Intelligent strong noise suppression method for deep seismic data
By employing a dual-channel parallel analysis and feature competition constraint method, the problem of accurately distinguishing noise from signal in deep seismic signals was solved, improving the noise suppression accuracy and signal fidelity of deep seismic data, and meeting the requirements of high-precision deep resource exploration.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INST OF EARTHQUAKE CHINA EARTHQUAKE ADMINISTRATION
- Filing Date
- 2026-04-09
- Publication Date
- 2026-06-23
AI Technical Summary
Existing deep learning methods cannot accurately distinguish between noise and effective signals in deep, weak signal scenarios, resulting in the discontinuity of the phase axis of reflected waves from deep targets and amplitude distortion, which makes it difficult to meet the technical requirements of high-precision deep resource exploration.
A dual-channel parallel analysis was used to extract global noise distribution features and local signal morphology features. A competitive suppression factor was generated through feature competition constraints. Based on the competitive suppression factor, local signal morphology features were weighted and screened and signals were reconstructed. Combined with data partitioning and signal confidence leveling, the signal reconstruction process of seismic data was gradually optimized.
It has improved the accuracy of noise suppression in deep seismic data, significantly improved the signal fidelity and continuity of reflection signals in deep seismic data, and met the needs of high-precision deep resource exploration.
Smart Images
Figure CN122260477A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of geophysical exploration technology, and in particular to a method for intelligent suppression of strong noise in deep seismic data. Background Technology
[0002] With the increasing demand for exploration of deep oil and gas reservoirs and deep mineral resources, reflection seismic data processing has become a core component of geophysical exploration. However, deep seismic signals experience significant energy attenuation during long-path propagation, exhibiting characteristics of weak amplitude and narrow frequency band. Furthermore, they are easily contaminated with strong noise during acquisition, such as environmental random noise, controllable source harmonic noise, and near-surface scattering noise. The effectiveness of noise suppression directly determines the usability of exploration data.
[0003] Current deep learning-based seismic data noise suppression techniques take noisy seismic data as direct input and learn the mapping relationship between noise and effective signal through an end-to-end black-box network model, demonstrating a certain processing efficiency in conventional seismic data denoising.
[0004] However, existing deep learning methods have significant drawbacks in deep, weak signal scenarios. The feature extraction process prioritizes learning and fitting strong noise features dominated by energy, making it impossible to accurately distinguish between low-energy effective reflection signals and high-energy noise. Consequently, during noise suppression, deep, weak effective signals are mistakenly identified as noise and removed or smoothed, leading to discontinuity of the phase axis of deep target reflection waves and amplitude distortion. This severely reduces the signal fidelity of deep seismic data and makes it difficult to meet the technical requirements of high-precision deep resource exploration. Summary of the Invention
[0005] This invention provides a method for intelligent suppression of strong noise in deep seismic data to solve the problems mentioned in the background art.
[0006] To achieve the above objectives, the present invention provides a method for intelligent suppression of strong noise in deep seismic data, comprising: S1. Obtain raw deep seismic data, and obtain noisy seismic data volume after preprocessing; S2. Perform dual-channel parallel analysis on the noisy seismic data volume to extract global noise distribution characteristics and local signal morphology characteristics, respectively; S3. Based on the feature difference between global noise distribution characteristics and local signal morphology characteristics, feature competition constraints are performed, and competition suppression factors are generated based on the constraint results; S4. Based on the competitive inhibition factor, the local signal morphological features are weighted and screened to retain the signal components with high confidence, thus obtaining the enhanced information body after competitive inhibition; S5. Perform signal reconstruction on the enhanced information volume to obtain seismic data after preliminary noise suppression; S6. Use the difference between the pre-suppressed noise seismic data and the noisy seismic data volume to reverse correct the parameters in S3 or S5. Repeat S3 or S5 and its subsequent steps until convergence is achieved to obtain the final noise-suppressed deep seismic data.
[0007] Preferably, in step S1, the raw deep seismic data is acquired and preprocessed to obtain a noisy seismic data volume, including: The raw deep seismic data is acquired, and the raw deep seismic data is filtered to remove missing and bad traces, so as to obtain a complete seismic record collection. Energy equalization processing is performed on the seismic record gathers to compensate for the energy attenuation of deep signals, resulting in an energy-equalized seismic data volume, which serves as the noisy seismic data volume.
[0008] Preferably, in step S2, a dual-channel parallel analysis is performed on the noisy seismic data volume to extract global noise distribution features and local signal morphology features, including: The noisy seismic data volume is partitioned into multiple data sub-regions; For each data sub-region, the first analysis channel and the second analysis channel are executed in parallel. The first analysis channel is used to extract the local noise energy distribution of the data sub-region, and the second analysis channel is used to extract the local signal in-phase axis shape of the data sub-region. The local noise energy distribution of each data sub-region is fused to form a global noise distribution feature, and the local signal phase axis morphology of each data sub-region is spliced together to form a local signal morphology feature.
[0009] Preferably, the parallel execution of the first analysis channel and the second analysis channel for each data sub-region includes: Within each data sub-region, the first analysis channel is executed to statistically analyze the energy distribution within the data sub-region along the time direction. Based on the level of energy distribution, the data sub-region is divided into a strong noise region and a weak noise region. Within each data sub-region, the second analysis channel is executed to trace the direction of the in-phase axis in the spatial direction and extract the continuity characteristics and amplitude variation characteristics of the in-phase axis. The division results of strong noise region and weak noise region are synchronously transmitted to the second analysis channel. The complete in-phase axis tracking results are retained in the weak noise region. In the strong noise region, the in-phase axis in the strong noise region is extended and fitted according to the in-phase axis direction of the adjacent weak noise region to obtain the corrected local signal in-phase axis shape.
