Full-band high-precision seismic inversion method

Abstract

The invention relates to a full-band high-precision seismic inversion method, which belongs to the technical field of reservoir prediction in oil and gas field exploration, and comprises the following steps: firstly, carrying out low and medium frequency filtering interpolation on logging data to establish a wave impedance initial model, and then carrying out wave impedance inversion based on seismic data; and carrying out waveform clustering on residual seismic disturbance obtained by subtracting actual seismic data from forward modeling data of the wave impedance inversion result, carrying out transverse high-frequency inversion on a logging high-frequency component according to a waveform clustering result, and finally fusing a low and medium frequency inversion result and a high-frequency inversion result to obtain a full-band high-precision inversion result. According to the method, the high-frequency component inversion is driven by the residual seismic disturbance obtained by subtracting the actual seismic data from the forward modeling data of the low and medium frequency inversion result, so that the difficulty of inaccurate high-frequency information inversion of the existing inversion method is effectively overcome, the full-band high-precision inversion driven by the seismic data is realized, the resolution and accuracy of the seismic inversion are greatly improved, and the method is suitable for large-scale popularization and application. And accurate identification of the thin reservoir in oil and gas field exploration and development is facilitated.

Description

Technical Field

[0001] This invention relates to the field of reservoir prediction technology in oil and gas seismic exploration, and in particular to a high-resolution seismic inversion method. Background Technology

[0002] Seismic inversion technology is a crucial tool for quantitative reservoir prediction in oil and gas seismic exploration. The results of seismic inversion lay a solid foundation for refined reservoir prediction and precise oil and gas exploration and development. As exploration and development deepen, the focus shifts from shallow to deep burial sites and from thick to thin reservoirs, demanding increasingly higher accuracy in reservoir prediction, particularly for the quantitative inversion of thin reservoirs. The theoretical resolution of seismic data is 1 / 4 wavelength. However, due to the band-limited nature of seismic data, high-frequency information is lacking. Currently, the actual reservoir thickness in oil and gas exploration and development is far less than 1 / 4 wavelength. Conventional sparse pulse inversion cannot identify thin layers with a resolution lower than that of seismic data. Therefore, identifying thin layers with a thickness less than 1 / 4 wavelength has always been a key objective of high-resolution seismic inversion.

[0003] For high-resolution seismic inversion of thin layers, there are currently four main types of methods:

[0004] (1) Model inversion: First, well-seismic calibration establishes the correspondence between time-domain seismic data and depth-domain logging curves; a low-frequency acoustic impedance background model is established using the logging curves; then, a synthetic seismic trace is established through forward modeling and compared with the actual seismic trace. The optimal result is obtained through continuous iteration, and finally, a suitable acoustic impedance is obtained (Russell & Hampson, 1991). The model-based post-stack inversion method has the advantages of high resolution and high consistency with logging information, and has been widely used in actual production.

[0005] (2) Geostatistical inversion: Geostatistical inversion was proposed by Bortoli et al. (1992) and Haas and Dubrule (1994), and further developed by Dubrule et al. (1998) and Rothman (1998). It is based on geostatistical information to express the prior probability density function of the solution space. Through the organic interaction between stochastic simulation and optimized simulation results, it fully leverages the advantages of stochastic modeling and seismic inversion, enabling the inversion results to effectively overcome the bandwidth limitations of seismic data and improve resolution.

[0006] (3) Waveform Indication Inversion: Seismic waveform classification technology and seismic sedimentology fully utilize the advantage of high lateral resolution of seismic data. By introducing high-frequency well logging curves, an intrinsic relationship is established between low-frequency seismic waveforms and high-frequency well logging curves. This method establishes a mapping relationship between the structure of seismic waveforms and the structure of high-frequency well logging curves through efficient dynamic clustering of seismic waveforms, thereby improving the vertical and lateral resolution of the inversion results and increasing the resolution of seismic inversion to 2-3m. By constructing Bayesian inversion frameworks for different seismic facies types, true facies-controlled inversion is achieved (Chen Yanhu, 2020).

