A method for modeling near-surface stratum Q value of micro-logging data in loess plateau region

Abstract

The application discloses a kind of near-surface stratum Q value modeling methods for micro-logging data in loess plateau region, which includes: step one: using improved micro-logging first arrival spectrum ratio method to obtain the near-surface micro-logging Q value of logging point;Step two: using Kriging method to carry out spatial interpolation on micro-logging Q value data, obtain low-frequency near-surface Q value model;Step three: using machine learning to predict the rest of the spatial position Q value through known logging Q value data, obtain high-frequency near-surface Q value model;Step four: scale fusion is carried out to low-frequency, high-frequency near-surface Q value model, and a fine near-surface Q value model is established;Step five: using fine near-surface Q value model to carry out absorption and attenuation compensation processing on seismic data.The application solves the problem that the extracted logging Q value is not accurate due to the poor coupling relationship of shotpoint and receiver, provides reliable key Q value data for solving the absorption and attenuation compensation problem in complex near-surface area, and can be used in industrial production of seismic exploration.

Description

Technical Field

[0001] This invention belongs to the field of seismic data processing for oil and gas exploration, and specifically relates to a method for near-surface formation Q-value modeling using micrologging data in the Loess Plateau region. Background Technology

[0002] The Loess Plateau region is characterized by dramatic topographic relief and numerous gullies, with significant variations in near-surface lithology and thickness: the loess layer varies in thickness from 0 to 300 meters, is loose, and exhibits weak elasticity; the loess layer on the slopes contains multiple layers of gravel of varying depths, thicknesses, and sizes; and bedrock and gravel are exposed at the bottom of the gullies. Due to the near-surface's characteristics of being a "free surface, low velocity, and high absorption," coupled with issues such as excitation factors in the work area and the layout of shot receivers, the received signals are distorted, severely impacting the quality of seismic data in the Loess Plateau region. Therefore, establishing a reasonable near-surface Q-value model and compensating for absorption attenuation in seismic data to improve its quality is crucial. Existing near-surface Q-value estimation methods can be broadly categorized into two types: borehole core sampling Q-value estimation and stratigraphic measurement Q-value estimation. The former can be further subdivided into three types based on different testing principles: stress-strain method, standing wave method, and traveling wave method; the latter can be subdivided according to different field observation techniques, mainly including small refraction method, first arrival wave method, and micro-logging method. Core sampling testing offers relatively high accuracy for Q estimation, but its practicality in large-scale seismic exploration is limited by factors such as sample limitations, inconsistencies between measurement and exploration frequencies, and high sampling costs. The small refraction method is affected by near-surface conditions and is unsuitable for areas with complex surface and subsurface structures. The first-arrival direct wave method is effective in estimating the average Q-value of low-velocity near-surface layers and can well describe the lateral variation of near-surface Q-values, but requires high-quality seismic data. Furthermore, the first-arrival method cannot reflect the attenuation characteristics of high-velocity near-surface layers. Due to the unique characteristics of the near-surface in the Loess Plateau and the poor data quality, neither the small refraction method nor the first-arrival direct wave method can be used for near-surface Q-value extraction in this region. Micrologging can obtain relatively accurate Q-values ​​near the observation site and has high vertical resolution, with fewer limitations, making it widely used in practical exploration. However, the poor coupling between shot and receiver points in the Loess Plateau region limits the accuracy of existing micrologging Q-value extraction methods.