[0010] Preferably, in step S3, feature competition constraints are performed based on the feature difference between global noise distribution characteristics and local signal morphological characteristics, and a competition suppression factor is generated based on the constraint results, including: At each spatial location of the noisy seismic data volume, the global noise distribution characteristics are compared with the local signal morphology characteristics at that location to determine the relative strength of their energy, which is used as the feature difference degree. Based on the characteristic difference degree, a competitive constraint is imposed on the global noise distribution characteristics and local signal morphology characteristics. The competitive constraint is as follows: when the relative strength of energy indicates that noise is dominant, noise characteristics are dominant; when the relative strength of energy indicates that signal is dominant, signal characteristics are dominant. Based on the dominant force at each position after the competition constraint, a competition inhibition factor is generated. The competition inhibition factor is used to identify the energy subject that is ultimately retained at each position.
[0011] Preferably, generating competition inhibition factors based on the dominant party at each position after competition constraints includes: After obtaining the dominant party identifiers at each position under competition constraints, a preliminary competition inhibition factor diagram is formed; Spatial continuity filtering is applied to the preliminary competition inhibition factor map to correct isolated dominant party identifiers to be consistent with the dominant parties in most of their surrounding locations. The revised competitive inhibition factor map is used as the final competitive inhibition factor.
[0012] Preferably, in step S4, the local signal morphological features are weighted and filtered according to a competitive inhibition factor to retain high-confidence signal components, resulting in an enhanced information body after competitive inhibition, including: Obtain the competition inhibition factor, which contains the energy principal identifier for each spatial location; Based on the energy subject identifier, a weighted filtering is performed on the local signal morphology features. The original signal components are retained where the energy subject identifier indicates that the signal is dominant, and the signal components are attenuated where the energy subject identifier indicates that the noise is dominant. The signal components at each position after weighted filtering are combined to form an enhanced information body after competition suppression.
[0013] Preferably, the step of performing weighted filtering on local signal morphology features based on energy entity identifiers includes: The energy entity identifier is decomposed into signal confidence levels, which are used to characterize the credibility of the signal components at each location. Based on the signal confidence level, a weighting coefficient is determined for each location, and the weighting coefficient is positively correlated with the signal confidence level. The weighting coefficients are multiplied by the signal components corresponding to each position in the local signal morphology features to obtain the weighted and filtered signal components.
[0014] Preferably, step S5 involves signal reconstruction of the enhanced information volume to obtain seismic data after preliminary noise suppression, including: The enhanced information body after competition suppression is obtained, which contains signal components after weighted filtering from multiple spatial locations; Signal domain mapping is performed on the enhanced information volume, and the weighted and filtered signal components at each spatial location are transformed from feature expression form to seismic data expression form to obtain the reconstructed seismic gather; The reconstructed seismic gathers are integrated to form seismic data after preliminary noise suppression.
[0015] Preferably, in step S6, the parameters in S3 or S5 are corrected in reverse using the difference between the pre-suppressed noise seismic data and the noisy seismic data volume. Steps S3 or S5 and subsequent steps are repeated until convergence is achieved, resulting in the final noise-suppressed deep seismic data, including: The difference volume between the pre-suppressed noise seismic data and the noisy seismic data volume is obtained. The difference volume reflects the amount of information remaining at each spatial location. Based on the distribution of residual information in the differential volume, determine the adjustment direction of feature competition constraints in S3 or the adjustment direction of signal reconstruction in S5, and correct the corresponding parameters according to the adjustment direction. Substitute the corrected parameters into S3 or S5, and repeat S3 or S5 and its subsequent steps until the differential volume meets the convergence condition, and output the final deep seismic data after suppressing noise.
[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. By using dual-channel parallel analysis, the global noise distribution characteristics and local signal morphology characteristics are accurately separated. Combined with feature competition constraints, a competition suppression factor is generated, which can clarify the energy dominance relationship between signals and noise at each spatial location. This prevents deep weak reflection signals from being submerged by strong noise characteristics from the root. While efficiently suppressing strong noise, it fully preserves and enhances the energy and morphology of deep weak signals, significantly improving the accuracy and signal fidelity of noise suppression in deep seismic data.
[0017] 2. By combining parallel processing of data partitioning with the synergistic effect of phase axis extension fitting in high-noise areas, the problem of broken signal phase axes in high-noise areas can be repaired. With the addition of signal confidence level weighted screening, high-confidence signal components are adaptively retained, further improving the continuity of deep reflection signal phase axes and optimizing the integrity and processing stability of seismic data signal reconstruction. Attached Figure Description
[0018] Figure 1 This is a flowchart illustrating a method for intelligent suppression of strong noise in deep seismic data according to an embodiment of the present invention.
[0019] Figure 2 This is a flowchart illustrating a method for generating competition factors according to an embodiment of the present invention. Detailed Implementation
[0020] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0022] Reference Figure 1 As shown, in this embodiment, the intelligent noise suppression method for deep seismic data includes: S1. Obtain raw deep seismic data, and obtain noisy seismic data volume after preprocessing; It should be noted that deep seismic raw data suffers from two major problems: data incompleteness due to missing or bad traces and significant attenuation of deep signal energy. Incomplete data will cause distortion in subsequent feature extraction, and uneven energy will amplify noise and weaken the effective signal, directly resulting in insufficient noise suppression accuracy and poor signal fidelity.
[0023] Therefore, in this embodiment of the invention, S1 involves acquiring raw deep seismic data, preprocessing it to obtain a noisy seismic data volume, including: The raw deep seismic data is acquired, and the raw deep seismic data is filtered to remove missing and bad traces, so as to obtain a complete seismic record collection. Energy equalization processing is performed on the seismic record gathers to compensate for the energy attenuation of deep signals, resulting in an energy-equalized seismic data volume, which serves as the noisy seismic data volume.
