schematic diagram of the process steps for forming the attenuation convolution expression in the earthquake data compensation method of this application;

[0045] Figure 3 This is a schematic diagram of the process steps for determining the reflection coefficient in the earthquake data compensation method of this application;

[0046] Figure 4This is a schematic diagram of the process steps for updating the reflection coefficient in the earthquake data compensation method of this application;

[0047] Figure 5 This is a schematic diagram of the basic data for earthquake simulation in this application;

[0048] Figure 6 This is a schematic diagram of the compensation results under different quality factors in this application;

[0049] Figure 7 This is a schematic diagram of the compensation results for earthquake data compensation methods in this application with a quality factor of 10 and a signal-to-noise ratio of 25.

[0050] Figure 8 This is a schematic diagram of the compensation results for earthquake data compensation methods in this application, where the quality factor is 10 and the signal-to-noise ratio is 15.

[0051] Figure 9 This is a schematic diagram of the compensation results for earthquake data compensation methods in this application with a quality factor of 10 and a signal-to-noise ratio of 5.

[0052] Figure 10 This is a schematic diagram showing the compensation results in the time domain and frequency domain dimensions compared in the earthquake data compensation method of this application;

[0053] Figure 11 This is a schematic diagram of the compensation results under a complex geological model in the earthquake data compensation method of this application;

[0054] Figure 12 This is a schematic diagram of the compensation results under a complex geological model with a quality factor of 10 and a signal-to-noise ratio of 15 in the earthquake data compensation method of this application.

[0055] Figure 13 This is a schematic diagram of the frequency domain compensation results under a complex geological model in the earthquake data compensation method of this application;

[0056] Figure 14 This is a schematic diagram of the neural network model architecture in the earthquake data compensation method of this application;

[0057] Figure 15 This is a schematic diagram of the functional structure of the earthquake data compensation system in this application. Detailed Implementation

[0058] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of the present invention and are therefore intended to limit the scope of protection of the present invention.

[0059] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application should have the ordinary meaning as understood by those skilled in the art to which this invention pertains.

[0060] like Figure 1 As shown, this application provides a method for compensating seismic data, the method including:

[0061] Step S10: Acquire seismic data based on the fundamental convolution model. The fundamental convolution model is a classic theoretical framework for seismic data processing. Its core logic is that seismic records are generated by convolution operations between seismic wavelets and subsurface reflection coefficients, reflecting the signal characteristics of seismic waves after propagation underground and reflection by the strata. Acquiring initial seismic data ensures that the data source conforms to the basic physical laws of seismic wave propagation, providing reliable raw input for subsequent attenuation compensation and avoiding compensation errors caused by deviations of the data basis from the physical mechanism.

[0062] Step S20 involves adding an attenuation operator to the basic convolution model to form an attenuation convolution expression. Due to the non-ideal elastic properties of the subsurface medium, seismic waves experience energy attenuation and high-frequency loss during propagation, effects not considered in the basic convolution model. By introducing an attenuation operator into the basic convolution model, the inelastic absorption characteristics of the formation are quantified and integrated into the model, resulting in the attenuation convolution expression. This expression accurately describes the attenuation pattern of seismic waves propagating in real subsurface media, making the model more closely reflect actual geological conditions.

[0063] Step S30: Synthesize attenuated seismic records based on the attenuation convolution expression. The attenuation convolution expression already includes the attenuation characteristics of seismic wave propagation. Calculations are performed using the attenuation convolution expression to generate theoretically attenuated seismic records. The synthesized attenuated seismic records simulate the energy attenuation and frequency band compression characteristics of seismic waves due to stratum absorption in actual observations. They can serve as a reference standard for comparison with actual observed attenuated seismic records, providing a basis for subsequent evaluation of differences and retrieval of reflection coefficients.

[0064] Step S40 involves evaluating the differences between observed attenuated seismic records and synthetic attenuated seismic records based on a pre-set neural network model, and obtaining the seismic reflection coefficient based on these differences. Observed attenuated seismic records are actually acquired signals with attenuation effects, while synthetic attenuated seismic records are reference signals generated based on theoretical models. By comparing the differences between the two, the reflection coefficient of the subsurface strata is derived in reverse. The reflection coefficient directly reflects the differences in the physical properties of the strata interface, such as changes in density and velocity, and is a key parameter for characterizing subsurface structures.

[0065] Step S50 involves convolving the reflection coefficient with the unattenuated seismic wavelet in the seismic data to construct a seismic attenuation compensation expression. Based on this expression, an attenuated seismic record is output. Step S40 then yields a more accurate representation of the reflection characteristics of the subsurface strata. The unattenuated seismic wavelet retains its original energy and high-frequency components, thus recovering the energy and high-frequency information lost due to strata attenuation. This compensates for the resolution reduction and interface blurring issues caused by attenuation in the original seismic data, resulting in a more clearly reflected subsurface structure and reservoir characteristics in the final attenuated seismic record.