[0003] To address the issue of poor Q-value extraction performance of micrologging Q-value extraction methods in the complex near-surface and subsurface structures of the Loess Plateau, where other Q-value extraction methods are limited and poor coupling between shot and receiver points leads to ineffective extraction, this paper first proposes a method for calculating near-surface formation Q-values ​​from micrologging data in the Loess Plateau region, reducing the impact of shot-receiver point coupling on Q-value extraction. The first-arrival spectral ratio method for Q-value extraction from micrologging is simple to calculate and has no usage limitations. Based on this, an improved spectral ratio method is derived theoretically, forming a method suitable for calculating near-surface formation Q-values ​​from micrologging data in the Loess Plateau region. Based on the Q-values ​​obtained from the improved spectral ratio method, lateral extrapolation and interpolation are performed to obtain a near-surface Q-value model. Kriging interpolation, a geostatistical interpolation method, can effectively extract low-frequency information. Machine learning can predict the Q-values ​​of the entire work area based on known micrologging Q-values, characterizing high-frequency components. The low-frequency and high-frequency components of the Q-value model were obtained by Kriging interpolation and machine learning, respectively, and then fused to form a near-surface Q-value model that can be used for absorption attenuation compensation.

[0004] No relevant literature has been published on how to use the improved first arrival spectral ratio method to eliminate the coupling effect of shot and receiver points and calculate the logging Q value using micrologging data. Summary of the Invention

[0005] The purpose of this invention is to provide a near-surface formation Q-value modeling method for micrologging data in the Loess Plateau region. By establishing a reasonable near-surface Q-value model and compensating for absorption attenuation of seismic data, high resolution of seismic data can be achieved. This overcomes the problems in actual work areas such as the Loess Plateau and complex mountainous areas, where the surface is undulating, the low-velocity layer is thick, and the lateral velocity changes are drastic. The near-surface absorption attenuation causes a decrease in seismic wave resolution, and the resulting changes in seismic attributes are much greater than those caused by formation and hydrocarbon factors.

[0006] To achieve the above objectives, the technical solution adopted by this invention is: a method for modeling near-surface formation Q-values ​​using micrologging data in the Loess Plateau region, comprising the following steps:

[0007] Step 1: Based on the traditional micro-logging first arrival spectrum ratio method, an improved method is derived, and the improved method is used to process the micro-logging data to obtain the near-surface micro-logging Q value of the logging point.

[0008] Step 2: Based on the calculation results in Step 1, spatial interpolation of the micrologging Q-value data is performed using the Kriging method to obtain a low-frequency near-surface Q-value model;

[0009] Step 3: Based on the calculation results in Step 2, use machine learning to predict the Q values ​​of other spatial locations using known well logging Q value data, and obtain a high-frequency near-surface Q value model containing high-frequency information;

[0010] Step 4: Perform scale fusion on the low-frequency near-surface Q-value model obtained in Step 2 and the high-frequency near-surface Q-value model obtained in Step 3, and extract the relatively reliable low-frequency and high-frequency components of the model respectively to establish a fine near-surface Q-value model that can be used for absorption attenuation compensation.

[0011] Step 5: Use the near-surface Q-value model constructed in Step 4 to perform absorption attenuation compensation processing on the seismic data.

[0012] Furthermore, in step one, the calculation process for the near-surface micrologging Q value is as follows:

[0013] The received frequency-domain micro-logging seismic records are characterized by the following:

[0014] y 11 =x 11 ×s1×r1,y 12 =x 12 ×s1×r2

[0015] y 21 =x 21 ×s2×r1,y 22 =x 22 ×s2×r2

[0016] y represents the actual first arrival record, x represents the record without coupling response, s represents the shot point coupling response, and r represents the receiver point coupling response;

[0017] According to the traditional first-arrival spectral ratio method, the following logarithmic spectral ratio can be obtained:

[0018]

[0019]

[0020]

[0021] Q1, Q2, and Q3 represent the Q-values ​​of the first layer, the second layer, and the equivalent Q-values ​​of the first two layers, respectively; A1(f) is the logarithmic spectral ratio used to calculate the Q-value of the first layer, affected by the detector coupling, and Δt1 is the time difference of s1 propagating to r1 and r2; A21(f) and A22(f) are the logarithmic spectral ratios used to calculate the Q-value of the second layer, affected by the shot point, and Δt2 is the time difference of s2 and s1 propagating to r1, Δt3 is the time difference of s2 and s1 propagating to r2; A3(f) is the logarithmic spectral ratio used to calculate the equivalent Q-values ​​of the first two layers, affected by the detector, and Δt4 is the time difference of s2 propagating to r1 and r2;