[0024] Specifically, for example, in a deep oil and gas exploration area, the seismic detector array deployed by the three-dimensional seismic observation system is used to collect raw deep seismic data of the oil and gas-bearing strata in the area. This raw data contains multiple seismic signal records.
[0025] Then, the acquired raw deep seismic data is screened, and the integrity of each seismic signal record is checked one by one. If a seismic signal record has missing data or signal distortion that exceeds the preset normal range, the record is determined to be a missing or bad record and is removed. Finally, a seismic record set composed of complete seismic signal records is obtained.
[0026] Finally, energy equalization processing is performed on the obtained seismic record gathers, and the energy distribution characteristics of each signal in the seismic record gathers are statistically analyzed along the time direction.
[0027] If the energy value of a certain signal is lower than the normal energy benchmark of the deep signal in the work area, energy compensation is performed on the signal to gradually increase its energy to the benchmark level. After the energy equalization process is completed, the energy-equalized seismic data volume is obtained, which is the noisy seismic data volume.
[0028] In summary, this approach eliminates processing biases caused by incomplete data and energy imbalance at the source, providing highly reliable input for subsequent dual-channel feature analysis and feature competition constraints. It fully preserves the deep effective signal substrate, avoids signal loss in the preprocessing stage, reduces the intensity of initial noise interference, and significantly improves the efficiency, accuracy, and signal fidelity of subsequent intelligent noise suppression, ensuring the stable operation of the overall method.
[0029] S2. Perform dual-channel parallel analysis on the noisy seismic data volume to extract global noise distribution characteristics and local signal morphology characteristics, respectively.
[0030] However, in the high-noise environment of deep earthquakes, the global noise distribution and local signal morphology cannot be extracted synchronously and accurately, resulting in low efficiency of single-channel analysis. High-noise areas are prone to causing breakage of the in-phase axis tracking and morphological distortion. Partition processing is prone to severing the spatial correlation of data, making it difficult to meet the dual requirements of globalizing noise features and localizing signal features.
[0031] Therefore, this scheme provides a preferred implementation method in which S2 performs dual-channel parallel analysis on the noisy seismic data volume to extract global noise distribution characteristics and local signal morphology characteristics, including: The noisy seismic data volume is partitioned into multiple data sub-regions; For each data sub-region, the first analysis channel and the second analysis channel are executed in parallel. The first analysis channel is used to extract the local noise energy distribution of the data sub-region, and the second analysis channel is used to extract the local signal in-phase axis shape of the data sub-region. The local noise energy distribution of each data sub-region is fused to form a global noise distribution feature, and the local signal phase axis morphology of each data sub-region is spliced together to form a local signal morphology feature.
[0032] Specifically, in a deep oil and gas exploration area, the three-dimensional seismic observation grid of the area is used as the basis for division. The noisy seismic data volume after energy equalization is divided into spatial grids. The entire data volume is divided into multiple continuous and non-overlapping rectangular data sub-areas according to the horizontal and vertical exploration line spacing. Each data sub-area corresponds to an independent exploration block under the surface of the area.
[0033] For example, based on the 50-meter spacing between the transverse and longitudinal exploration lines in the work area, the noisy seismic data volume is divided into dozens of adjacent data sub-regions, with each data sub-region covering all seismic gather data within the corresponding exploration block.
[0034] Furthermore, within each defined data sub-region, the first and second analysis channels are simultaneously activated. The first analysis channel performs local noise energy distribution extraction on all seismic trace data within the sub-region, while the second analysis channel performs local signal phase axis morphology extraction on all seismic trace data within the sub-region.
[0035] For example, if a certain data sub-area corresponds to Block A underground in the work area, the first analysis channel counts the energy values of each seismic trace in the sub-area, and the second analysis channel tracks the in-phase axis direction of the seismic signals in the sub-area. The two channels are executed simultaneously, and data interaction only occurs in the final result integration stage.
[0036] Finally, after completing the dual-channel analysis of all data sub-regions, the local noise energy distribution data obtained from each data sub-region are spliced and integrated according to their original spatial relationships to form a complete energy distribution map covering the entire underground area of the work area. This map is the global noise distribution characteristic.
[0037] The local signal phase axis morphology data obtained from each data sub-region are then spliced together according to their original spatial positional relationship to form a set of phase axis morphology covering the entire underground area of the work area. This set is the local signal morphology feature.
[0038] In this embodiment of the invention, for each data sub-region, the first analysis channel and the second analysis channel are executed in parallel, including: Within each data sub-region, the first analysis channel is executed to statistically analyze the energy distribution within the data sub-region along the time direction. Based on the level of energy distribution, the data sub-region is divided into a strong noise region and a weak noise region. Within each data sub-region, the second analysis channel is executed to trace the direction of the in-phase axis in the spatial direction and extract the continuity characteristics and amplitude variation characteristics of the in-phase axis. The division results of strong noise region and weak noise region are synchronously transmitted to the second analysis channel. The complete in-phase axis tracking results are retained in the weak noise region. In the strong noise region, the in-phase axis in the strong noise region is extended and fitted according to the in-phase axis direction of the adjacent weak noise region to obtain the corrected local signal in-phase axis shape.
[0039] Specifically, within each data sub-region, the first analysis channel statistically analyzes the energy values of each seismic trace at different sampling points along the time direction.
[0040] For example, if the energy value at a certain time sampling point is higher than the preset noise energy benchmark, the seismic trace where the sampling point is located is determined to be a strong noise associated trace; if the energy value is lower than or equal to the benchmark, it is determined to be a weak noise associated trace.