[0066] Here, convolution can be understood as a type of convolution, and the convolution model can be understood as a type of convolution model; the basic convolution model can be understood as the classic convolution model, which was proposed by Enders Robinson in 1967.

[0067] In this embodiment, after acquiring seismic data based on the basic convolution model, an attenuation convolution expression is formed by adding an attenuation operator, thereby synthesizing attenuated seismic records. Then, a pre-set neural network model is used to evaluate the difference between the observed attenuated seismic records and the synthesized attenuated seismic records to obtain reflection coefficients. Finally, these reflection coefficients are convolved with unattenuated seismic wavelets to construct a compensation expression, outputting the attenuated-compensated seismic records. By introducing attenuation operators to accurately characterize effects such as amplitude energy attenuation, high-frequency component loss, and bandwidth narrowing, the reflection coefficients obtained based on these differences more closely approximate the actual stratigraphic characteristics. Combined with the compensation operation of the unattenuated wavelet, the attenuated energy and high-frequency information can be effectively recovered, improving the resolution of the seismic records, clearly displaying the reflection characteristics of stratigraphic interfaces, and preventing the submersion of thin reservoir signals. This improves the accuracy of subsurface structure identification and reservoir prediction, providing reliable data support for key aspects of subsequent oil and gas exploration and ensuring the smooth progress of oil and gas exploration work.

[0068] like Figure 2 As shown, the steps to form a decaying convolution expression by adding a decay operator to the basic convolution model include:

[0069] Step S210: Obtain the expression for the unattenuated seismic wave based on the basic convolution model. The basic convolution model is a classic framework for seismic data processing. Obtaining the expression for the unattenuated seismic wave based on the basic convolution model means that the seismic record is generated by the convolution operation of the seismic wavelet matrix and the reflection coefficient. The seismic wave expression reflects the ideal state of seismic wave propagation and reflection when it is not affected by the inelastic attenuation of the strata.

[0070] Step S220: The seismic wave expression is converted into a frequency domain expression. To facilitate the analysis of the frequency characteristics of the seismic wave and the subsequent introduction of attenuation operators, the unattenuated seismic wave expression is converted from the time domain to the frequency domain. Through this conversion, the convolution operation in the time domain is transformed into a multiplication operation in the frequency domain, providing mathematical convenience for accurately characterizing the frequency dependence of the attenuation effect.

[0071] Step S230: Add an attenuation operator to the frequency domain expression to form an attenuation-convolution expression. Due to the inelastic properties of the actual subsurface medium, seismic wave propagation involves energy attenuation and phase distortion. An attenuation operator is introduced into the frequency domain expression to form an attenuation-convolution frequency domain expression. The frequency domain expression of the attenuation operator includes parameters such as amplitude attenuation factor and phase rotation factor, used to quantify the attenuation effect of the formation on seismic waves. Then, an inverse Fourier transform is used to convert it back to the time domain to obtain the attenuation-convolution expression, which can accurately describe the attenuation characteristics of seismic waves propagating in real strata.

[0072] Furthermore, the expression for an unattenuated seismic wave is s(t) = W(t) * r(t), where s(t) is the seismic record, W(t) is the seismic wavelet matrix, r(t) represents the reflection coefficient, * denotes the convolution operation, and t is the total propagation and reflection time of the seismic wave. Seismic waves originate from a seismic source, such as an artificially generated or natural source, and propagate underground. When they encounter interfaces between different geological strata, some of the seismic waves are reflected back to the surface and captured by the receiving device. The time elapsed from the moment the seismic wave is generated at the source until the reflected wave is received is the total propagation and reflection time t of the seismic wave.

[0073] The frequency domain expression is s(ω) is the frequency domain representation of the seismic record amplitude, W(ω) is the frequency domain representation of the seismic wavelet matrix, A(ω,t) is the frequency domain representation of the attenuation operator, and ω is the frequency; e -iωt It represents the Fourier transform, which realizes the mathematical transformation from the time domain to the frequency domain. Through the frequency domain transformation, the formation attenuation is embedded into the model in the form of an operator.

[0074] The expression for decaying convolution is: This is the actual model after introducing attenuation. W a (t) Attenuated seismic wavelet, W a = W × A, where W is the original seismic wavelet and A is the attenuation operator. By modifying the wavelet, W is transformed into W0. a By incorporating the formation attenuation effect into the basic convolution model, an expression for the seismic record that reflects the actual energy loss of the formation is obtained.