[0022] To eliminate the effects of shot-receiver coupling, the formula for calculating the logarithmic spectral ratio of the second-layer Q-value is derived:

[0023]

[0024] Similarly, the Q value of each layer except the first layer is calculated using this method;

[0025] Assuming the initial amplitude of the seismic data is A1, and the initial amplitude after attenuation through the first two layers of medium is A2, according to Futeman attenuation, the following relationship holds:

[0026]

[0027] ΔT1, ΔT2, and ΔT3 represent the propagation times of the first, second, and two layers, respectively. Simplifying, the relationship between Q1 and Q3 is given by the following equation:

[0028]

[0029] Substituting the above equation into:

[0030] This allows us to obtain the first layer of Q-values ​​that eliminate the coupling between the shot and the receiver, thus obtaining the accurate Q-values ​​for micro-logging.

[0031] Furthermore, in step two, the Kriging interpolation method used in the interpolation process treats the X, Y coordinates and the Z-coordinate of elevation together as spatial constraints to complete the data interpolation.

[0032] Furthermore, in step three, machine learning is used to predict the LSTM network trained using the Q-values ​​calculated from all known well logs, and the training data is normalized to unify the dimensions of the network input and output data, thereby accelerating network convergence.

[0033] Furthermore, the LSTM network is based on LSTM unit modules, which have two hidden layers. Each layer has 256 LSTM units, which are connected by skip connections. The output of the last LSTM unit is connected to a fully connected layer to control the output dimension.

[0034] Furthermore, the spatial location of the well logging Q-value is input into the LSTM, and the corresponding formation Q-value is output to complete the training of the LSTM. During the training process, the error between the output formation Q-value and the sample label is defined by Loss, which is characterized by the following:

[0035] Loss1 = mean(abs(Q_label - Q))

[0036] label is the label, mean is the average operator, abs is the absolute value operator, and Q is the Q value of the strata near the surface.

[0037] Furthermore, in step four, the scale fusion process is as follows: multi-scale decomposition is performed using curvelet transform, but in actual processing, only two-scale decomposition is performed, namely, decomposition into high-frequency scale and low-frequency scale; the low-frequency scale signal from Kriging interpolation decomposition and the high-frequency scale signal from the LSTM network prediction result are respectively taken to form a new data volume, and then the inverse curvelet transform is performed on the new data volume to obtain the fused model.

[0038] Furthermore, in step five, the absorption attenuation compensation process employs an adaptive gain-limited inverse Q-filtering method, and the compensation formula is shown below:

[0039]

[0040] In the formula: t represents travel time; w d (t) is the cutoff angular frequency within the effective frequency range; It is the amplitude compensation function; θ(t,w) is the phase compensation function; c(t) = A(t,w) L ) is the time-varying gain limit of the amplitude compensation function A(t,w), in dB; it is related to the cutoff angular frequency ω of the effective frequency range of the seismic wave. L Correspondingly; c 2 (t)-2c(t) -1 It is a stabilizing factor used to suppress high-frequency noise and the Gibbs effect; U(t,w) and U(t,w) are the seismic record wavefields at time t before and after inverse Q filtering, respectively.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] First, the reliability of the methodology. It primarily utilizes the first-arrival spectrum ratio method, Kriging interpolation, and machine learning algorithms, all of which have mature theoretical foundations and proven effectiveness. Theoretical derivations and innovations are made based on the traditional first-arrival spectrum ratio method, ensuring accurate acquisition of micrologging Q-values. Based on accurate logging Q-values, the model obtained by interpolating micrologging Q-values ​​using the Kriging method exhibits good spatial constraint capabilities, ensuring stable and reliable large-scale trends. The model predicted by machine learning helps recover high-frequency components of complex structures. The fusion of Kriging interpolation and machine learning interpolation in the near-surface Q-value model demonstrates high reliability in principle. Furthermore, compared to the conventional first-arrival spectrum ratio method, the improved method effectively addresses receiver coupling issues under all circumstances.