[0041] The distribution range of strong noise associated channels within each sub-region is statistically analyzed. The part with a continuous distribution range that covers most of the sub-region is classified as a strong noise region, and the remaining area is classified as a weak noise region. For example, if the energy value of most sampling points in a data sub-region is higher than the baseline, then most of the sub-region is a strong noise region, and a small part of the edge is a weak noise region.
[0042] Furthermore, within each data sub-region, the second analysis channel traces the direction of the phase axis of the seismic signal along the spatial direction one by one. If the direction of the phase axis changes continuously between adjacent seismic channels, the continuity of the phase axis is determined to be good. If the direction changes abruptly beyond the preset continuity judgment benchmark, the continuity is determined to be poor.
[0043] Simultaneously, the amplitude values of the in-phase axis on each seismic trace are recorded, the trend of amplitude change is statistically analyzed, and the characteristics of amplitude change are extracted. For example, if the in-phase axis in a certain data sub-region has a basically consistent direction from the first trace to the tenth trace, and the amplitude decreases slowly, then the in-phase axis has good continuity and the amplitude shows a slow decay change characteristic.
[0044] Finally, the strong noise region and weak noise region division results obtained from the first analysis channel are synchronously transmitted to the second analysis channel. The second analysis channel directly retains the original in-phase axis tracking results in the weak noise region and in the strong noise region.
[0045] The in-phase axis orientation of the adjacent weak noise zone is selected as a reference. The in-phase axis in the strong noise zone is extended along the reference orientation. If the extended in-phase axis matches the amplitude characteristics of the seismic signal in the strong noise zone, the shape of the extended in-phase axis is determined, the correction of the in-phase axis in the strong noise zone is completed, and the corrected local signal in-phase axis shape is obtained.
[0046] For example, if the in-phase axis of a weak noise region A adjacent to a strong noise region extends laterally, then the in-phase axis is extended within the strong noise region with this direction as a reference. If the amplitude after extension matches the signal characteristics within the strong noise region, then the correction mode is determined.
[0047] Overall, this solution employs data partitioning and dual-channel parallel analysis to significantly improve feature extraction efficiency. Partition fusion takes into account both global and local data attributes. Noise partitioning-guided signal correction solves the problem of distortion and breakage of the in-phase axis in strong noise areas, ensuring the integrity and continuity of the signal shape. Accurate separation of global noise and local signal features provides high-quality input for subsequent feature competition constraints, adapting to deep, strong noise scenarios and significantly improving the accuracy of noise suppression and signal fidelity.
[0048] S3. Based on the feature difference between global noise distribution characteristics and local signal morphology characteristics, feature competition constraints are performed, and competition suppression factors are generated based on the constraint results.
[0049] In practice, under the strong noise environment of deep earthquakes, global noise and local signal characteristic energy are intertwined and mixed, making it difficult to accurately determine the dominance of signals / noise at each location. This can easily lead to the false suppression of effective signals and residual noise. Point-by-point feature constraints can easily produce isolated anomaly markers, lacking spatial continuity, resulting in distorted feature differentiation and directly reducing the accuracy of subsequent noise suppression.
[0050] To solve the above problems, refer to Figure 2 As shown, S3 performs feature competition constraints based on the feature difference between global noise distribution characteristics and local signal morphological characteristics, and generates competition suppression factors based on the constraint results, including: At each spatial location of the noisy seismic data volume, the global noise distribution characteristics are compared with the local signal morphology characteristics at that location to determine the relative strength of their energy, which is used as the feature difference degree. Based on the characteristic difference degree, a competitive constraint is imposed on the global noise distribution characteristics and local signal morphology characteristics. In the competitive constraint, where the relative energy strength indicates that noise is dominant, noise characteristics take precedence; where the relative energy strength indicates that signal is dominant, signal characteristics take precedence. Based on the dominant force at each position after the competition constraint, a competition inhibition factor is generated. The competition inhibition factor is used to identify the energy subject that is ultimately retained at each position.
[0051] In one specific implementation, in order to achieve energy competition constraints between global noise distribution characteristics and local signal morphology characteristics at each spatial location, the present invention quantifies the feature difference degree and generates a competition suppression factor in the following manner.
[0052] First, calculate each spatial location. Energy competition coefficient :
[0053] in, This represents the local signal morphological characteristic value at that location. This represents the characteristic value of the global noise distribution at that location. To prevent the elimination of the zero constant, the energy competition coefficient is then compared with a preset competition threshold. Compare and generate the dominant party identifier. :
[0054] Among them, when This indicates that the location is dominated by signal characteristics. This indicates that the location is dominated by noise characteristics.
[0055] Specifically, for example, in a deep oil and gas exploration area, for each spatial location of the noisy seismic data volume, which corresponds to a seismic detection point at a certain depth underground in the area, the local signal morphology characteristic value of that location is first obtained. This value comes from the amplitude statistics of the local signal phase axis morphology after correction at that location. Then, the global noise distribution characteristic value of that location is obtained. This value comes from the energy statistics of the corresponding location in the global noise distribution characteristic map. The two values are directly compared to determine the relative strength of their energy. This relationship is the characteristic difference degree of that location.
[0056] For example, if the local signal morphology characteristic value of a certain detection point in the work area is greater than the global noise distribution characteristic value, it is determined that the signal energy at that location is stronger than the noise energy, which is the characteristic difference degree at that location.
