A method for frequency conversion of well-seismic combined seismic data

Abstract

This invention discloses a well-seismic combined seismic data overlay method, relating to the field of geophysical exploration processing and interpretation technology in oil and gas field exploration. The method includes the following steps: S1, acquiring acoustic and density logging data, calculating the wave impedance curve and converting it to reflection coefficients, selecting a wavelet close to the seismic data, convolving it with the reflection coefficients to form a conventional synthetic record, and performing well-seismic calibration; S2, determining the target wavelet for overlay, and convolving it with the reflection coefficients to form a broadband synthetic record; S3, combining well-side seismic data, conventional synthetic records, and broadband synthetic records to construct a sample set, and balancing the number of samples; S4, expanding the sample set through mutation; S5, establishing a nonlinear relationship for overlay based on the sample set; S6, applying the nonlinear relationship to conventional seismic data to obtain high-resolution seismic data. This invention utilizes the advantages of wide bandwidth of well logging data and seismic frequency bands to enhance high-frequency seismic data and broaden the seismic frequency band, thereby improving seismic resolution.

Description

Technical Field

[0001] This invention relates to the field of geophysical exploration processing and interpretation technology for oil and gas field exploration, and more specifically to a method for frequency extension of seismic data combining well and seismic data. Background Technology

[0002] In oil and gas seismic exploration, seismic data is widely used to estimate structural conditions, stratigraphic thickness, and reservoir development. As oil and gas exploration progresses, reservoir thickness becomes increasingly thinner, with many areas having thicknesses below 5 meters, making current seismic data resolution often insufficient. Conventional seismic data processing often limits the dominant frequency and bandwidth of the data to facilitate interpretation, reduce interpretation difficulty, and simplify stratigraphic models, thus ensuring a high signal-to-noise ratio in the dominant frequency bands. Furthermore, differences in processing levels lead to the loss or distortion of high-frequency information during processing, resulting in low overall resolution, especially with the amount of high-frequency information far lower than the actual amount of data acquired. In reality, due to the increased propagation depth of seismic waves, high-frequency noise energy exceeds the effective signal energy, yet it is still present in the acquired data. Therefore, restoring high-frequency signal energy, broadening the effective frequency band, and improving seismic resolution are key tasks in the field of seismic exploration.

[0003] In the development of seismic exploration technology over the past few decades, researchers have continuously studied high-fidelity, high-resolution processing techniques to enhance the usability of seismic data. Various high-resolution processing methods for seismic data have been proposed, including single-mode deconvolution and signal decomposition and reconstruction techniques; well-constrained and well-free methods; and frequency extension techniques using neural networks as the main algorithm. For well-constrained methods, it is necessary to establish the relationship between conventional seismic data and well data to enhance information in frequency bands beyond the dominant frequency bands of the seismic data, while maintaining good correlation with wells. However, when used in a specific block, this well-constrained processing technique may have few constrained wells and is greatly affected by sample quality. Poor-quality samples will reduce the fitting and prediction effects of the mapping algorithm; therefore, this is an aspect that urgently needs improvement. In the prior art, the applicant's patent application with publication number CN103389513B discloses a method for improving seismic data resolution using constrained inversion of acoustic logging data. The core of this method lies in establishing a relationship between high-resolution synthetic records of logging data and seismic data to enhance the energy of high-frequency components in the seismic data, thereby improving the seismic data resolution. Because the high-frequency components of seismic data are submerged in noise, conventional deconvolution methods for improving resolution are often ineffective and have limited resolution-enhancing capabilities. By combining high-resolution synthetic records of logging data with seismic data, high-frequency energy can be effectively extracted.