[0043] Secondly, the operation is simple and easy to implement. The code for the first-arrival spectral ratio method, Kriging interpolation, and hybrid density network algorithm is readily available. By using the improved first-arrival spectral ratio method to obtain the micro-logging Q-value calculation results of the actual exploration area, it is easy to interpolate and obtain the Kriging interpolation model and machine learning prediction model. By directly fusing the two interpolation results at multiple scales, it is possible to model the near-surface Q-value model under the complex tectonic conditions of the Loess Plateau.

[0044] Third, the reliability of the calculation results. During the calculation process, the influence of shot-receiver coupling on the extraction of micrologging Q-values ​​was reduced, and the constraints of measured micrologging information were fully utilized. The near-surface Q-value model after multi-scale fusion ensures the reliability of the reconstruction. Compared with other Q-value modeling methods, this algorithm has higher lateral resolution and is based on reliable micrologging data, making the reconstruction results more reliable than other methods. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of a single-well micro-logging.

[0046] Figure 2 It is an LSTM network used for training.

[0047] Figure 3 This is a flowchart of a method for modeling near-surface formation Q-values ​​using micrologging data in the Loess Plateau region.

[0048] Figure 4 This is a flowchart of the scale fusion process between the low-frequency near-surface Q-value model and the high-frequency near-surface Q-value model.

[0049] Figure 5 This is an overview of area A; (a) is the actual micrologging data of the work area, and (b) is the actual micrologging distribution map of the work area.

[0050] Figure 6 The results are the near-surface Q-value modeling results for region A in this invention; where (a) is the Q-value calculated by micro-logging, (b) is the Kriging interpolation result, (c) is the prediction result of the machine learning network, and (d) is the fusion result.

[0051] Figure 7 These are comparison images of single-shot arrangement and horizontally superimposed profile absorption attenuation compensation in area A of this invention; where (a) is the pre-stack shot record, (b) is the absorption attenuation compensation effect, (c) is the horizontally superimposed profile record, and (d) is the absorption attenuation compensation effect.

[0052] Figure 8 These are the near-surface Q-value modeling results for region B in this invention; where (a) is the Kriging interpolation result, (b) is the machine learning prediction result, and (c) is the fusion result.

[0053] Figure 9These are comparison images of single-shot arrangement and horizontally superimposed profile absorption attenuation compensation in area B of this invention; where (a) is the pre-stack shot record, (b) is the absorption attenuation compensation effect, (c) is the horizontally superimposed profile record, and (d) is the absorption attenuation compensation effect. Detailed Implementation

[0054] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0055] like Figure 1-4 As shown, a method for near-surface formation Q-value modeling using micrologging data in the Loess Plateau region includes the following steps:

[0056] Step 1: Estimate the logging Q value using micrologging data, and calculate the accurate Q value after reducing poor coupling between the shot and receiver points using the improved first arrival spectrum ratio method.

[0057] Step 2: Based on the calculation results in Step 1, spatial interpolation of the micrologging Q-value data is performed using the Kriging method to obtain a low-frequency near-surface Q-value model.

[0058] Step 3: Based on the calculation results in Step 2, use machine learning to predict the Q values ​​of other spatial locations using known well logging Q value data, and obtain a near-surface Q value model containing high-frequency information.

[0059] Step 4: Perform scale fusion of the low-frequency near-surface Q-value model obtained in Step 2 and the high-frequency near-surface Q-value model obtained in Step 3 to obtain a more refined near-surface model.

[0060] Step 5: Use the near-surface Q-value model constructed in Step 5 to perform absorption attenuation compensation processing on the seismic data.