[0057] Then, feature competition constraints are carried out based on the feature difference degree of each position. If the feature difference degree of a certain position indicates that the signal is dominant, that is, the energy of the local signal morphology feature of that position is stronger than the energy of the global noise distribution feature, then the local signal morphology feature is dominant and the signal correlation feature of that position is retained first. If the feature difference degree of a certain position indicates that the noise is dominant, that is, the energy of the global noise distribution feature of that position is stronger than the energy of the local signal morphology feature, then the global noise distribution feature is dominant and the noise correlation feature of that position is highlighted first.
[0058] For example, if the signal energy at a certain detection point in the middle of the work area is significantly stronger than the noise energy, then the focus is on the morphological characteristics of the in-phase axis at that location, with signal characteristics as the primary factor.
[0059] It should be noted that a zero constant for prevention and control also needs to be preset. This constant is determined by testing the seismic data of the work area in the early stage. A very small fixed value is selected to avoid the inability to calculate when the global noise distribution characteristic value is zero. For example, one-thousandth of the minimum noise energy value of the work area is selected as the zero constant for prevention and control. Then, the local signal morphology characteristic value and the global noise distribution characteristic value are obtained at each location. The two values are divided and the zero constant for prevention and control is added to obtain the energy competition coefficient at that location.
[0060] For example, if the local signal morphology characteristic value at a certain location is 4, the global noise distribution characteristic value is 2, and the zero constant is 0.001, then multiplying 4 by 2 and adding 0.001 gives the energy competition coefficient of that location as 8.001.
[0061] It is also necessary to preset a competition threshold. This threshold can be determined by the minimum value of the energy competition coefficient of the location where the signal is dominant in the historical seismic data of the work area. This minimum value is used as the competition threshold. Then, the energy competition coefficient of each location is compared with this threshold. If the energy competition coefficient is greater than or equal to the threshold, the competition suppression factor is 1, indicating that the location is dominated by signal characteristics; if the energy competition coefficient is less than the threshold, the competition suppression factor is 0, indicating that the location is dominated by noise characteristics.
[0062] For example, if the preset competition threshold is 5, and the energy competition coefficient at a certain position is 8.001, which is greater than the threshold, then its competition suppression factor is 1, and the signal characteristics dominate at this position.
[0063] As a preferred implementation, a competition inhibition factor is generated based on the dominant party at each position after competition constraints, including: After obtaining the dominant party identifiers at each position under competition constraints, a preliminary competition inhibition factor diagram is formed; Spatial continuity filtering is applied to the preliminary competition inhibition factor map to correct isolated dominant party identifiers to be consistent with the dominant parties in most of their surrounding locations. The revised competitive inhibition factor map is used as the final competitive inhibition factor.
[0064] Specifically, the dominant party identifiers determined after extracting the competition constraints at each location are used. When the dominant party identifier is 1, it indicates that the location is dominated by signal characteristics, and when the identifier is 0, it indicates that the location is dominated by noise characteristics. The dominant party identifiers of all locations are arranged one-to-one according to their spatial positions in the noisy seismic data volume to form a preliminary competition suppression factor map covering the entire exploration range of the work area.
[0065] For example, the dominant party identifiers of the detection points in the eastern part of the work area are mostly 1, while those in the western part are mostly 0. After arranging them according to this spatial distribution, a preliminary competition inhibition factor map is obtained.
[0066] Spatial continuity filtering is applied to the preliminary competition suppression factor map. Eight neighboring detection points around each location are selected as references. The number of dominant party identifiers at these eight neighboring locations is counted. If the dominant party identifier at a location is inconsistent with the identifiers of most of the surrounding neighboring locations, the dominant party identifier at that location is determined to be an isolated identifier and is corrected to the dominant party identifier of most of the surrounding neighboring locations.
[0067] For example, if the dominant identifier of a certain position is 0, and six of its eight neighboring positions are identified as 1, then the identifier of that position is corrected to 1, and so on, until all isolated identifiers are corrected.
[0068] After correcting all isolated dominant party identifiers, a competition suppression factor map covering the entire detection range of the work area without isolated identifiers is obtained. This corrected competition suppression factor map is used as the final competition suppression factor.
[0069] For example, in the revised competitive inhibition factor map, the dominant force in the central region of the work area is marked as 1, while the edge region is marked as 0, with no scattered isolated markers. This map is the final competitive inhibition factor used to identify the energy subjects at each location.
[0070] Overall, this scheme adaptively and accurately distinguishes between the dominant signals and noise, fundamentally avoiding the problems of effective signals being falsely suppressed and strong noise not being suppressed; spatial continuity filtering eliminates isolated outliers, making the factor distribution conform to geological spatial patterns and ensuring the rationality of feature constraints; it provides accurate identifiers for subsequent signal weighting and screening, significantly improving the accuracy of noise suppression and signal fidelity in deep, high-noise scenarios.
[0071] S4. Based on the competitive inhibition factor, the local signal morphological features are weighted and screened to retain the signal components with high confidence, thus obtaining the enhanced information body after competitive inhibition.
[0072] In this embodiment of the invention, in step S4, local signal morphological features are weighted and filtered according to a competitive suppression factor to retain high-confidence signal components, resulting in an enhanced information body after competitive suppression, including: Obtain the competition inhibition factor, which contains the energy principal identifier for each spatial location; Based on the energy subject identifier, a weighted filtering is performed on the local signal morphology features. The original signal components are retained where the energy subject identifier indicates that the signal is dominant, and the signal components are attenuated where the energy subject identifier indicates that the noise is dominant. The signal components at each position after weighted filtering are combined to form an enhanced information body after competition suppression.
[0073] In this embodiment of the invention, weighted filtering of local signal morphology features is performed based on the energy subject identifier, including: The energy entity identifier is decomposed into signal confidence levels, which are used to characterize the credibility of the signal components at each location. Based on the signal confidence level, a weighting coefficient is determined for each location, and the weighting coefficient is positively correlated with the signal confidence level. The weighting coefficients are multiplied by the signal components corresponding to each position in the local signal morphology features to obtain the weighted and filtered signal components.