[0004] The aforementioned patents obtain high-resolution profiles or data volumes by establishing the relationship between synthetic records and seismic records and applying it to seismic traces. However, some of the technical steps have shortcomings. For example, how to solve the problem of establishing a more reasonable relationship model between well data and seismic data, how to solve the problem of insufficient sample size, diversity, and unbiasedness, and how to solve the problem of reasonably and faithfully expanding the seismic frequency band and improving resolution. Summary of the Invention

[0005] To overcome the shortcomings of the existing technology, this invention discloses a well-seismic combined seismic data extension method. The purpose of this invention is to address how to more rationally establish a relationship model between well data and seismic data, solving the problems of insufficient sample size, diversity, and unbiasedness; and to solve the technical problem of rationally and faithfully widening the seismic bandwidth and improving resolution. This invention utilizes the advantages of wide bandwidth of well logging data and seismic bandwidth to enhance high-frequency seismic data and widen the seismic bandwidth, thereby improving seismic resolution. By acquiring acoustic and density logging data, wave impedance curves are calculated and converted into reflection coefficients. A wavelet close to the seismic data is selected and convolved with the reflection coefficient to form a conventional synthetic record, completing well-seismic calibration. The target wavelet for extension is determined and convolved with the reflection coefficient to form a broadband synthetic record. The well-side seismic data, conventional synthetic record, and broadband synthetic record are combined to form a sample, and the sample size is balanced. The existing sample set is expanded by variation. A nonlinear relationship for extension is established based on the sample set. The established relationship is applied to conventional seismic data to obtain high-resolution seismic data.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for frequency conversion of well-seismic combined seismic data includes the following steps: I. Formation of Conventional Synthetic Records S1. Acquire acoustic and density logging data, calculate the wave impedance curve and convert it into reflection coefficient, select a wavelet that is close to the seismic data, convolve it with the reflection coefficient to form a conventional synthetic record and perform well-seismic calibration. Preferably, in step S1, the wavelet includes the Lake wavelet, the wavelet extracted jointly from well logging and seismic data, the seismic statistical wavelet, the simulated wavelet, and the hypothetical wavelet.

[0007] Step S1 is a commonly used technique in the industry for creating synthetic seismic records and for calibrating well-seismic-time-depth relationships. The wavelet of the synthetic record can be the Rick wavelet, the wavelet extracted from well logging and seismic data, the seismic statistical wavelet, the simulated wavelet, the hypothetical wavelet, etc., but it should be as close as possible to the wavelet of the seismic data itself.

[0008] II. Broadband Synthesis Record Formation S2. Determine the target wavelet for frequency extension and convolve it with the reflection coefficient to form a broadband synthetic record; Preferably, in step S2, the primary frequency and bandwidth of the target wavelet are determined according to the actual stratum resolution requirements. The target wavelet has a higher primary frequency and wider bandwidth than the wavelet in step S1. The target wavelet includes the Lake wavelet, the Yu wavelet, and the bandpass wavelet.

[0009] Step S2 is used to create the target overlay wavelet for broadband synthesis recording. Its dominant frequency and bandwidth are determined according to the actual stratigraphic resolution requirements. It has a higher dominant frequency and wider bandwidth than the wavelet in step S1. It is generally an analytical wavelet such as the Lake wavelet, Yu wavelet, or bandpass wavelet.

[0010] III. Constructing Samples S3. Combine the well-side seismic data, conventional synthetic records, and broadband synthetic records to construct a sample set, and balance the number of samples. Preferably, in step S3, the step of combining well-side seismic data, conventional synthetic records, and broadband synthetic records to construct a sample set includes: for well sections with good target layer calibration results, combining well-side seismic data and broadband synthetic records to form a sample, wherein the well-side seismic data is the input sample and the broadband synthetic records are the target sample; combining conventional synthetic records and broadband synthetic records to form a sample, wherein the conventional synthetic records are the input sample and the broadband synthetic records are the target sample.

[0011] In step S3, the two types of samples are combined, which can enhance the diversity and unbiasedness of the sample set.

[0012] Preferably, in step S3, the original sample of each well is split into samples of equal length, including: setting the sample length and the splitting step size, splitting the sample well segment of each well, and finally forming a sample set of equal length.

[0013] Preferably, in step S3, the process of balancing the number of samples includes: for wells with a small number of samples, selecting samples from their existing samples to repeatedly supplement them until the number of samples from all wells in the sample set is equal; the selection rules include random sampling and equally spaced sampling.

[0014] In step S3, it is necessary to balance the sample size of different wells. Because the selected sample well segments have different lengths, the sample size varies from well to well. Well segments with fewer samples have weaker constraints when establishing relationships. Therefore, it is necessary to supplement the samples from wells with fewer samples to create balanced constraints for each well and enhance the balance of the sample set. The method is as follows: for wells with fewer samples, select some samples from their existing samples and repeat this process until the sample size of all wells in the sample set is roughly equal. The selection rule can be random sampling or equal-interval sampling.