[0061] In step one, considering the influence of the shot point and the receiver point, the received frequency domain micro-logging seismic record can be characterized as follows:

[0062] y 11 =x 11 ×s1×r1,y 12 =x 12 ×s1×r2

[0063] y 21 =x 21 ×s2×r1,y 22 =x 22 ×s2×r2

[0064] y represents the actual first arrival record, x represents the record without coupling response, s represents the shot point coupling response, and r represents the receiver point coupling response. Based on the traditional first arrival spectral ratio method, the following logarithmic spectral ratio can be obtained:

[0065]

[0066]

[0067]

[0068] Q1, Q2, and Q3 represent the first-layer Q-value, the second-layer Q-value, and the equivalent Q-values ​​of the first two layers, respectively; A1(f) is the logarithmic spectral ratio used to calculate the first-layer Q-value, affected by detector coupling, where Δt1 is the time difference between s1 and r1 / r2; A21(f) and A22(f) are the logarithmic spectral ratios used to calculate the second-layer Q-value, affected by the shot point, where Δt2 is the time difference between s2 and s1 and r1, and Δt3 is the time difference between s2 and s1 and r2; A3(f) is the logarithmic spectral ratio used to calculate the equivalent Q-values ​​of the first two layers, affected by the shot point, where Δt4 is the time difference between s2 and r1 / r2; to eliminate the influence of shot-spot coupling, the formula for calculating the logarithmic spectral ratio of the second-layer Q-value is derived:

[0069]

[0070] Similarly, the Q value of each layer except the first layer can be calculated using this method, eliminating the influence of shot-receiver coupling;

[0071] Assuming the initial amplitude of the seismic data is A1, and the initial amplitude after attenuation through the first two layers of medium (vertical propagation) is A2, according to Futeman attenuation, the following relationship exists:

[0072]

[0073] ΔT1, ΔT2, and ΔT3 represent the propagation times of the first, second, and two layers, respectively. Simplifying, the relationship between Q1 and Q3 is given by the following equation:

[0074]

[0075] Substituting the above equation into:

[0076] This allows us to obtain the first layer of Q-values ​​that eliminate the coupling between the shot and the receiver, thus obtaining the accurate Q-values ​​for micro-logging.

[0077] In step two, the Kriging interpolation method used is the ordinary Kriging interpolation algorithm. However, during the interpolation process, the X, Y coordinates and the elevation Z coordinate are used together as spatial constraints to complete the data interpolation.

[0078] In step three, the machine learning prediction mainly uses an LSTM network trained with the Q-values ​​calculated from all known well logs, and normalizes the training data, unifying the dimensions of the network input and output data to accelerate network convergence.

[0079] like Figure 2 As shown, the LSTM network is mainly composed of LSTM unit modules, with two hidden layers. Each layer has 256 LSTM units, which are connected by skip connections. The output of the last LSTM unit is connected to a fully connected layer to control the output dimension. The spatial location of the well logging Q-value is input into the LSTM, and the corresponding formation Q-value is output to complete the training of the LSTM. During the training process, the error between the output formation Q-value and the sample label is defined by Loss, which is characterized by the following:

[0080] Loss1 = mean(abs(Q_label - Q))

[0081] label is the label, mean is the average operator, abs is the absolute value operator, and Q is the Q value of the strata near the surface.

[0082] In step four, curvelet transform is used for multi-scale decomposition. In actual processing, only two-scale decomposition is performed, namely, decomposition into high-frequency scale and low-frequency scale. The low-frequency scale signal from Kriging interpolation decomposition and the high-frequency scale signal from the LSTM network prediction result are respectively used to construct a new data volume. Then, inverse curvelet transform is performed on the new data volume to obtain the fused model. The fusion process is as follows: Figure 4 As shown.

