CN100349012C - Method for pressing random noise in seismological record with low SNR - Google Patents

Abstract

The present invention relates to a method for suppressing medium low signal noise ratio random noise in earthquakes. The operator length is selected on a single frequency point to calculate a prediction operator; in the frequency field, data on different earthquake traces on the same frequency point is processed by a single-point frequency prediction operator in a noise attenuation mode so as to eliminate the random noise which is distributed irregularly; the data of the elimination of the random noise is used as effective signals, the width of the frequency band is selected, the dominant frequency band prediction operator is counted, and the non-dominant frequency band prediction operator is determined; dominant frequency band extrapolation random noise suppression processing is completed. The present invention can accurately determine and suppress low signal noise ratio random noise in earthquakes according to different signal noise ratios in different frequency ranges in the earthquake data frequency space field.

Description

The method of random noise in the compacting low signal-to-noise ratio seismologic record

Technical field

The present invention relates to the geophysical survey seismic data processing technology, is the method for random noise in a kind of compacting low signal-to-noise ratio seismologic record that improves geological data signal to noise ratio (S/N ratio), resolution.

Background technology

In the earthquake data acquisition process, because subterranean strata anisotropy, in not homologous pints generation scattering, and the influence of the wind in human factor and the natural conditions, thunder, electricity etc., each frequency band has comprised very strong random noise disturbance in the geological data of gathering, had a strong impact on seismic data analysis and processing, made the seismic section precise decreasing.So in the seismic data with low signal-to-noise ratio processing procedure, must suppress and decay to random noise.

In the high-resolution seismic exploration data handling procedure, the high frequency random noise utmost point seriously influences earthquake useful signal frequency band and widens.Because the earth absorbs and the influence of other factors, received seismic wavelet radio-frequency component decay is serious, random noise for depth of stratum a little less than the influence of reflected signal particularly serious, so seismologic record high frequency region signal to noise ratio (S/N ratio) is normally lower.Widen the frequency band of significant wave, improve resolution, should at first consider how to improve the signal to noise ratio (S/N ratio) of seismologic record high frequency region, eliminate the influence of high frequency noise.

Conventional territory, frequency space random noise decay (F-X territory) forecasting techniques is based on the linear lineups of seismic signal and can predicts in the territory, frequency space, and the uncertain principle of random noise, adopt the plural number method of one-step prediction forward, extract predictable linear lineups, separate random noise simultaneously, though this technology can be suppressed random noise to a certain extent, the effect of its compacting random noise mainly relies on the signal to noise ratio (S/N ratio) of input record itself.Because the different frequency range of geological data has different signal to noise ratio (S/N ratio)s, so F-X territory forecasting techniques just is difficult to bring into play the effect of suppressing noise for the frequency range that signal to noise ratio (S/N ratio) is lower than to a certain degree.As seen, this technology its Frequency point in processing procedure is single all the time, the predictive operator of current Frequency point is only estimated by the information of a Frequency point, its precision depends on the original signal to noise ratio (S/N ratio) of this Frequency point, if only ask the signal to noise ratio (S/N ratio) of single-frequency point by a seismic trace, its value is not only different, and the signal to noise ratio (S/N ratio) curve is jagged to a great extent.Show that predictive operator changes in its reliability of different Frequency points, especially after frequency acquires a certain degree, along with increasing of frequency, signal to noise ratio (S/N ratio) progressively reduces, the reliability of operator is variation progressively also, behind signal to noise ratio (S/N ratio) (after the denoising before the value/denoising of useful signal amplitude energy the amplitude energy value of useful signal)＜1/6, the signal of prediction is false.

Summary of the invention

Technical matters to be solved by this invention provide a kind of can be at the different signal to noise ratio (S/N ratio)s of different frequency range in the territory, geological data frequency space, accurately determine and compacting low signal-to-noise ratio seismologic record in the method for random noise.