[0075] Furthermore, the frequency domain expression of the attenuation operator is: Wherein, γ=(1 / πQ), γ is the frequency dependence index, which is associated with the formation quality factor Q and controls the rate attenuation as a function of frequency. The amplitude attenuation factor (Q) decreases exponentially with increasing frequency and propagation time; the smaller the Q value, the faster the attenuation. iωt The phase rotation factor, For frequency-dependent compensation factors, ω h Using a reference angular frequency, a frequency-dependent compensation factor corrects for attenuation differences at different frequencies, making the model more closely resemble actual geological formations. Q is the quality factor, a crucial physical parameter in seismic exploration and related wave theory, used to quantitatively describe the energy loss characteristics of a medium during wave propagation. The quality factor Q is a dimensionless parameter reflecting the ratio of stored energy to dissipated energy within a vibrating system in the medium over one vibration cycle. In the context of seismic wave propagation, the quality factor measures the degree of energy attenuation of seismic waves by the subsurface medium. Specifically, a higher Q value means less energy loss by the medium, allowing the seismic wave to maintain relatively strong energy and a better waveform during propagation; conversely, a lower Q value indicates greater absorption and dissipation of seismic wave energy by the medium, resulting in faster energy attenuation, more severe loss of high-frequency components, and significant waveform distortion.

[0076] The relationship between seismic wave frequency, propagation time and stratum attenuation is described by attenuation operators. Attenuation operators simulate the energy attenuation and phase distortion of seismic waves during propagation due to the inelastic properties of the stratum.

[0077] like Figure 3 As shown, the differences include residuals, which are usually error functions, such as mean square error, which quantify the differences between the two in terms of amplitude, phase, and other characteristics.

[0078] The steps for evaluating the differences between observed attenuated seismic records and synthetic attenuated seismic records based on a pre-configured neural network model include:

[0079] Step S410: Output the initial reflection coefficient based on the pre-set neural network model; input the relevant seismic data into the neural network model, and the neural network model directly outputs the preliminary estimate of the reflection coefficient of the underground strata, i.e. the initial reflection coefficient, according to the mapping relationship, to provide basic data for subsequent simulated seismic records.

[0080] Step S420: Based on the initial reflection coefficients and the attenuated wavelet in the attenuation convolution expression, a convolutional synthesis of an attenuated seismic record is performed; the initial reflection coefficients obtained in step S410 are convolved with the known attenuated wavelet. The convolution operation simulates the process of seismic records being generated by the interaction between the wavelet and the stratum reflection coefficient when seismic waves propagate underground, and a simulated attenuated seismic record is synthesized through calculation.

[0081] Step S430: Determine the residual between the observed attenuated seismic record and the synthesized attenuated seismic record, and minimize the residual to determine the reflection coefficient. Obtain the attenuated seismic record obtained from actual observation, and then calculate the residual between the observed and synthesized attenuated seismic records. Subsequently, an optimization algorithm, such as gradient descent, is used to continuously adjust the value of the reflection coefficient with the goal of minimizing the residual. During the iterative optimization process, new attenuated seismic records are continuously synthesized and the residual is calculated until the residual reaches its minimum or meets the preset accuracy requirements. At this point, the determined reflection coefficient closely matches the actual geological conditions and accurately reflects the reflection characteristics of the underground strata.

[0082] In one embodiment of this application, the step of representing the residual as a loss function and minimizing the residual to determine the reflection coefficient includes:

[0083] Step S431: Calculate the loss function between the observed attenuated seismic record and the synthesized attenuated seismic record, and update the neural network parameters through backpropagation to update the reflection coefficient. First, calculate the loss function between the actually observed attenuated seismic record and the attenuated seismic record synthesized from the initial reflection coefficient to quantify the data difference. Then, use the backpropagation algorithm to update and adjust the neural network parameters using the loss function. Since the reflection coefficient is output by the neural network, the update of the network parameters will directly lead to the optimization of the reflection coefficient, ultimately achieving a precise update of the reflection coefficient, making the synthesized attenuated seismic record more closely match the observed attenuated seismic record.

[0084] In one embodiment of this application, the loss function is J(r). The actual observed attenuation records satisfy the following:

[0085] Minimizing the loss function can be understood as making the synthetic attenuation record infinitely close to the real record. The reflection coefficient at this point is the reflection coefficient that is closest to the real stratum.

[0086] In this process, by combining optimization algorithms such as gradient descent, the derivative of the loss function with respect to the reflection coefficient is calculated to guide the iterative direction of the reflection coefficient.

[0087] In one embodiment of this application, the step of backpropagating to update the neural network parameters to update the reflection coefficients includes:

[0088] Step S4311 involves calculating the derivative of the loss function according to a pre-defined chain rule to update the network parameters of the neural network. Specifically, first, the partial derivatives of the loss function with respect to the parameters of each layer of the neural network are calculated using the pre-determined chain rule to obtain the gradient direction for parameter updates. Then, using optimization strategies such as gradient descent, the calculated gradients are used to adjust the weights, biases, and other parameters of the neural network, making the reflection coefficients output by the neural network more closely match the actual geological conditions. The chain rule is applied to propagate the error of the loss function back to the parameters of each layer of the neural network, thereby calculating the gradient for parameter updates. Simply put, the chain rule transforms the error in the seismic record into an adjustment amount for the network parameters. Through iterative optimization of the parameters using gradient descent, the neural network becomes increasingly adept at predicting the reflection coefficients of the actual geological formations.