[0074] In one specific implementation, the present invention combines a competitive suppression factor with local signal morphological features to generate weighted coefficients with smooth transitions. Specifically, for each spatial location... Weighting coefficients Determined by the following formula:
[0075] in, As a competitive inhibitor, This represents the local signal morphological characteristic value at that location. It represents the maximum value of the local signal morphological characteristics across all spatial locations. and The preset weighting coefficients are then used to weight and filter the original local signal morphological features, resulting in weighted and filtered signal components. :
[0076] in, This represents the original local signal morphological characteristics.
[0077] By employing the above methods, signal components are preserved or enhanced at locations with high signal confidence, while moderate attenuation is applied at locations where noise is dominant, thus achieving a smooth boundary transition.
[0078] Specifically, the final competition suppression factor is first extracted. This competition suppression factor is a competition suppression factor map covering the entire detection range of the work area. The energy main identifier of each spatial location in the map is read position by position. When the identifier is 1, it indicates that the signal is dominant at that location, and when it is 0, it indicates that the noise is dominant at that location.
[0079] For example, the energy main identifiers of the detection points in the southern part of the work area are mostly 1, while those in the northern part are mostly 0. The energy main identifiers of all locations are obtained in full for subsequent weighted filtering.
[0080] Furthermore, the energy subject identifier at each location is decomposed into a signal confidence level. When the energy subject identifier is 1, the signal confidence level is set to high, indicating that the signal component at that location is highly reliable. When the energy subject identifier is 0, the signal confidence level is set to low, indicating that the signal component at that location is low reliable.
[0081] For example, if the energy source at a certain location is identified as 1, its signal confidence level is high, and the credibility meets the requirement of preserving the original signal.
[0082] It should be noted that when two weighting coefficients are preset, they are determined through testing of early seismic data in the work area, and two fixed values that sum to 1 are selected to ensure that the weighting coefficient calculation is reasonable.
[0083] For example, select 0.6 as the first weighting coefficient and 0.4 as the second weighting coefficient. Then, count the local signal morphology feature values of all spatial locations, select the largest value as the maximum value of the local signal morphology feature, obtain the competition suppression factor and local signal morphology feature value of each location, multiply the first weighting coefficient by the competition suppression factor of the location, add the second weighting coefficient multiplied by the ratio of the local signal morphology feature value of the location to the maximum value, and obtain the weighting coefficient of the location.
[0084] The weighting coefficient increases as the confidence level of the signal increases. For example, if the competition suppression factor at a certain location is 1, the local signal morphological characteristic value is 8, and the maximum value is 10, then 0.6 is multiplied by 1, and 0.4 is multiplied by the ratio of 8 to 10 to obtain the weighting coefficient of 0.92 at that location.
[0085] Then, the original local signal morphological features are obtained position by position. The weighting coefficient of the position is multiplied by the corresponding original local signal morphological features to obtain the weighted and filtered signal components. If the signal confidence at the position is high, the weighting coefficient is close to 1, and the original signal components are retained; if noise is dominant, the weighting coefficient is close to 0, and the signal components are attenuated.
[0086] For example, if the weighting coefficient at a certain location is 0.92 and the original local signal morphology feature is 8, multiplying the two together gives 7.36, which is the signal component after weighted filtering at that location, thus achieving the preservation of signal components.
[0087] Finally, all spatially weighted signal components are combined according to their original spatial locations in the noisy seismic data volume to form a set that covers the entire exploration area and retains only high-confidence signal components. This set is the enhanced information volume after competition suppression.
[0088] For example, after screening all the detection points in the work area, the signal components are arranged according to the horizontal and vertical exploration line positions to form a complete enhanced information body, highlighting the signal components and attenuating the noise components.
[0089] In summary, this solution accurately distinguishes between high and low confidence signals, fundamentally avoiding the problems of false suppression of effective signals and noise residue; hierarchical weighting achieves a smooth transition, eliminating the harshness of the screening boundary; and adaptive intelligent screening is achieved by relying on competitive suppression factors, which greatly improves the signal purity and integrity of the enhanced information body, provides high-quality input for subsequent signal reconstruction, and significantly improves the accuracy of deep strong noise suppression and signal fidelity.
[0090] S5. Reconstruct the signal from the enhanced information volume to obtain seismic data after preliminary noise suppression.
[0091] This is to achieve seamless adaptation from the feature domain to the seismic data domain, avoiding signal distortion and gather misalignment, and fully restoring the effective signal shape. Therefore, in this embodiment of the invention, signal reconstruction is performed on the enhanced information volume to obtain seismic data after preliminary noise suppression, including: The enhanced information body after competition suppression is obtained, which contains signal components after weighted filtering from multiple spatial locations; Signal domain mapping is performed on the enhanced information volume, and the weighted and filtered signal components at each spatial location are transformed from feature expression form to seismic data expression form to obtain the reconstructed seismic gather; The reconstructed seismic gathers are integrated to form seismic data after preliminary noise suppression.
[0092] Specifically, the enhanced information body after competition suppression is first obtained. This enhanced information body contains the weighted and filtered signal components of all seismic detection points in the work area. The filtered signal components of each spatial detection point are read location by location. For example, the weighted and filtered signal components of the detection points in the south of the work area are 7.36 and those in the north are 0.8. The signal components of all detection points are obtained completely to ensure the integrity and accuracy of the enhanced information body.