The method of random noise in the compacting low signal-to-noise ratio seismologic record of the present invention comprises following treatment step:

(1) acquiring seismic data transforms to frequency field with the multiple tracks geological data from time domain, forms the vector data of spatial domain;

(2) on single frequency point, selected operator length is asked for predictive operator;

(3) in frequency field, with single-point frequency predication operator the data of different earthquake road same frequency point are carried out noise attentuation, eliminate the random noise of irregular distribution;

(4) the data of eliminating random noise as useful signal, selected frequency span, in whole frequency field, add up the amplitude energy of the forward and backward different earthquake of denoising road same frequency band width respectively, utilize the amplitude energy ratio of the forward and backward same frequency band width of denoising to ask for signal to noise ratio (S/N ratio) S/N, the maximum frequency band of S/N value promptly is confirmed to be the dominant frequency band scope than the high s/n ratio frequency band;

(5) statistics dominant frequency band predictive operator is determined non-dominant frequency band predictive operator;

(6) finish dominant frequency band extrapolation random noise compression process,

(7) geological data is transformed into time domain from frequency field.

The present invention adopts Fourier Tranform to realize the mutual conversion of geological data between time domain and frequency field.

The present invention adopts conventional F-X territory predictive filtering method to ask for predictive operator at the single frequency point of frequency field, and the method for promptly using Wiener filtering is asked for the predictive operator of single-frequency point.

Linear lineups according to seismic signal are foreseeable (comprising line noise and the linear lineups that do not continue) in the territory, frequency space, and random noise is the dominant frequency band that uncertain principle is asked for seismic signal, use conventional F-X territory forecasting techniques, adopt the Forecasting Methodology of plural number to back, extract predictable linear lineups, separate random noise, thereby improve the signal to noise ratio (S/N ratio) and the continuity of seismic section, concrete steps are as follows:

X in objective function (1) _n(f) be the single-frequency raw data function that contains random disturbance, and in formula (1) It is conventional F-X territory prediction denoising result function.(f) has so if the signal to noise ratio (S/N ratio) of single-point frequency is E:

$E\left(f\right)=\frac{\left|{\hat{S}}_n\left(f\right)\right|}{X_n\left(f\right)}---\left(1\right)$

Wherein || the mould of expression plural number.The signal to noise ratio (S/N ratio) of single-point frequency does not have practical significance, and what need is the signal to noise ratio (S/N ratio) in territory, frequency space, so have:

$E\left(\mathrm{\Δf}\right)=\frac{\underset{l=1}{\overset{n}{\mathrm{\Σ}}}\left|{\hat{S}}_n(\mathrm{\Δf},x_l)\right|}{\underset{l=1}{\overset{n}{\mathrm{\Σ}}}\left|X_n(\mathrm{\Δf},x_l)\right|}---\left(2\right)$

The bandwidth of self-defined dominant frequency band is k, and then the dominant frequency band signal to noise ratio (S/N ratio) is:

$E\left(\mathrm{K\Δf}\right)=\frac{\underset{i=1}{\overset{K}{\mathrm{\Σ}}}\underset{l=1}{\overset{n}{\mathrm{\Σ}}}\left|{\hat{S}}_n(f_i,x_l)\right|}{\underset{i=1}{\overset{K}{\mathrm{\Σ}}}\underset{l=1}{\overset{n}{\mathrm{\Σ}}}\left|X_n(f_i,x_l)\right|}---\left(3\right)$

Wherein f represents frequency, and Δ f represents sampling interval, f _iRepresent different Frequency points, x _iRepresent different seismic traces, can search for the dominant frequency band signal to noise ratio (S/N ratio) of given bandwidth so automatically by Frequency point.