[0007] (4) Deep learning seismic inversion: In recent years, deep learning algorithms have received widespread attention in the field of geophysical inversion due to their superior nonlinear representation capabilities. Based on neural networks, the connection between seismic waveforms and target curves is established. Through various network models and optimization algorithms, end-to-end mapping is achieved, training data is used to establish inversion models, thereby enabling the direct prediction of target curves from seismic waveforms. Mosser et al. (2020) proposed a stochastic seismic waveform inversion method using generative adversarial networks (GANs) as geological priors. Wu et al. (2020) presented a wave impedance inversion method using a fully convolutional residual network (FCRN) combined with a transfer learning strategy. Liu et al. (2022) proposed a model-driven seismic inversion method based on small sample data. Su et al. (2023) proposed a wave impedance inversion method based on geophysically constrained closed-loop Unet with spatial background information. Yang et al. (2025) proposed a three-dimensional seismic wave impedance inversion method based on semi-supervised V-Net and transfer learning.

[0008] Based on thorough research and practical data application, this invention identifies the following problems with existing high-resolution seismic inversion methods:

[0009] (1) Although the results of model inversion, geostatistical inversion, waveform indicator inversion and neural network inversion all have the advantages of high resolution and high consistency with well logging information, the accuracy of high frequency components in their inversion results is low and they cannot characterize the real lateral changes in the formation.

[0010] (2) The high-frequency components in the inversion results of the model inversion come from the high-frequency information in the initial model established by the well logging data. In the actual inversion iteration, only the information of the medium and low frequencies in the seismic frequency band is iterated and modified. Therefore, its high-frequency information is essentially the result of well interpolation. The high-frequency components of geostatistical inversion come from the random simulation of well logging data. Neither of the two methods uses seismic data to drive the inversion of high-frequency information. The high-frequency information is the result of well logging interpolation or random simulation. Therefore, its lateral resolution is low and its reliability is low.

[0011] (3) Waveform indication inversion adopts the idea of ​​model wave impedance inversion in the low and medium frequencies. The high frequency components in the inversion results are obtained by directly establishing a relationship between the high frequency information of well logging and the seismic data. The high frequency information inversion results are driven by the seismic data and can reflect the real changes in the lateral strata to a certain extent. However, the seismic data is a comprehensive response of the real strata (which includes both low-frequency thick and thin layer information in the full frequency band) and mainly reflects the low and medium frequency information. Therefore, directly establishing a relationship between the high frequency information of well logging and the waveform of the comprehensive response of the real strata to drive the inversion of high frequency information has certain defects. Its high frequency inversion results reflect the low-frequency lateral variation characteristics more.

[0012] (4) Neural network inversion establishes a nonlinear mapping relationship between well logging data and seismic data through deep learning theory. On the one hand, it lacks the support of geophysical theory and geological laws. On the other hand, due to the huge difference between the number of actual drilled wells and the number of seismic data, the neural network inversion method has serious multi-solution problems and poor generalization ability of the training model. The high-frequency components of its inversion results are bound to the low-frequency components, and the high-frequency inversion results reflect the low-frequency lateral variation characteristics more. Summary of the Invention

[0013] To address the problem of low accuracy of high-frequency components in existing seismic inversion methods, which fail to accurately characterize the lateral variations of real strata, this invention proposes a full-band high-precision seismic inversion method. This method first establishes an initial wave impedance model by performing mid-to-low frequency filtering and interpolation on well logging data. Then, wave impedance inversion is performed based on the seismic data. The residual seismic disturbance obtained by subtracting the forward model data from the actual seismic data is subjected to waveform clustering. Based on the waveform clustering results, the high-frequency components of the well logging data are subjected to lateral high-frequency inversion. Finally, the mid-to-low frequency inversion results are fused with the high-frequency inversion results to obtain a full-band high-precision inversion result. This achieves seismic data-driven full-band high-precision inversion, significantly improving the resolution and accuracy of seismic inversion. The specific technical solution is as follows:

[0014] S1: First, the well logging impedance curves of the study area are separated into low-frequency and high-frequency impedance curves by bandpass filtering.

[0015] S2: Inversion of mid-to-low frequency components:

[0016] S2-1: For the target layer in the study area, an initial wave impedance inversion model is established based on the interpolation of low-frequency wave impedance curves in well logging. The modeling process combines geological sedimentary characteristics to select interpolation methods such as parallel to the top, parallel to the bottom, and equal division.

[0017] S2-2: Extract the seismic wavelet of the target segment, and use the sparse pulse inversion method to perform inversion iteration on the initial wave impedance inversion model. After the inversion is completed, the wave impedance inversion results of medium and low frequency are obtained.