[0093] Then, signal domain mapping is performed on the enhanced information volume. By referring to the standard expression format of seismic data in this work area, the weighted and filtered signal components at each spatial location are converted from eigenvalue form into amplitude time series signals that can be identified by the seismic trace.
[0094] If the signal component after weighted filtering at a certain location is 7.36, then the value is converted into a time-series waveform with the corresponding amplitude. The conversion is completed position by position to obtain the reconstructed seismic trace set composed of the time-series signals of each seismic trace.
[0095] Finally, the reconstructed seismic gathers are integrated. Seismic gathers of the same exploration line are spliced together in sequence according to the horizontal and vertical exploration lines of the three-dimensional seismic observation grid in the work area. Gathers of different exploration lines are integrated according to their spatial location to ensure that the spatial location of each seismic gather corresponds one-to-one with the detection point in the work area. After integration, seismic data covering the entire detection range of the work area is formed. This data is the seismic data after preliminary noise suppression.
[0096] S6. Use the difference between the pre-suppressed noise seismic data and the noisy seismic data volume to reverse correct the parameters in S3 or S5. Repeat S3 or S5 and its subsequent steps until convergence is achieved to obtain the final noise-suppressed deep seismic data.
[0097] In this embodiment of the invention, step S6 uses the difference between the pre-suppressed noise seismic data and the noisy seismic data volume to reverse-correct the parameters in S3 or S5, and repeats S3 or S5 and its subsequent steps until convergence, to obtain the final noise-suppressed deep seismic data, including: The difference volume between the pre-suppressed noise seismic data and the noisy seismic data volume is obtained. The difference volume reflects the amount of information remaining at each spatial location. Based on the distribution of residual information in the differential volume, determine the adjustment direction of feature competition constraints in S3 or the adjustment direction of signal reconstruction in S5, and correct the corresponding parameters according to the adjustment direction. Substitute the corrected parameters into S3 or S5, and repeat S3 or S5 and its subsequent steps until the differential volume meets the convergence condition, and output the final deep seismic data after suppressing noise.
[0098] Specifically, the signal values of the seismic data after initial noise suppression and the noisy seismic data volume are compared at each spatial location. The difference between the two values is taken as the information residual amount at that location. The information residual amounts of all locations are arranged one-to-one according to their original spatial locations within the detection range of the work area, forming a complete difference volume that reflects the information residual situation at each location.
[0099] For example, the initial suppressed data value at a certain location in the middle of the work area is 7.36, and the value of the noisy data volume is 9.2. The difference between the two is 1.84. This value is the information remaining amount at that location. The information remaining amounts at all locations are integrated into the difference volume.
[0100] Then, the information residual distribution at each position in the difference body is statistically analyzed. If the information residual in a certain region is generally high, the adjustment direction of the feature competition constraint in S3 is determined to be to increase the dominant weight of the signal features. If the information residual in a certain region is concentrated in the signal-dominant position, the adjustment direction of the signal reconstruction in S5 is determined to be to optimize the transformation rule of the signal domain mapping.
[0101] Adjust the corresponding parameters according to the adjustment direction, such as raising the competition threshold of S3, or adjusting the splicing and connection method of the set integration in S5, to ensure that the parameter adjustment fits the distribution characteristics of the remaining information.
[0102] Finally, substitute the corrected parameters into S3 or S5, re-execute S3 and subsequent steps, obtain the difference volume again and check the information residual distribution. If the information residual at all locations in the difference volume is lower than the preset convergence benchmark, the convergence condition is met, the iteration stops and the final deep seismic data after noise suppression is output. For example, if the preset convergence benchmark is that the information residual is lower than 0.5, and the information residual at all locations in the difference volume after iteration is lower than this value, the convergence is completed and the final data is output.
[0103] Overall, this scheme adaptively matches the noise characteristics of different regions, solving the problem of poor adaptability of fixed parameters and eliminating noise residue and signal distortion at the source; closed-loop iteration continuously optimizes the denoising effect until stable convergence, significantly improving the accuracy of deep strong noise suppression and signal fidelity; it realizes intelligent self-correction of parameters without manual intervention in parameter tuning, improving processing efficiency and process stability, and ensuring that the final deep seismic data quality meets the standards.
[0104] In the several embodiments provided by this invention, it should be understood that the disclosed method can be implemented in other ways.
[0105] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.
[0106] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, and technology that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.
[0107] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.
Claims
1. A method for intelligent suppression of strong noise in deep seismic data, characterized in that, The method includes the following steps: S1. Obtain raw deep seismic data, and obtain noisy seismic data volume after preprocessing; S2. Perform dual-channel parallel analysis on the noisy seismic data volume to extract global noise distribution characteristics and local signal morphology characteristics, respectively; S3. Based on the feature difference between global noise distribution characteristics and local signal morphology characteristics, feature competition constraints are performed, and competition suppression factors are generated based on the constraint results; S4. Based on the competitive inhibition factor, the local signal morphological features are weighted and screened to retain the signal components with high confidence, thus obtaining the enhanced information body after competitive inhibition; S5. Perform signal reconstruction on the enhanced information volume to obtain seismic data after preliminary noise suppression; S6. Use the difference between the pre-suppressed noise seismic data and the noisy seismic data volume to reverse correct the parameters in S3 or S5. Repeat S3 or S5 and its subsequent steps until convergence is achieved to obtain the final noise-suppressed deep seismic data.
2. The intelligent noise suppression method for deep seismic data as described in claim 1, characterized in that, In step S1, raw deep seismic data is acquired, and after preprocessing, a noisy seismic data volume is obtained, including: The raw deep seismic data is acquired, and the raw deep seismic data is filtered to remove missing and bad traces, so as to obtain a complete seismic record collection. Energy equalization processing is performed on the seismic record gathers to compensate for the energy attenuation of deep signals, resulting in an energy-equalized seismic data volume, which serves as the noisy seismic data volume.