[0015] IV. Mutation and Expansion S4. Perform mutation and expansion on the sample set to increase its size; Preferably, in the variation expansion of step S4, the original reflection coefficient of the sample composed of conventional synthetic records and broadband synthetic records is perturbed.

[0016] Preferably, in the variation expansion of step S4, the reflection coefficient perturbation includes: randomly generating some positions on the reflection coefficient sequence and generating random changes at these positions to modify the original reflection coefficient.

[0017] Preferably, in the mutation expansion of step S4, the reflection coefficient perturbation includes random scaling and conditional scaling: Random scaling: A portion of the reflection coefficient values ​​are enlarged or reduced by the coefficient, and the position and scaling ratio are random. Scaling based on conditions: Set conditions to amplify the reflection coefficient if the condition is met, and reduce the reflection coefficient if the condition is not met.

[0018] Preferably, in the mutation expansion of step S4, the mutation method is used alone or in combination to generate a new reflection coefficient. After the new reflection coefficient is generated, it is convolved with the conventional wavelet and the broadband wavelet to form a mutation sample, which is then added to the existing sample set to expand the existing sample set.

[0019] In step S4, using only actual well samples to construct the sample set results in a lack of scale and diversity, leading to poor robustness of the subsequently established frequency extension relationships. To improve this, the existing sample set is expanded through mutation. The method involves perturbing the original reflection coefficients of samples composed of conventional and broadband synthetic records. This perturbation can be random or systematic, including but not limited to the following: 1. Randomly generating positions in the reflection coefficient sequence and applying random changes at these positions to modify the original reflection coefficients; 2. Magnifying or reducing the values ​​of a portion of the reflection coefficients by a factor, with the position and scaling ratio being random; alternatively, a condition can be set so that reflection coefficients meeting the condition are magnified, while those not meeting the condition are reduced. These mutation methods can be used individually or in combination. After generating a certain number of new reflection coefficients, they are convolved with conventional and broadband wavelets to form mutated samples, which are then added to the existing sample set, thus expanding the existing sample set.

[0020] V. Establishment of Nonlinear Relationships S5. Establish nonlinear relationships for frequency extension based on sample sets; Preferably, in step S5, a nonlinear relationship of frequency extension is established using a machine learning method, which includes a neural network method for fitting the nonlinear mapping relationship between all input samples and the target sample.

[0021] In step S5, there are no restrictions on the method for establishing the nonlinear relationship of the frequency extension. Machine learning methods are a good choice, such as various neural network methods, which are used to fit the nonlinear mapping relationship between all input samples and target samples.

[0022] VI. Application of Relational Extrapolation S6. Apply nonlinear relationships to conventional seismic data to obtain high-resolution seismic data.

[0023] Preferably, in step S6, the nonlinear relationship of frequency extension established in step S5 is applied to conventional seismic data to obtain frequency-extended seismic data, which is high-resolution data.

[0024] The beneficial effects of this invention are: In this invention, two types of data are used as inputs to construct samples. One type consists of well-side seismic data and broadband synthetic records, while the other consists of conventional synthetic records and broadband synthetic records. Combining these two types of samples enhances the diversity and unbiasedness of the sample set.

[0025] In this invention, the method for balancing the number of samples after the samples are constructed involves selecting some samples from the existing samples of wells with a small number of samples and repeating the process until the number of samples in all wells in the sample set is equal. This results in a balanced constraint on the samples of each well and enhances the balance of the sample set.

[0026] In this invention, by generating new samples through a certain degree of variation in the reflection coefficient, the sample set is expanded. This method is beneficial for increasing the size and diversity of the sample set and enhancing the robustness of the subsequently established frequency extension relationship.

[0027] Compared with existing technical solutions, the sample set establishment, balancing and expansion method proposed in this invention increases the number of samples used to establish correct frequency extension relationships by several times, improves the balance, diversity and unbiasedness of the sample set, and thus improves the reliability and stability of the method.