[0018] S3: High-frequency component inversion:

[0019] S3-1: The mid-to-low frequency wave impedance inversion results are convolved with the seismic wavelet of the target layer to obtain the forward seismic data of the mid-to-low frequency wave impedance inversion results.

[0020] S3-2: Subtract the actual seismic data from the forward modeling seismic data obtained by inverting the impedance of medium and low frequency waves to obtain the residual seismic disturbance data;

[0021] S3-3: Perform waveform classification on the residual seismic disturbance data. The number of waveform classification categories is equal to the number of actual wells drilled in the study area, and obtain the waveform classification results.

[0022] S3-4: Match the waveform classification result of each trace with the actual well category. If the waveform category is the same as a certain actual well category, assign the high-frequency impedance curve of that well to that seismic trace to obtain the high-frequency impedance inversion result.

[0023] S4: The impedance inversion results of low- and medium-frequency waves are added and fused with the impedance inversion results of high-frequency waves, and then smoothed to output high-resolution seismic inversion results across the entire frequency band.

[0024] Compared with the prior art, the present invention can achieve the following positive technical effects:

[0025] 1) Compared with model inversion and geostatistical inversion, which obtain high-frequency information through well logging interpolation or stochastic simulation, this invention uses seismic data to drive the high-frequency information inversion process, which significantly improves the lateral resolution and accuracy of the inversion results.

[0026] 2) Compared with waveform indication inversion and neural network inversion, this invention uses the residual seismic disturbance to drive the high-frequency component inversion by subtracting the forward modeling data of the low-frequency inversion results from the actual seismic data. This achieves independent inversion of high-frequency components and low-frequency components without affecting each other, making the low-frequency and high-frequency information in the inversion results more accurate.

[0027] 3) A full-band high-precision seismic inversion method is provided. This method divides seismic inversion into low-to-medium frequency inversion and high-frequency inversion, both of which are driven by seismic data and have a reliable theoretical basis, greatly improving the resolution and accuracy of seismic inversion. The inversion results have the advantages of high resolution and high consistency with well logging information, and can characterize the actual lateral changes in the formation, which is helpful for the accurate identification of thin reservoirs in oil and gas field exploration and development. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the present invention;

[0029] Figure 2This is a series of well-connected seismic profiles of the study area;

[0030] Figure 3 This is a profile of the impedance inversion results for the mid-to-low frequency waves in the study area;

[0031] Figure 4 This is a profile of the high-frequency wave impedance inversion results for the study area;

[0032] Figure 5 This is a profile of the full-band wave impedance inversion results for the study area. Detailed Implementation

[0033] Example 1

[0034] A full-band high-resolution seismic inversion method, the specific steps of which are as follows:

[0035] Step 1: Perform well-seismic calibration based on the depth-domain logging impedance curves to obtain the time-domain impedance curves of the target layer. n is the sequence number of the actual drilled well, and t is the time corresponding to the seismic trace sampling point. The highest effective frequency is determined based on the seismic data. Low-pass and high-pass filters were used respectively to obtain the impedance of the mid-to-low frequency waves. and high-frequency impedance curve ;

[0036] Step 2, inversion of mid-to-low frequency components, frequency band from 0 Hz to... hertz;

[0037] Step 2-1: Based on the geological and sedimentary characteristics of the target layer, select the interpolation method of parallel to the top, parallel to the bottom, or equal division, and interpolate according to the low-frequency impedance curve in the well logging for the target layer. Establish an initial model for wave impedance inversion of the target segment. ;

[0038] Step 2-2, Input earthquake data Extracting seismic wavelets from the target segment The model inversion method is used to perform inversion iterations on the initial model of wave impedance inversion, and the minimization of the L1 norm is used as the termination condition for the inversion iteration.