[0083] In step five, the pre-stack seismic data is subjected to absorption attenuation compensation processing based on the obtained near-surface Q-value model to improve its resolution. This mainly employs an adaptive gain-limited inverse Q-filtering method, and the compensation formula is shown below:

[0084]

[0085] In the formula: t represents travel time; w d (t) is the cutoff angular frequency within the effective frequency range; It is the amplitude compensation function; θ(t,w) is the phase compensation function; c(t) = A(t,w) L ) is the time-varying gain limit of the amplitude compensation function A(t,w), in dB; it is related to the cutoff angular frequency ω of the effective frequency range of the seismic wave. L Correspondingly; c 2 (t)-2c(t) -1 It is a stabilization factor, mainly used to suppress high-frequency noise and the Gibbs effect in order to solve the instability problem of inverse Q filtering; U(t,w) and U(t,w) are the seismic record wavefields at time t before and after inverse Q filtering, respectively.

[0086] To verify the near-surface formation Q-value modeling method using micrologging data in the Loess Plateau region and its effect after absorption compensation, the following analysis takes the Q-value modeling process using seismic micrologging data in a certain western region A as an example and the results after absorption compensation.

[0087] Figure 5 (a) is the actual micro-logging data for area A. Figure 5 (b) shows the location distribution information of micrologging wells in area A. The receiver spacing is 20 meters, and there are a total of 305,745 receivers to be interpolated. The micrologging data were calculated using the improved first-arrival spectral ratio method to obtain the near-surface Q-value information of the logging locations. This information was used as known data for near-surface Q-value model reconstruction.

[0088] Figure 6 This demonstrates the logging Q-value calculated using the improved spectral ratio first-arrival method. Figure 6 (a)), Q-value model results obtained by Kriging interpolation ( Figure 6 (b) and machine learning prediction models ( Figure 6 (c)). From Figure 6 It can be seen that Kriging interpolation results are dominated by low-frequency components, but high-frequency components are severely lacking, while machine learning prediction results are rich in high-frequency information. Using... Figure 6 The two models are fused together to obtain, for example... Figure 6 (d) Model Results. The near-surface Q-value model established in this invention was used for absorption attenuation compensation processing, and horizontally stacked profiles were obtained. The results were compared between the original seismic data and the absorption attenuation compensation effect of the model of this invention, as shown below. Figure 7 The results are shown. A comparison of the results shows that the invention significantly improves the resolution of near-surface data, and deep energy is fully compensated.

[0089] To further illustrate the effectiveness of the method, micrologging data from another region, B, were also processed in the same way. The resulting near-surface Q-value model is shown below. Figure 8 As shown.

[0090] Figure 7 , Figure 9 A comparison was made between the original seismic data and the absorption attenuation compensation effect of the model of this invention. The comparison results show that the near-surface Q-model obtained by this invention greatly improves the resolution of the data, broadens the effective frequency band, and compensates for deep energy.

[0091] Experiments using data from regions A and B revealed that, based on well logging Q-values ​​calculated using the improved first-arrival spectral ratio method, a method combining Kriging interpolation and machine learning predictions can effectively obtain near-surface Q-value models suitable for absorption attenuation compensation. Comparison with the original data showed that this method effectively improves data resolution.

[0092] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the scope of protection of the present invention in any way, and all technical solutions obtained by equivalent substitution or other means fall within the scope of protection of the present invention.

[0093] All parts not covered in this invention are the same as or can be implemented using existing technologies.