[0028] This invention can significantly enhance the high-frequency effective signals in seismic data, increase the dominant frequency and bandwidth of the seismic data, and achieve a high correlation coefficient with well data, thereby reasonably improving resolution. With sufficient protection of high-frequency signals in conventional data, it can increase the dominant frequency of seismic data by 20-40Hz and broaden the effective bandwidth to over 100Hz. Compared to existing technologies, the sample set establishment, balancing, and expansion methods proposed in this invention multiply the number of samples used to establish correct frequency extension relationships, improving the balance and diversity of the sample set, thus enhancing the reliability and stability of the method.

[0029] This invention is applicable to situations in certain oil and gas fields where fine processing of seismic data is required. This method provides a powerful solution to the challenges of thin reservoirs and has been applied in multiple oil and gas fields and structures with good results. For this method to be applicable, at least one well needs to have high-quality logging data to meet the conditions for combined well-seismic application. Attached Figure Description

[0030] Figure 1 This is a technical flowchart of the well-seismic combined seismic data frequency extension method of the present invention; Figure 2 This invention provides a comparison between the predicted output and the expected output for a portion of the samples. Figure 3 This is a comparison of the high-resolution processing of the over-well profile before and after stacking, as presented in this invention. Detailed Implementation

[0031] The following will provide a clear and complete description of the concept, specific structure, and technical effects of the present invention in conjunction with the embodiments and accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention.

[0032] A method for frequency conversion of well-seismic combined seismic data, such as Figure 1 As shown, it includes the following steps: 1. Acquire acoustic and density logging data, calculate the wave impedance curve and convert it into reflection coefficient, select a wavelet with a frequency close to the dominant frequency of the seismic data, convolve it with the reflection coefficient to form a conventional synthetic record and complete well-seismic calibration; 2. Determine the target wavelet for frequency extension and convolve it with the reflection coefficient to form a broadband synthesized record; 3. For well sections with good target layer calibration results, combine the well-side seismic data, conventional synthetic records, and broadband synthetic records to form a sample, and balance the number of samples. 4. Expand the existing sample set by mutation; 5. Establish nonlinear relationships in frequency extension based on sample sets; 6. Apply the established relationships to regular seismic data to obtain high-resolution seismic data.

[0033] This embodiment relates to high-resolution processing of seismic data from a region in the Sichuan Basin. The dominant frequency of the conventional data for this block is 25Hz, with a predominant frequency band of 5-70Hz. The target layer and its sub-layers cannot be effectively resolved using seismic phase axes. High-resolution combined well-seismic processing was performed according to the procedure of this invention. For example... Figure 1 As shown, the specific implementation process is as follows: Step 1: Acquire acoustic and density logging data for the region, calculate the wave impedance curve and convert it into reflection coefficients. Select a wavelet with a frequency close to the dominant frequency of the seismic data, convolve it with the reflection coefficients to form a conventional synthetic record, and complete well-seismic calibration. This step utilizes commonly used synthetic seismic record production techniques and well-seismic time-depth relationship calibration methods in the industry. Based on the spectral analysis of the seismic data, the wavelet used in the synthetic record is a 25Hz zero-phase Ricker wavelet, which is close to the wavelet of the seismic data itself.

[0034] Step 2: Based on the resolution requirements of the target layer in this region, under the condition of a wavelet with a main frequency of about 55Hz, the target layer of the synthetic record can resolve a set of sublayers. Therefore, the target wavelet for the upscaling is determined to be 55Hz. Here, the zero-phase Lake wavelet is also used, and then it is convolved with the reflection coefficient to form a broadband synthetic record.

[0035] Step 3: Select 15 wells with good matching between synthetic records and well-side seismic data as sample wells, and select well segments with good calibration results for the target layer. In this example, a total of 23 well segments were selected. The well-side seismic data and broadband synthetic records are combined to form a sample, with the well-side seismic data serving as the input sample and the broadband synthetic records as the target sample. Furthermore, the conventional synthetic records and broadband synthetic records are combined to form a sample, with the conventional synthetic records serving as the input sample and the broadband synthetic records as the target sample.

[0036] Sample splitting: The sample length was set to approximately one and a half wavelet lengths, about 45 ms, and the splitting step size was set to two sample points. Then, the original samples of the selected well sections of each well were split according to this length and step size. In this example, a total of 1782 samples were formed after splitting. Since two types of input data, well-side seismic data and conventional synthetic records, were used, the total number of samples was 3564.