[0039]

[0040]

[0041] in, The reflection coefficient is calculated from the wave impedance. * represents the convolution operation, used for the initial model of wave impedance inversion. After iteration, the wave impedance inversion results for the mid-to-low frequencies are obtained. ;

[0042] Step 3, high-frequency component inversion, bandwidth greater than The range of Hertz;

[0043] Step 3-1: Invert the impedance results of the mid-to-low frequency waves. Seismic wavelet of the target layer Perform forward modeling of the fold:

[0044]

[0045]

[0046] in, Impedance inversion results for mid-to-low frequency waves The calculated reflection coefficient, where * represents the convolution operation. Forward modeling seismic data for impedance inversion results of medium and low frequency waves;

[0047] Step 3-2: Convert the forward-modeled seismic data from the impedance inversion results of the medium and low frequency waves. Compared with actual earthquake data Subtraction yields residual seismic disturbance data ;

[0048]

[0049] Step 3-3: Transfer the residual seismic disturbance data Waveform classification is performed, assuming there are N drilled wells in the work area, with each well representing a separate waveform category. The number of waveform categories is N, resulting in a waveform classification plane. ;

[0050] Steps 3-4: Match the waveform classification result of each channel with the actual drilled well category. If the waveform category of a channel matches a certain actual drilled well category, then the high-frequency impedance curve of that well is... The values ​​were assigned to the seismic trace to obtain the high-frequency impedance inversion results. ;

[0051]

[0052] Step 4: Invert the impedance results of the mid-to-low frequency waves. With high-frequency impedance inversion results Perform addition and fusion, and with The time window length is smoothed to output full-band high-resolution seismic inversion results. .

[0053]

[0054]

[0055] Example 2

[0056] Figures 2 to 5 This is an example of full-band high-precision seismic impedance inversion based on well logging and seismic data from a study area in the Sichuan Basin. Figure 2 This is a well-connected seismic profile of the study area. Figure 3 The impedance inversion results for mid-to-low frequency waves after applying this invention, Figure 4 The high-frequency impedance inversion results after applying this invention, Figure 5 The results of full-band high-precision wave impedance inversion after applying the present invention show that the seismic inversion results after applying the present invention have the advantages of high longitudinal and lateral resolution, good continuity, and high degree of agreement with well logging curves. The inversion results are driven by seismic data and can characterize the real longitudinal and lateral changes of the strata, and the ability to identify thin layers is significantly improved.

[0057] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A full-band high-precision seismic inversion method comprising the specific steps of: Step 1, well seismic calibration based on depth domain logging wave impedance curve to obtain time domain target interval wave impedance curve , n is the serial number of the drilled well, t is the time corresponding to the sampling point of the seismic trace, and the highest effective frequency is determined according to the seismic data , low-pass filtering and high-pass filtering are respectively used to obtain low-frequency wave impedance and high-frequency wave impedance curve ; Step 2, mid-low frequency component inversion, frequency band range 0 hertz to hertz; Step 2-1, select parallel to top, parallel to bottom, equal division interpolation mode according to the geological deposition characteristics of the target layer, and interpolate the low-frequency wave impedance curve in the well log for the target layer Establish the wave impedance inversion initial model of the target layer ; Step 2-2, input seismic data extracting a seismic wavelet of a target interval , using a model inversion method to iteratively invert the initial model of wave impedance, and taking the minimum L1 norm as the termination condition of the inversion iteration, ； ； Wherein, The reflection coefficient calculated by the wave impedance, represents the convolution operation on the initial model of the wave impedance inversion After the iteration is completed, the wave impedance inversion result of the medium-low frequency is obtained ; Step 3, high frequency component inversion, frequency band greater than Hertz range; Step 3-1, impedance inversion result of low and medium frequency wave seismic wavelet of the target layer convolution forward: ； ； wherein, is the low-frequency wave impedance inversion result The calculated reflection coefficient represents the convolution operation, is the forward seismic data of the low-frequency wave impedance inversion result; Step 3-2, forward seismic data of the low frequency wave impedance inversion result Subtracting the actual seismic data to obtain residual seismic disturbance data ; ； Step 3-3, residual seismic disturbance data is removed Waveform classification is performed, assuming that there are N actual drilled wells in the work area, each well is a separate waveform class, and the number of waveform classification classes is N, and a waveform classification plane result is obtained ; Step 3-4: match the class of the waveform classification result of each trace with the class of the real drilled well, and if the class of the waveform of the trace is the same as the class of the real drilled well, then assign the high-frequency wave impedance curve of the well to the trace to the seismic trace, and obtain the high-frequency wave impedance inversion result ; ； Step 4, add the low-frequency wave impedance inversion result and the high-frequency wave impedance inversion result together, and perform smoothing processing with a time window length of to output the full-band high-resolution seismic inversion result ; ； 。

Images