After the present invention tries to achieve dominant frequency band, discern the random noise of non-dominant frequency band.By the corresponding predictive operator component of dominant frequency band is added up, according to the reflection of the predictive operator of signal is signal space continuity direction, and the principle that signal space continuity direction does not change with frequency, the predictive operator component of different frequency has space-time linear geometry form similar or that slowly change with frequency, and has certain relevance, adjacent its similarity degree of frequency predication operator component is also high more, it shows a kind of simple sight line sexual intercourse on frequency axis, just can extrapolate the predictive operator component of non-dominant frequency band thus.If dominant frequency band and non-dominant frequency band predictive operator component correlation coefficient are Q, then have:

$F_{j,m}\left(\mathrm{k\Δf}\right)=\underset{m=1}{\overset{C_{L-j}}{\mathrm{\Σ}}}{Q}_{j,m}{F}_{j,m}\left[(k\±m)\mathrm{\Δf}\right]---\left(4\right)$

(j＝1，2，...，L)

In the formula (4): F _{J, m}Be the predictive operator component of a certain frequency, with common Wiener filtering method can in the hope of, C is a composite symbol, $C_{L-j}=C_L^{L-j};$ Δ f is the frequency sampling interval; K is the bandwidth of advantage frequency range, i.e. the operator extrapolation length; L is an operator length; In the formula+and expression is to the low frequency extrapolation, and-expression is extrapolated to high frequency.Correlation coefficient is that Q tries to achieve from (5) formula.

$\left(\begin{array}{c}{Q}_{j,1}\\ Q_{j,2}\\ Q_{j,3}\\ ...\\ Q_{j,k}\end{array}\right)={\left(\begin{array}{ccccc}{F}_{j,1}& F_{j,2}& F_{j,3}& ...& F_{j,k}\\ F_{j,2}& F_{j,3}& F_{j,4}& ...& F_{j,k+2}\\ F_{j,3}& F_{j,4}& F_{j,5}& ...& F_{j,k+3}\\ ...& ...& ...& ...& ...\\ F_{j,m}& ...& ...& ...& F_{j,m+k-1}\end{array}\right)}^{-1}\·\left(\begin{array}{c}{\underset{\‾}{F}}_{j,k+1}\\ F_{j,k+2}\\ F_{j,k+3}\\ ...\\ F_{j,k+m}\end{array}\right)---\left(5\right)$

The present invention is behind the predictive operator of trying to achieve dominant frequency band and non-dominant frequency band, on the single-point frequency of frequency field, for the prediction seismic trace, utilize the seismic trace of its predictive operator length in front and back to multiply each other, again the single-point frequency of multiplied result addition as the prediction seismic trace with the predictive operator component is corresponding.Its principle is similar to time domain the amplitude addition of the same sampled point of front and back seismic trace is averaged as the amplitude of the same sampled point of prediction seismic trace.

The present invention for earthquake data before superposition, needs carry out normal moveout correction to data earlier during to the compression process of non-dominant frequency band random noise.

For geological condition complexity, bigger area, lineups inclination angle, when carrying out the random noise compacting, shorten pre-operator length, widen dominant frequency band.

In the territory, frequency space, putting its predictive operator at side frequency is similar or slow the variation according to effective seismic wavelet in the present invention.Based on the high frequency range data of signal to noise ratio (S/N ratio), excellent frequency range predictive operator component is carried out comprehensive statistics, recursion goes out excellent frequency range predictive operator component in addition successively, finish the prediction denoising of all frequency contents, the predictive operator of low signal-to-noise ratio frequency range can reach reliable precision like this, more effectively estimated signal is separated random noise, reaches the purpose that improves geological data signal to noise ratio (S/N ratio) and resolution.The present invention can accurately determine the scope of earthquake data edge frequency band.When removing random noise, more effectively protected the useful signal of medium-high frequency band, under the situation especially lower, that the predictive operator precision is not high in signal to noise ratio (S/N ratio) than conventional F-X prediction.