[0089] Step S4312: Based on the updated network parameters, a nonlinear mapping between observed attenuated seismic records and reflection coefficients is established to predict and update reflection coefficients based on neural networks. Based on the optimized network parameters, the nonlinear mapping relationship from observed attenuated seismic records to reflection coefficients is reconstructed. This mapping relationship can more accurately capture the complex correlation between seismic records and the reflection characteristics of subsurface strata. At this point, by processing the input observed attenuated seismic records using a neural network, the updated reflection coefficients can be directly predicted and output. Through this process, the reflection coefficients are iteratively optimized, continuously reducing the difference between the synthesized attenuated seismic records and the observed records, ultimately yielding reflection coefficient results that better reflect the actual geological conditions.

[0090] The derivative of the loss function satisfies the following:

[0091] w new Updated neural network parameters, w old The neural network parameters before the update, where 'a' is the learning rate. The learning rate controls the step size of the parameter updates, avoiding updates that are too large or too small.

[0092] In one embodiment of this application, the seismic attenuation compensation expression is: Updated reflectance coefficient;

[0093] Net() represents a neural network, and w represents the parameters of the neural network. It transforms geological problems in oil and gas exploration into data-driven optimization problems, using a neural network to learn the true reflection information in attenuation records and then inversely compensate for formation attenuation. First, the neural network is used to derive the true reflection coefficients from the attenuation records, and then convolution is used to generate compensated seismic records, making the seismic data closer to the true appearance of the formation.

[0094] Figure 5This is a set of basic data diagrams for earthquake simulation, showing the process from stratigraphic model to earthquake record and then to attenuation effect. From left to right and from top to bottom, it corresponds to the process of stratigraphic model construction, earthquake record generation, and attenuation simulation.

[0095] Specifically, Figure (a): Velocity Model, showing the P-wave velocity distribution of the strata; different colors indicate different velocities. The drawing shows three horizontal strata, with different colors representing strata with different velocities. Figure (b): Density Model, used in conjunction with the velocity model, together determine the reflection coefficient of the strata. Again, it shows three horizontal strata, with the density distribution corresponding to the velocity model. Together with the velocity model, the reflection coefficient is calculated. Figure (c): Seismic Data, a simulated seismic record without attenuation using the velocity and density models. The horizontal axis represents the trace number, the vertical axis represents time, and the color represents amplitude (warm colors for positive amplitude, cool colors for negative amplitude). The phase axis, i.e., continuous and clear color bands, reflects the horizontal reflection interface of the layered strata; without attenuation, there is no energy loss. Figure (d): Attenuated Seismic Data, a record simulating the attenuation effect of the strata based on the original seismic record. Amplitude decays over time, meaning the color of the deeper phase axis becomes lighter, indicating loss of high-frequency components. This simulates the attenuation effects of strata absorbing and scattering seismic wave energy. Figure (e): Seismic Data vs Attenuated. A single seismic record (e.g., Trace Number = 100) is extracted, comparing the differences between attenuated and non-attenuated records. One record can be understood as a complete seismic signal acquisition record. The horizontal axis represents time, and the vertical axis represents amplitude. Black represents the original record, without attenuation, with large and sharp amplitudes; blue represents the attenuated record, with small and gentle amplitudes, and the vibration center shifts relative to the original record. The amplitude of the attenuated record decreases over time, and the waveform widens, visually demonstrating the impact of attenuation on the seismic signal.

[0096] To comprehensively evaluate the attenuation compensation effect of the proposed method, three commonly used indicators, signal-to-noise ratio (SNR), structural similarity index (SSIM), and mean square error (MSE), were selected to quantitatively analyze the difference between the compensated data and the reference data.

[0097] Mean square error is used to measure the overall amplitude difference between the compensation result and the reference data, and its formula is expressed as:

[0098] in, Indicates reference earthquake data, For the compensation result, N = H × M is the total number of pixels in the image. MSE is used to measure the overall magnitude error between two images at the pixel level; the smaller the error value, the higher the compensation accuracy.

[0099] The signal-to-noise ratio formula is expressed as:

[0100] A higher SNR value indicates higher data quality in the compensation results.

[0101] The SSIM structural similarity index is used to measure the degree of similarity between two data points. The SSIM structural similarity formula is expressed as:

[0102]

[0103] in, and s re and s com The mean, and For the equation, This represents the covariance. c1 and c2 are stabilizing terms to prevent the denominator from being zero. The SSIM value ranges from [0, 1], and the closer it is to 1, the better the compensation effect.