[0037] Sample size balance: Across all wells, the maximum number of samples that can be extracted from a single well section is 132, and the minimum is 33. Wells with fewer than 132 samples need to be supplemented to have approximately 132 samples. In this example, for each well section of each well, the difference between the existing sample size and 132 is used... Randomly select from the existing sample The samples were duplicated and added to the well sample. After balancing, the sample size increased to 23 * 132 * 2 = 6072.

[0038] Step 4: Expand the existing sample set through mutation. In this example, for samples composed of conventional and broadband synthetic records, a certain degree of perturbation is applied to their original reflection coefficients using random perturbation. Currently, there are a total of 3036 samples composed of conventional and broadband synthetic records. In this example, 10% of these are randomly selected, rounded down to 303 samples for mutation. The number of mutation points in each sample is randomly generated between 5% and 20% of the total number of sample points (45), and the mutation location is also randomly generated. The mutation amount is a random value that can be positive or negative, and its absolute value is set to 20% of the maximum absolute value of the reflection coefficient of each sample. The mutation expansion is set to 10 times for each sample, meaning that 10 mutated samples are added to each sample. After generating new reflection coefficients, they are convolved with conventional and broadband wavelets to form mutated samples, which are then added to the existing sample set, thus expanding the existing sample set. After mutation expansion, 303 * 10 = 3030 new samples are added. At this point, the sample set contains a total of 6072 + 3030 = 9102 samples, an increase of 7320 samples compared to the unexpanded sample set, and the sample size has expanded more than four times.

[0039] Step 5: Establish the nonlinear relationship of the frequency extension based on the sample set. In this example, a deep neural network was chosen to establish the relationship. The main structure of the neural network is an 8-layer one-dimensional convolutional autoencoder structure; the input length of the network is 45, which is equal to the sample length; the output length is equal to the input length. After 600 iterations, the root mean square error of the training reached 0.00024, and the test error reached 0.003. The predicted curve closely matches the expected curve with a high degree of accuracy and small error. Figure 2 As shown, this indicates that the network has good predictive capabilities.

[0040] Step Six: Apply the established relationships to conventional seismic data to obtain high-resolution seismic data. The processed data's dominant frequency is increased from approximately 25Hz to 55Hz, and its advantageous bandwidth is expanded from 65Hz to approximately 95Hz, resulting in an overall resolution improvement of more than 100%.

[0041] Effect Analysis: Taking a profile of an unconstrained area in this region as an example, the results are compared with conventional data (see attached). Figure 3 a) High-resolution data of the sample set without augmentation (attached) Figure 3 b) and high-resolution data post-processed according to the invention after expanding the sample set (see appendix) Figure 3 c), to illustrate the improvement effect of expanding the sample set. First, high-resolution profile attachment Figure 3 c. Comparison with conventional cross-sections (attached) Figure 3 a) In comparison, while maintaining the original wave group energy relationships, the resolution is significantly improved. The boundary of the sub-segment of Long-1 above the bottom boundary of the fifth peak of the target layer is more obvious, and the fault breakpoint is clearer. Secondly, compared with the attached... Figure 3 b and appendix Figure 3 c. The energy relationships of the processed results without expanded sample sets differ significantly from those of conventional data: large sections of weak energy exist in the middle of the profile, resulting in a low signal-to-noise ratio; large clusters of strong energy reflections are present above the profile, leading to significant lateral contrast, which does not conform to actual geological understanding; and reasonable waveforms cannot be correctly extrapolated and predicted near the target layer, especially in areas without sample well constraints. In contrast, the profile processed with expanded sample sets maintains the same energy relationships, has a high signal-to-noise ratio, uniform lateral energy, and a smooth transition, making it a profile with better interpretability, and performing even better in areas with well constraints. Therefore, expanding the sample set can significantly improve its balance and diversity, especially the robustness and generalization ability of the algorithm establishing the nonlinear relationship of the frequency extension in step five, thereby enhancing the reliability and stability of the overall technical solution and optimizing the effect of this combined well-seismic high-resolution processing method.

[0042] The embodiments of the present invention have been described in detail above, but the present invention is not limited to the described embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention, and these equivalents or substitutions are all included within the scope defined by the claims of the present invention.