Description of drawings

Fig. 1 (a) is the theoretical model data;

Fig. 1 (b) is the seismic trace that Fig. 1 (a) adds noise;

Fig. 1 (c) dominant frequency band operator extrapolation compacting noise processed result;

Fig. 1 (d) is master pattern Fig. 1 (a) frequency and amplitude spectrogram;

Fig. 1 (e) adds illustraton of model 1 (b) the frequency and amplitude spectrogram of making an uproar;

Fig. 1 (f) is a frequency and amplitude spectrogram after the dominant frequency band operator extrapolation process, 1-useful signal wherein, and the 2-useful signal is pressed, and the 3-useful signal is resumed;

Fig. 2 (a) is the post-stack seismic data section;

Fig. 2 (b) carries out seismic section after the dominant frequency band operator extrapolation compacting noise processed to post-stack seismic data;

Fig. 2 (c) is a geological data signal to noise ratio (S/N ratio) section;

Fig. 2 (d) is through the geological data signal to noise ratio (S/N ratio) section after the dominant frequency band operator extrapolation compacting noise processed;

Fig. 2 (c), (d) cathetus 4 is zero shellfish markings, the markings top is signal to noise ratio (S/N ratio)＞1;

Fig. 3 (a) is the preceding CRP gather of denoising after the prestack normal moveout correction;

Fig. 3 (b) is a CRP gather after the dominant frequency band operator extrapolation denoising after the prestack normal moveout correction;

Embodiment

Embodiment 1

Dominant frequency band extrapolation noise suppression methods of the present invention applies to gross data, and Fig. 1 (a) is a high spud angle stratum theoretical model, and the noiseless interference, and four lineups that can follow the trail of are continuously wherein arranged, and comprises a flat seam and three layer positions that the inclination angle is different.Fig. 1 (b) is the seismic trace that Fig. 1 (a) adds noise, is gone up as seen by Fig. 1 (b): because signal to noise ratio (S/N ratio) is lower, useful signal is wherein flooded by noise basically.Fig. 1 (d) shows that its useful signal of Fig. 1 (a) concentrates on the 10-100Hz frequency range basically, Fig. 1 (e) shows that random noise mainly concentrates on more than the 40Hz, dominant frequency band is essentially 10-50Hz, low, the intermediate frequency that have so just guaranteed model have higher signal to noise ratio (S/N ratio), thereby are close with actual geological data.

The multiple tracks geological data of gathering is transformed to frequency field from time domain, form the vector data of spatial domain; As the selected predictive operator length of Fig. 1 (b) is 7, and 7 seismic traces that input contains random noise geological data Fig. 1 (b) carry out Fourier Tranform to 7 seismic channel data then, form the vector data of spatial domain.

Ask for the Wiener filtering operator at each single frequency point then, the data of different earthquake road same frequency point are carried out noise attentuation, eliminate the random noise of irregular distribution with single-point frequency operator.

As useful signal, is the data after the denoising 41 to Fig. 1 (b) selective advantage frequency span, adds up the amplitude energy of different earthquake road, denoising front and back same frequency band respectively,

By formula (3)

$E\left(\mathrm{K\Δf}\right)=\frac{\underset{i=1}{\overset{K}{\mathrm{\Σ}}}\underset{l=1}{\overset{n}{\mathrm{\Σ}}}\left|{\hat{S}}_n(f_i,x_l)\right|}{\underset{i=1}{\overset{K}{\mathrm{\Σ}}}\underset{l=1}{\overset{n}{\mathrm{\Σ}}}\left|X_n(f_i,x_l)\right|}$

Utilize the amplitude energy ratio of denoising front and back same frequency band to ask signal to noise ratio (S/N ratio), be confirmed to be the dominant frequency band scope than the high s/n ratio frequency band, what calculate is 10-50Hz than the high s/n ratio frequency band, can determine that thus Fig. 1 (b) geological data dominant frequency band is 10-50Hz.