[0104] See Figure 6 As shown in the figure, this set of figures shows the seismic attenuation compensation results under different Q values. By controlling the Q value of the variable, the generalization of the compensation method of this application is verified. The smaller the Q, the stronger the attenuation.

[0105] Figures (A), (B), and (C) represent attenuation records for different Q values. Figure (A): Q = 100 (weak attenuation), the phase axis of the attenuation record is relatively clear, the color is deep, and the continuity is good. Figure (B): Q = 50 (medium attenuation), the ambiguity of the phase axis increases, the color becomes lighter, and it is affected by attenuation. Figure (C): Q = 10 (strong attenuation), the phase axis is extremely blurry, the color is very pale, and the energy loss is severe. Figures (D), (E), and (F) represent the compensation results of the compensation method of this application, and show quantified indicators. It can be seen that regardless of whether Q = 100, Q = 50, or Q = 10, the compensation method of this application can effectively compensate.

[0106] See Figure 7As shown, with quality factor Q=10 and SNR=25, the differences between the original data and the attenuated data and the compensation results of different algorithms are visualized through 6 sub-figures to verify the compensation effect of the compensation method of this application. Among them, Figure (1): Reference (no attenuation reference record). Figure (2): Attenuated (attenuated seismic record), the observation record after stratum attenuation. The phase axis is blurred, the color becomes lighter, the continuity is poor, the amplitude decreases with time, the high frequency component is lost, and the phase axis becomes wider. Figure (3): Traditional (compensation result of traditional algorithm), the phase axis is clearer than Figure (2), but compared with Figure (a), it is still blurred, such as the color is not bright enough and the continuity is insufficient, indicating that the compensation effect of traditional algorithm is limited. Figure (4): Reference-Traditional (error of traditional algorithm), it can be seen that there are many color bands and large residuals, especially in the deep and edge areas, indicating that the traditional algorithm has not fully compensated for the attenuation and there are many residual errors. Figure (5): Ours (compensation method of this application). It can be seen that the phase axis is clear and sharp, and the color is close to that of Figure (1) (without attenuation reference), indicating that the compensation algorithm of this application has better effect and the energy and waveform are closer to the ideal state. Figure (6): Reference-Ours (error of this application). The residuals of Figure (1) and Figure (5) are compared. It can be seen that there are fewer color bands, the residuals are small, and most areas are close to 0, that is, colorless, indicating that the compensation method of this application is highly consistent with the ideal record and the error is significantly smaller than that of the traditional algorithm. By visualizing the residuals, the compensation accuracy of different algorithms is quantified. It can be seen that the compensation method of this application is closer to the ideal record, the residuals are smaller, the compensation is more thorough, and it is closer to the real reflection of the formation.

[0107] See Figure 8The comparative experimental results of the seismic attenuation compensation algorithm are shown in the scenario of SNR=15 and Q=10. Among them, Figure (7) Reference (no attenuation reference), the same as the previous figure (1), is the ideal no-attenuation seismic record (Ground Truth). Figure (8) Attenuated, simulating the observation record under more severe conditions Q=10, SNR=15: low signal-to-noise ratio and more noise. The phase axis is extremely blurry, the color is very light, and it is greatly affected by noise, resulting in more severe energy attenuation. Figure (9) Traditional (traditional algorithm compensation), the phase axis is partially recovered, but the noise is extremely high, the color is messy, and a large amount of error remains after compensation. Figure (10) Reference-Traditional (traditional algorithm residual), the residual between the traditional algorithm result and the reference record. It can be seen that the color bands are dense and messy, indicating that the traditional algorithm has a very poor compensation effect under severe conditions and cannot distinguish between noise and attenuation. Figure (11) Ours (compensation method of this application), it can be seen that the phase axis is clearly recovered, close to the reference record, and the noise is effectively suppressed. Figure (12) Reference-Ours (Residual of the Compensation Method of this Application) shows that there are very few color bars, indicating that the compensation method of this application can still accurately compensate for attenuation and suppress noise under harsh conditions, and is close to the ideal record.

[0108] Figure 9 As shown, the seismic attenuation compensation comparison results under Q=10 and SNR=5 are verified by six sub-figures to verify the robustness of the algorithm. Among them, Figure (13) Reference (no attenuation reference). Figure (14) Attenuated, simulated Q=10, SNR=5, extremely low signal-to-noise ratio, noise completely drowns out the weak signal observation record. The phase axis almost disappears, the color is very light, severely interfered with by noise, and the stratum reflection is covered by noise. Figure (15) Traditional (compensation result of traditional algorithm), it can be seen that the phase axis cannot be effectively recovered, the color is messy, noise and signal are mixed, and the record after compensation has almost no usable information. Figure (16) Reference-Traditional (residual of traditional algorithm), it can be seen that the color bands are dense and the whole picture is distorted, indicating that the traditional algorithm completely fails under extreme conditions, cannot distinguish noise and attenuation, and the compensation logic collapses. Figure (17) Ours (compensation method of this application), the phase axis is clearly recovered, close to the reference record, the noise is effectively suppressed, the color bands are sharp, and the stratum reflection is reproduced. Figure (18) Reference-Ours (residuals of the compensation method of this application) shows that there are very few color bars, indicating that the compensation method of this application can still accurately compensate for attenuation and suppress noise under extreme conditions, and is close to the ideal attenuation-free record.