Claims

1. A method for modeling near-surface formation Q-values ​​using micrologging data in the Loess Plateau region, characterized in that, Includes the following steps: Step 1: Based on the traditional micro-logging first arrival spectrum ratio method, an improved method is derived, and the improved method is used to process the micro-logging data to obtain the near-surface micro-logging Q value of the logging point. The calculation process for the Q value of near-surface micrologging is as follows: The received frequency-domain micro-logging seismic records are characterized by the following: This is an actual initial arrival record. Records that do not have a coupling response. For the shot point coupled response, This is the coupling response at the detector point; According to the traditional first-arrival spectral ratio method, the following logarithmic spectral ratio can be obtained: Q1, Q2, and Q3 are the Q-values ​​of the first layer, the second layer, and the equivalent Q-values ​​of the first two layers, respectively. The logarithmic spectral ratio used to calculate the Q-value of the first layer is affected by detector coupling. for spread to Time difference; , These are the logarithmic spectral ratios used to calculate the Q-values ​​of the second layer, affected by the shot point. for and spread to Time difference, for and spread to Time difference; The logarithmic spectral ratio used to calculate the equivalent Q-values ​​of the first two layers is affected by the detector. for spread to Time difference; To eliminate the effects of shot-receiver coupling, the formula for calculating the logarithmic spectral ratio of the second-layer Q-value is derived: Similarly, the Q value of each layer except the first layer is calculated using this method; Assuming the initial amplitude of the seismic data is A1, and the initial amplitude after attenuation through the first two layers of medium is A2, according to Futeman attenuation, the following relationship holds: The propagation times for the first, second, and two layers are respectively given. Simplifying, the relationship between Q1 and Q3 is given in the following equation: Substituting the above equation into: ; This allows us to obtain the first layer of Q-values ​​that eliminate the coupling between the shot and receiver, thus obtaining the accurate Q-values ​​for micro-logging. Step 2: Based on the calculation results in Step 1, spatial interpolation of the micro-logging Q-value data is performed using the Kriging method to obtain a low-frequency near-surface Q-value model. The Kriging interpolation method used here treats the X, Y coordinates and the Z-coordinate of elevation as spatial constraints during the interpolation process to complete the data interpolation. Step 3: Based on the calculation results in Step 2, use machine learning to predict the Q values ​​of other spatial locations using known well logging Q value data, and obtain a high-frequency near-surface Q value model containing high-frequency information; The machine learning prediction uses an LSTM network trained with Q-values ​​calculated from all known well logs, and normalizes the training data, unifying the dimensions of the network input and output data to accelerate network convergence. Step 4: Perform scale fusion on the low-frequency near-surface Q-value model obtained in Step 2 and the high-frequency near-surface Q-value model obtained in Step 3, and extract the relatively reliable low-frequency and high-frequency components of the model respectively to establish a refined near-surface Q-value model that can be used for absorption attenuation compensation. The scale fusion process is as follows: use curvelet transform to perform multi-scale decomposition. In actual processing, only two-scale decomposition is performed, that is, decomposed into high-frequency scale and low-frequency scale. The low-frequency scale signal from Kriging interpolation decomposition and the high-frequency scale signal from the LSTM network prediction results are used to construct a new data volume. Then, the inverse curvelet transform is performed on the new data volume to obtain the fused model. Step 5: Use the near-surface Q-value model constructed in Step 4 to perform absorption attenuation compensation processing on the seismic data.

2. The method for near-surface formation Q-value modeling based on micrologging data in the Loess Plateau region according to claim 1, characterized in that: The LSTM network is based on LSTM unit modules, which have two hidden layers. Each layer has 256 LSTM units, which are connected by skip connections. The output of the last LSTM unit is connected to a fully connected layer to control the output dimension.

3. A method for near-surface formation Q-value modeling based on micrologging data in the Loess Plateau region according to claim 1 or 2, characterized in that: The spatial location of the well logging Q-value is input into the LSTM, and the corresponding formation Q-value is output to complete the training of the LSTM. During the training process, the error between the output formation Q-value and the sample label is defined by Loss, which is characterized by the following: It's a tag. It is an average operator. It is an absolute value operator, and Q is the Q value of the stratum near the surface.

4. The method for near-surface formation Q-value modeling based on micrologging data in the Loess Plateau region according to claim 1, characterized in that: In step five, the absorption attenuation compensation process employs an adaptive gain-limited inverse Q-filtering method, and the compensation formula is shown below: In the formula: For travel; The cutoff angular frequency is the effective frequency range. It is the amplitude compensation function; It is a phase compensation function; It is an amplitude compensation function The time-varying gain limit, dB; Its cutoff angular frequency relative to the effective frequency range of seismic waves Correspondingly; It is a stabilizing factor used to suppress high-frequency noise and the Gibbs effect; and Before and after inverse Q filtering, respectively The seismic record wave field at a given moment.

Images