Utilize the predictive operator of dominant frequency band then, through type (4),

$F_{j,m}\left(\mathrm{k\Δf}\right)=\underset{m=1}{\overset{C_{L-j}}{\mathrm{\Σ}}}{Q}_{j,m}{F}_{j,m}\left[(k\±m)\mathrm{\Δf}\right]$

(j＝1，2，...，L)

Formula (5),

$\left(\begin{array}{c}{Q}_{j,1}\\ Q_{j,2}\\ Q_{j,3}\\ ...\\ Q_{j,k}\end{array}\right)={\left(\begin{array}{ccccc}{F}_{j,1}& F_{j,2}& F_{j,3}& ...& F_{j,k}\\ F_{j,2}& F_{j,3}& F_{j,4}& ...& F_{j,k+2}\\ F_{j,3}& F_{j,4}& F_{j,5}& ...& F_{j,k+3}\\ ...& ...& ...& ...& ...\\ F_{j,m}& ...& ...& ...& F_{j,m+k-1}\end{array}\right)}^{-1}\·\left(\begin{array}{c}{F}_{j,k+1}\\ F_{j,k+2}\\ F_{j,k+3}\\ ...\\ F_{j,k+m}\end{array}\right)$

Recursion goes out the predictive operator of non-dominant frequency band.

The predictive operator of the non-dominant frequency band that goes out with the predictive operator and the recursion of dominant frequency band carries out the noise compression process to seismic channel data at last, on the single-point frequency, seismic trace for a predictive operator length, utilize corresponding the multiplying each other of predictive operator of three seismic traces in front and three seismic traces in back, again the single-point frequency of multiplied result addition as prediction seismic trace the 4th road, the 4th seismic trace can be predicted comes out.

Use the same method then and dope other seismic channel data, carry out dominant frequency band extrapolation random noise compression process, compacting noise result such as Fig. 1 (c), the dominant frequency band operator postpones outward, owing to be subjected to the influence of 10-50Hz frequency content extrapolation, the predictive operator precision is improved greater than the predictive operator of the radio-frequency component of 50Hz.The ability of compacting random noise strengthens greatly, comparison diagram 1 (d), and (e), (f) decay of random noise makes that downtrodden useful signal is recovered well as can be known.At last geological data is transformed into time domain from frequency field, the useful signal of the random noise that has been eliminated.

Embodiment 2

For the geological data of geological condition more complicated, shown in Fig. 2 (a), the complicacy of itself causes predictability not high, and the geological data regularity is very poor, if the operator of prediction is long more, estimation range is big more, and it is just inaccurate more to predict the outcome.So this moment, predictive operator should suitably be selected weak point, but if operator is too short, from prediction theory as can be known, the burbling noise ability weakens, so operator length is selected and must be determined through overtesting.Fig. 2 (a) left side section lineups poor continuity, there are the high spud angle lineups on the right, and operator length selects 5, and the dominant frequency band width is 60.Adopt 1 described identical method prediction and compacting random noise with embodiment, result such as Fig. 2 (b), comparison diagram 2 (c), (d) as seen, signal to noise ratio (S/N ratio) has had large increase before and after the denoising, has reached the purpose that improves resolution.

Embodiment 3

The present invention can also handle earthquake data before superposition except can handling the poststack seismologic record, prerequisite is to carry out after normal moveout correction is handled.Fig. 3 (a) is a CRP gather after the normal moveout correction, adopts the disposal route identical with embodiment 1, and Fig. 3 (b) is the result after this method processing.Because before the normal moveout correction on common reflection point lineups be hyperbola distribution, lineups linearly distribute on common reflection point after the normal moveout correction.This method can predict that to the lineups of form linearly the hyperbolic curve lineups approach with straight line in forecasting process, predicting the outcome is not accurate especially, and predictive operator is long more, and error is big more, handles so must carry out normal moveout correction before the denoising.