[0109] See Figure 10As shown, this set of figures compares the attenuation compensation effects of traditional algorithms and the compensation method of this application under different signal-to-noise ratio (SNR) scenarios from two dimensions: time-domain waveform and frequency-domain spectrum. Figures A1, C1, and E1 show the time-domain waveform comparison, with the horizontal axis representing time, reflecting the travel time of seismic wave propagation, and the vertical axis representing amplitude, reflecting the energy intensity of the seismic wave. Reference (black) represents the reference record without attenuation. SNR is shown in blue, representing attenuation records at different SNRs. Traditional (yellow) represents the waveform after compensation using the traditional algorithm. Ours (red) represents the waveform using the compensation method of this application. It can be seen that the smaller the SNR, the greater the difference between the attenuation record and the reference waveform. After compensation using the traditional algorithm, the waveform is still affected by noise; after compensation using the method of this application, the waveform is closer to the reference record.

[0110] Additionally, Figures B1, D1, and F1 show a comparison of frequency domain spectra. The horizontal axis represents frequency, reflecting the frequency components of seismic waves. The vertical axis represents amplitude, reflecting the energy intensity of each frequency component. The curves have the same meaning as the previous figure, comparing the recovery of spectral energy after compensation by different algorithms. The smaller the SNR, the more severe the high-frequency energy loss; the curve dips in the high-frequency band, indicating loss of stratigraphic details. After compensation by the traditional algorithm (yellow), the spectral energy recovery is limited, especially in the high-frequency band, still lower than the reference record, indicating that the traditional method is difficult to compensate for high-frequency attenuation and is easily affected by noise interference. After compensation by the method of this application (red), the spectral energy is closer to the reference record, and the high-frequency band is well recovered, indicating that the compensation method of this application can accurately compensate for attenuation across the entire frequency band, including high-frequency and low SNR scenarios where the traditional algorithm fails. Figures A1 and B1 show that at SNR=25, the attenuation record differs little from the reference; the traditional algorithm can partially compensate, but the waveform and spectrum of the compensation method of this application are closer to the reference. Figures C1 and D1 show at SNR=5, where the attenuation record is heavily noiseed, and traditional algorithms still result in noise interference after compensation; the compensation method of this application clearly restores the waveform, and the spectral energy is close to the reference. Figures E1 and F1 show at SNR=1, where the attenuation record is completely submerged by noise, and traditional algorithms result in no usable waveform after compensation; the compensation method of this application can still restore the waveform outline, and the spectral energy is significantly improved.

[0111] See Figure 11The figures show the seismic attenuation compensation results under a complex geological model (non-horizontal layered). Figure a1 shows the Velocity Model, representing the actual velocity distribution of the strata. This is a non-horizontal layered model with structural variations such as faults and lithological gradations. Color gradations reflect spatial velocity variations, such as high velocity on the left and low velocity on the right, potentially corresponding to faults or abrupt lithological changes. Figure b1 shows the Reference (no attenuation reference record), a simulated seismic record without attenuation. The phase axis is non-horizontal, reflecting the reflection interface of complex strata, such as faults causing phase axis discontinuity or lithological gradations causing phase axis bending. Figure c1 shows the Attenuated seismic record, at Q=10, simulating the observation record after strata attenuation and complex structures based on the non-uniform velocity model. The phase axis is blurred, the color becomes lighter, and the continuity is poor. Furthermore, due to the complex structures, the attenuation pattern is more difficult to predict. Figure d1: Ours (the compensation method of this application) shows that the phase axis is clearly restored, close to the reference record without attenuation, and the reflection characteristics of complex structures are accurately reproduced, proving that the compensation method of the application has good generalization to non-uniform strata and is suitable for real geology.

[0112] See Figure 12 As shown, this is a comparison of non-horizontal layered structures under the scenarios of Q=10 and SNR=15. Figure a2: Attenuated, showing the non-horizontal layered structure under Q=10 and SNR=15. It can be seen that the phase axis is blurred, but complex structures are still identifiable. Figure b2: Traditional (compensation result using the traditional algorithm), showing that the phase axis is partially recovered, but noise is not effectively suppressed, colors are messy, and complex structural details are lost, such as blurred discontinuities at breaks and discontinuities at bends. Figure c2: Ours (compensation result using the compensation method of this application), showing that the phase axis is clearly recovered, colors are vibrant and continuous, complex structural details are fully preserved, and noise is effectively suppressed.