Claims

1. A method for frequency conversion of well-seismic combined seismic data, characterized in that, Includes the following steps: S1. Acquire acoustic and density logging data, calculate the wave impedance curve and convert it into reflection coefficient, select a wavelet that is close to the seismic data, convolve it with the reflection coefficient to form a conventional synthetic record and perform well-seismic calibration. S2. Determine the target wavelet for frequency extension and convolve it with the reflection coefficient to form a broadband synthetic record; S3. Combine the well-side seismic data, conventional synthetic records, and broadband synthetic records to construct a sample set, and balance the number of samples. S4. Perform mutation and expansion on the sample set to increase its size; S5. Establish nonlinear relationships for frequency extension based on sample sets; S6. Apply nonlinear relationships to conventional seismic data to obtain high-resolution seismic data.

2. The well-seismic combined seismic data overlay method as described in claim 1, characterized in that, In step S1, the wavelet includes the Lake wavelet, the wavelet extracted jointly from well logging and seismic data, the seismic statistical wavelet, the simulated wavelet, and the hypothetical wavelet.

3. The well-seismic combined seismic data frequency extension method as described in claim 1, characterized in that, In step S2, the primary frequency and bandwidth of the target wavelet are determined according to the actual stratum resolution requirements. The target wavelet has a higher primary frequency and wider bandwidth than the wavelet in step S1. The target wavelet includes the Lake wavelet, the Yu wavelet, and the bandpass wavelet.

4. The well-seismic combined seismic data overlay method as described in claim 1, characterized in that, In step S3, the process of combining well-side seismic data, conventional synthetic records, and broadband synthetic records to construct a sample set includes: for well sections with good target layer calibration results, combining well-side seismic data and broadband synthetic records to form a sample, wherein the well-side seismic data is the input sample and the broadband synthetic records are the target sample; combining conventional synthetic records and broadband synthetic records to form a sample, wherein the conventional synthetic records are the input sample and the broadband synthetic records are the target sample.

5. The well-seismic combined seismic data overlay method as described in claim 4, characterized in that, In step S3, the original sample of each well is split into samples of equal length, including: setting the sample length and the splitting step size, splitting the sample well segment of each well, and finally forming a sample set of equal length.

6. The well-seismic combined seismic data overlay method as described in claim 4, characterized in that, In step S3, the process of balancing the number of samples includes: for wells with a small number of samples, selecting samples from their existing samples to repeatedly supplement them until the number of samples from all wells in the sample set is equal; the selection rules include random sampling and equally spaced sampling.

7. The method for frequency spreading of well-seismic combined seismic data as described in claim 1, characterized in that, In the variation expansion of step S4, the original reflection coefficients of the samples composed of conventional synthetic records and broadband synthetic records are perturbed.

8. The well-seismic combined seismic data overlay method as described in claim 7, characterized in that, In the mutation expansion of step S4, the perturbation of the reflection coefficient includes: randomly generating some positions on the reflection coefficient sequence and generating random changes at these positions to modify the original reflection coefficient.

9. The method for frequency spreading of well-seismic combined seismic data as described in claim 7, characterized in that, In the mutation expansion of step S4, the reflection coefficient perturbation includes random scaling and conditional scaling: Random scaling: A portion of the reflection coefficient values ​​are enlarged or reduced by the coefficient, and the position and scaling ratio are random. Scaling based on conditions: Set conditions to amplify the reflection coefficient if the condition is met, and reduce the reflection coefficient if the condition is not met.

10. The well-seismic combined seismic data overlay method as described in claim 7, characterized in that, In the mutation expansion of step S4, the mutation method is used alone or in combination to generate new reflection coefficients. These new coefficients are then convolved with conventional wavelets and broadband wavelets to form mutated samples, which are then added to the existing sample set to expand the existing sample set.

11. The well-seismic combined seismic data overlay method as described in claim 1, characterized in that, In step S5, a nonlinear relationship for frequency extension is established using machine learning methods, including neural network methods, which are used to fit the nonlinear mapping relationship between all input samples and target samples.

12. The well-seismic combined seismic data overlay method as described in claim 1, characterized in that, In step S6, the nonlinear relationship established in step S5 is applied to conventional seismic data to obtain the seismic data after frequency extension, which is high-resolution data.

Images