Claims (4)

1, a kind of method of suppressing random noise in the low signal-to-noise ratio seismologic record comprises following treatment step:

(1) acquiring seismic data adopts Fourier Tranform that the multiple tracks geological data is transformed to frequency field from time domain, forms the vector data of spatial domain;

(2) the single frequency point in frequency field adopts conventional F-X territory predictive filtering method to ask for predictive operator,

(3) in frequency field, with single-point frequency predication operator the data of different earthquake road same frequency point are carried out noise attentuation, eliminate the random noise of irregular distribution;

(4) the data of eliminating random noise as useful signal, the width of selected frequency band, in whole frequency field, add up the amplitude energy of different earthquake road, denoising front and back same frequency band width respectively, utilize the amplitude energy ratio of denoising front and back same frequency band width to ask for signal to noise ratio (S/N ratio), the frequency band of snr value maximum promptly is confirmed to be dominant frequency band than the high s/n ratio frequency band;

(5) by the corresponding predictive operator component of dominant frequency band is added up,, extrapolate the predictive operator component of non-dominant frequency band, and then try to achieve the predictive operator of non-dominant frequency band according to the relevance of dominant frequency band and non-dominant frequency band predictive operator component;

Dominant frequency band and non-dominant frequency band predictive operator component correlation coefficient are Q, then have:

$F_{j,m}\left(\mathrm{k\Δf}\right)=\underset{m=1}{\overset{C_{L-j}}{\mathrm{\Σ}}}{Q}_{j,m}{F}_{j,m}\left[(k\±m)\mathrm{\Δf}\right]---\left(4\right)$

(j＝1，2，…，L)

In the formula (4): F _{J, m}Be the predictive operator component of a certain frequency, try to achieve that C is a composite symbol with common Wiener filtering method, $C_{L-j}=C_L^{L-j};$ Δ f is the frequency sampling interval; K is the bandwidth of advantage frequency range, is the operator extrapolation length; L is an operator length; In the formula+and expression is to the low frequency extrapolation, and-expression is extrapolated to high frequency,

Correlation coefficient is that Q tries to achieve from following formula:

$\left(\begin{array}{c}{Q}_{j,1}\\ Q_{j,2}\\ Q_{j,3}\\ \·\·\·\\ Q_{j,k}\end{array}\right)={\left(\begin{array}{ccccc}{F}_{j,1}& F_{j,2}& F_{j,3}& \·\·\·& F_{j,k}\\ F_{j,2}& F_{j,3}& F_{j,4}& \·\·\·& F_{j,k+2}\\ F_{j,3}& F_{j,4}& F_{j,5}& \·\·\·& F_{j,k+3}\\ \·\·\·& \·\·\·& \·\·\·& \·\·\·& \·\·\·\\ F_{j,m}& \·\·\·& \·\·\·& \·\·\·& F_{j,m+k-1}\end{array}\right)}^{-1}\·\left(\begin{array}{c}{F}_{j,k+1}\\ F_{j,k+2}\\ F_{j,k+3}\\ \·\·\·\\ F_{j,k+m}\end{array}\right)---\left(5\right)$

(6) finish dominant frequency band extrapolation random noise compression process;

(7) geological data is transformed into time domain from frequency field, the useful signal of the random noise that is eliminated generates seismic cross-section.

2, the method for random noise in the compacting low signal-to-noise ratio seismologic record according to claim 1, after it is characterized in that trying to achieve the predictive operator of dominant frequency band and non-dominant frequency band, on frequency field single-point frequency, for the prediction seismic trace, utilize the seismic trace of its predictive operator length in front and back to multiply each other, again the single-point frequency of multiplied result addition as the prediction seismic trace with the predictive operator component is corresponding.

3, the method for random noise in the compacting low signal-to-noise ratio seismologic record according to claim 1 when it is characterized in that the random noise compression process to non-dominant frequency band, for earthquake data before superposition, needs advanced action treatment for correcting.

4, the method for random noise in the compacting low signal-to-noise ratio seismologic record according to claim 1 is characterized in that for complicated geological situation, bigger area, lineups inclination angle, during the compacting random noise, should suitably shorten predictive operator length.

Images