[0113] See Figure 13 As shown, this is a frequency domain comparison under complex structural scenarios, further verifying the advantages of the compensation method proposed in this application in frequency domain energy compensation. The horizontal axis represents frequency, and the vertical axis represents amplitude. Reference (black) represents the spectrum of the unattenuated reference record, ideally with complete energy across the entire frequency band. Attenuated (green) represents the spectrum of the attenuated record under complex structures with Q=10 and SNR=15, where energy decreases with frequency, especially with severe loss in the high-frequency band. Traditional (blue) represents the spectrum after compensation using traditional algorithms, where energy is partially recovered, but the high-frequency band is still lower than the reference and is subject to noise interference, resulting in an uneven spectrum. Ours (red) represents the compensation method proposed in this application, where the compensated spectrum energy is close to the reference record, with good coverage across the entire frequency band, especially accurate recovery of the high-frequency band.

[0114] In the compensation method of this application, the high-frequency energy is close to the reference, proving that the compensation method can accurately compensate for the full-band attenuation in complex scenarios, especially high-frequency details, key information such as small stratigraphic structures and lithological changes. In the low-frequency band (below 50Hz), the energy of the compensation method of this application is consistent with the reference, indicating that the compensation method of this application accurately controls the low-frequency compensation intensity, avoiding over-compensation or under-compensation.

[0115] Furthermore, based on the time-domain conclusions, in the time domain, the compensation method of this application results in a clear in-phase axis and preserves the complex structure; in the frequency domain, the spectrum of the compensation method of this application is close to the reference.

[0116] See Figure 14 As shown, the neural network model of this application is further explained. The neural network model of this application is a seismic attenuation compensation / reflection coefficient inversion model that integrates CNN (convolutional neural network) and Transformer.

[0117] The neural network model undergoes closed-loop processing, sequentially passing through the input layer, encoding layer, decoding layer, and loss layer. The input layer provides attenuated seismic data (attenuated wavelet) and the original wavelet (wavelet) as the data foundation for the model.

[0118] The encoding layer extracts features through CNN and Transformer. Path Embedding is mainly used to encode path-related information in the data, while Position Embedding mainly provides the model with location information of the data, helping the model to better process and understand earthquake data. The Transformer Encoder performs global correlation and uses Multi-Head Attention to capture globally correlated features. MLP (Multilayer Perceptron) is used for non-linear feature transformation to enhance the model's expressive power. Norm is used for stable training to ensure the stable distribution of features in each layer and improve the model's robustness.

[0119] The decoding layer uses a CNN-based decoder to reconstruct the compensated seismic record or the optimized reflection coefficient.

[0120] The loss layer compares the compensation result with a reference without attenuation, calculates the loss function (Loss), and drives iterative optimization of the model. The smaller the Loss, the closer the compensation effect is to the ideal state.

[0121] Attenuated wavelets are decayed seismic wavelets. Wavelets are undecayed original seismic wavelets. Convolutional preprocessing uses CNNs to extract local features, providing initial feature mappings for the Transformer. Fully connected layers (FCNs) connect the Encoder and Decoder, mapping the global features extracted by the Transformer to the decoding dimension.

[0122] See Figure 15 As shown, this application also provides a seismic data compensation system, which includes: an acquisition module, an addition module, a synthesis module, an evaluation module, and a compensation module.

[0123] The acquisition module is used to acquire seismic data based on the basic convolutional model. Based on this model, it extracts initial seismic data from the raw information acquired in actual data collection or generated through simulation. This provides clean input unaffected by attenuation for subsequent attenuation compensation processes, ensuring that subsequent modules have reliable basic data support.

[0124] The add module is used to add attenuation operators to the basic convolution model to form an attenuated convolution expression; the add module introduces attenuation operators into the basic convolution model to construct an attenuated convolution expression. This simulates the attenuation effect of real geological formations.

[0125] The synthesis module is used to synthesize attenuated seismic records based on the attenuation convolution expression. The synthesis module utilizes the attenuation convolution expression constructed by the addition module to combine basic seismic data with attenuation operators, generating attenuated seismic records through convolution operations. This provides the evaluation and compensation modules with clear correction targets.

[0126] The evaluation module assesses the differences between observed and synthetic attenuated seismic records based on a pre-set neural network model, and obtains the seismic reflection coefficients based on these differences. The module calculates the difference between the actual observed and synthetic attenuated records using a loss function, and optimizes the reflection coefficients based on this difference, thereby driving the reflection coefficients to approximate the true stratigraphic characteristics.