3. The intelligent noise suppression method for deep seismic data as described in claim 1, characterized in that, In S2, a dual-channel parallel analysis is performed on the noisy seismic data volume to extract global noise distribution features and local signal morphology features, including: The noisy seismic data volume is partitioned into multiple data sub-regions; For each data sub-region, the first analysis channel and the second analysis channel are executed in parallel. The first analysis channel is used to extract the local noise energy distribution of the data sub-region, and the second analysis channel is used to extract the local signal in-phase axis shape of the data sub-region. The local noise energy distribution of each data sub-region is fused to form a global noise distribution feature, and the local signal phase axis morphology of each data sub-region is spliced together to form a local signal morphology feature.
4. The intelligent noise suppression method for deep seismic data as described in claim 3, characterized in that, The parallel execution of the first analysis channel and the second analysis channel for each data sub-region includes: Within each data sub-region, the first analysis channel is executed to statistically analyze the energy distribution within the data sub-region along the time direction. Based on the level of energy distribution, the data sub-region is divided into a strong noise region and a weak noise region. Within each data sub-region, the second analysis channel is executed to trace the direction of the in-phase axis in the spatial direction and extract the continuity characteristics and amplitude variation characteristics of the in-phase axis. The division results of strong noise region and weak noise region are synchronously transmitted to the second analysis channel. The complete in-phase axis tracking results are retained in the weak noise region. In the strong noise region, the in-phase axis in the strong noise region is extended and fitted according to the in-phase axis direction of the adjacent weak noise region to obtain the corrected local signal in-phase axis shape.
5. The intelligent noise suppression method for deep seismic data as described in claim 1, characterized in that, In step S3, feature competition constraints are performed based on the feature difference between global noise distribution characteristics and local signal morphological characteristics, and a competition suppression factor is generated based on the constraint results, including: At each spatial location of the noisy seismic data volume, the global noise distribution characteristics are compared with the local signal morphology characteristics at that location to determine the relative strength of their energy, which is used as the feature difference degree. Based on the feature difference degree, a competitive constraint is imposed on the global noise distribution features and local signal morphology features. In the competitive constraint, when the relative strength of energy indicates that noise is dominant, the noise features are dominant; when the relative strength of energy indicates that the signal is dominant, the signal features are dominant. Based on the dominant force at each position after the competition constraint, a competition inhibition factor is generated. The competition inhibition factor is used to identify the energy subject that is ultimately retained at each position.
6. The intelligent noise suppression method for deep seismic data as described in claim 5, characterized in that, The step of generating competition inhibition factors based on the dominant party at each position after competition constraints includes: After obtaining the dominant party identifiers at each position under competition constraints, a preliminary competition inhibition factor diagram is formed; Spatial continuity filtering is applied to the preliminary competition inhibition factor map to correct isolated dominant party identifiers to be consistent with the dominant parties in most of their surrounding locations. The revised competitive inhibition factor map is used as the final competitive inhibition factor.
7. The intelligent noise suppression method for deep seismic data as described in claim 1, characterized in that, In step S4, local signal morphological features are weighted and filtered based on a competitive suppression factor to retain high-confidence signal components, resulting in an enhanced information body after competitive suppression, including: Obtain the competition inhibition factor, which contains the energy principal identifier for each spatial location; Based on the energy subject identifier, a weighted filtering is performed on the local signal morphology features. The original signal components are retained where the energy subject identifier indicates that the signal is dominant, and the signal components are attenuated where the energy subject identifier indicates that the noise is dominant. The signal components at each position after weighted filtering are combined to form an enhanced information body after competition suppression.
8. The intelligent noise suppression method for deep seismic data as described in claim 7, characterized in that, The step of performing weighted filtering on local signal morphology features based on energy entity identifiers includes: The energy entity identifier is decomposed into signal confidence levels, which are used to characterize the credibility of the signal components at each location. Based on the signal confidence level, a weighting coefficient is determined for each location, and the weighting coefficient is positively correlated with the signal confidence level. The weighting coefficients are multiplied by the signal components corresponding to each position in the local signal morphology features to obtain the weighted and filtered signal components.
9. The intelligent noise suppression method for deep seismic data as described in claim 1, characterized in that, In step S5, signal reconstruction is performed on the enhanced information volume to obtain seismic data after preliminary noise suppression, including: The enhanced information body after competition suppression is obtained, which contains signal components after weighted filtering from multiple spatial locations; Signal domain mapping is performed on the enhanced information volume, and the weighted and filtered signal components at each spatial location are transformed from feature expression form to seismic data expression form to obtain the reconstructed seismic gather; The reconstructed seismic gathers are integrated to form seismic data after preliminary noise suppression.
10. The intelligent noise suppression method for deep seismic data as described in claim 9, characterized in that, In step S6, the parameters in S3 or S5 are corrected in reverse using the difference between the pre-suppressed noise seismic data and the noisy seismic data volume. S3 or S5 and subsequent steps are repeated until convergence is achieved, resulting in the final noise-suppressed deep seismic data, including: The difference volume between the pre-suppressed noise seismic data and the noisy seismic data volume is obtained. The difference volume reflects the amount of information remaining at each spatial location. Based on the distribution of residual information in the differential volume, determine the adjustment direction of feature competition constraints in S3 or the adjustment direction of signal reconstruction in S5, and correct the corresponding parameters according to the adjustment direction. Substitute the corrected parameters into S3 or S5, and repeat S3 or S5 and its subsequent steps until the differential volume meets the convergence condition, and output the final deep seismic data after suppressing noise.