[0127] The compensation module convolves the reflection coefficient with the unattenuated seismic wavelet in the seismic data to construct a seismic attenuation compensation expression. Based on this expression, it outputs attenuated-compensated seismic records. Alternatively, the compensation module, based on the optimized reflection coefficient from the evaluation module, convolves it with the unattenuated seismic wavelet to construct the attenuation compensation expression, ultimately outputting the compensated seismic records to complete the attenuation compensation task.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for compensating seismic data, characterized in that, The compensation method includes: Seismic data is acquired based on a fundamental convolutional model; Based on the aforementioned basic convolution model, a decay operator is added to form a decay convolution expression; Attenuated seismic records are synthesized based on the aforementioned attenuation convolution expression; The difference between the observed attenuated seismic record and the synthetic attenuated seismic record is evaluated based on a pre-set neural network model, and the reflection coefficient of the earthquake is obtained based on the difference. The reflection coefficient is convolved with the unattenuated seismic wavelet in the seismic data to construct a seismic attenuation compensation expression, and the attenuated compensated seismic record is output according to the seismic attenuation compensation expression.

2. The compensation method according to claim 1, characterized in that, The steps of adding a decay operator to the basic convolution model to form a decay convolution expression include: The expression for the unattenuated seismic wave is obtained based on the aforementioned basic convolution model; Convert the seismic wave expression into a frequency domain expression form; Add an attenuation operator to the frequency domain expression to transform it into an attenuation convolution expression.

3. The compensation method according to claim 2, characterized in that, The expression for an unattenuated seismic wave is s(t) = W(t) * r(t), where s(t) is the seismic record, W(t) is the seismic wavelet matrix, r(t) represents the reflection coefficient, * represents the convolution operation, and t is the total propagation and reflection time of the seismic wave. The frequency domain expression is s(ω)=W(ω)∫r(t)A(ω,t)e -iωt dt, s(ω) is the frequency domain representation of the seismic record amplitude, W(ω) is the frequency domain representation of the seismic wavelet matrix, A(ω,t) is the frequency domain representation of the attenuation operator, and ω is the frequency; The decay convolution expression is: W a (t) Attenuated seismic wavelet, W a =W×A, where W is the original seismic wavelet and A is the attenuation operator.

4. The compensation method according to claim 3, characterized in that, The frequency domain expression of the attenuation operator is: Where γ = (1 / πQ), γ is the frequency dependence exponent. e is the amplitude attenuation factor. iωt The phase rotation factor, For frequency-dependent compensation factors, ω h Here, Q is the reference angular frequency, and Q is the quality factor.

5. The compensation method according to claim 1, characterized in that, The differences include residuals; The steps of evaluating the differences between observed attenuated seismic records and the synthetic attenuated seismic records based on a pre-set neural network model include: The initial reflection coefficients are output based on a pre-set neural network model; Attenuated seismic records are synthesized by convolution based on the initial reflection coefficient and the attenuated wavelet in the attenuation convolution expression; Determine the residual between the observed attenuated seismic record and the synthesized attenuated record, and minimize the residual to determine the reflection coefficient.

6. The compensation method according to claim 5, characterized in that, The residual is expressed as a loss function; The steps to minimize the residuals to determine the reflection coefficient include: The loss function between the observed attenuated seismic record and the synthetic attenuated seismic record is calculated, and the neural network parameters are updated via backpropagation to update the reflection coefficient.

7. The compensation method according to claim 6, characterized in that, The loss function is J(r), The actual observed attenuation records satisfy the following:

8. The compensation method according to claim 6, characterized in that, The step of backpropagating to update the neural network parameters to update the reflection coefficients includes: The loss function is differentiated according to a pre-defined chain rule, and the network parameters of the neural network are updated accordingly. Based on the updated network parameters, a nonlinear mapping between the observed attenuated seismic records and the reflection coefficients is established, so as to predict and update the reflection coefficients based on the neural network. The derivative of the loss function satisfies the following: w new Updated neural network parameters, w old The neural network parameters before the update, where 'a' is the learning rate.

9. The compensation method according to claim 6, characterized in that, The seismic attenuation compensation expression is as follows: Updated reflectance coefficient; Net() represents a neural network, and w represents the parameters of the neural network.

10. A seismic data compensation system, characterized in that, The compensation system includes: The acquisition module is used to acquire seismic data based on the basic convolutional model; Add a module to add a decay operator based on the basic convolution model to form a decay convolution expression; The synthesis module is used to synthesize attenuated seismic records based on the attenuation convolution expression; The evaluation module is used to evaluate the differences between the observed attenuated seismic records and the synthetic attenuated seismic records based on a pre-set neural network model, and to obtain the reflection coefficient of the earthquake based on the differences. The compensation module is used to convolve the reflection coefficient with the unattenuated seismic wavelet in the seismic data to construct a seismic attenuation compensation expression, and output attenuation-compensated seismic records based on the seismic attenuation compensation expression.

Images