| In the recent years, with the increasing demand for oil and gas, the prospecting maturity has been advanced ceaselessly, thus the object of the seismic prospecting becomes more and more complicated: On the one side, the difficulty of the prospecting has been increasing ceaselessly because the study orientation has turned from simple big-middle structure oil and gas store fields to multiple ones. On the other side, the visual field of prospecting study has advanced to the earth deep-layer. It means that the complex deep structure and geosphere have greater relationship with the oil and gas production, delivery, collection, storage and so on. The complexity of seismic data is made from not only the variational underground geologic structure but also the earth'surface geologic condition ( such as desert, hill and forest, etc). Complicated character is going with the seismic-wave medium flexibility and geometry. And the seismic-wave transmit path, quiver intensity and wave shape have to permeate them. Additionally the attenuation is different with different frequency wave absorbed by the underground medium. Therefore, the seismic information incepted on earth surface is typical non-stationary signal. Such signal frequency spectrum and multifarious statistic characteristic is changed by the time changing. And the changes just reflect the abundance information of underground echo layer sequence.The dissertation briefly presents the develop history of time-frequency analysis, and how the existing various methods are applied in seismic prospecting. And then the dissertation carries out matching pursuit wavelet packet time-frequency method. By these time frequency analysis, I calculate instant spectrum attribute. From amplitude attribute and frequency attribute to instant spectrum attribute, I get some new cognition. In the dissertation, the following four aspects are studied. And some new understandings are obtained. They should do some contribute in promoting development of the time-frequency analysis method study work and application in seismic prospecting:1. Sum up existing time-frequency analysis method: hilbert transform, short window discrete fourier transform, wigner-ville distribution, match pursuit arithmetic, wavelet analysis, wavelet packet analysis. And introduce how they are applied in seismic prospecting.2. Seismic prospecting signal is changing fleetly with time. Even in the same frequency area, signals have different energy distribution at different moment. The best wavelet packet arithmetic does not conform to this kind of time changing. The dissertation develops wavelet packet analysis, and devises the matching pursuit wavelet packet method and in detail presents the method flow and process. Using the nonlinear best matching pursuit method, the signal time-frequency character can be expressed well through finding the best atom in the atom storeroom built up by wavelet packet. The wavelet packet in the method is computed by Daubechies filter. It requires that the signal length is 2n. Otherwise two sides"0"extending will be done. We can choose rule of stopping iteration according to different aim on decomposing signal. As shown in time-frequency plane, every atom is represented by the wigner-ville distribution. Considering interfering item, we can do the corresponding calculation after getting the resoled signal. We can use the smoothed pseudo wigner-ville time-frequency distribution. directly for the better resolution. The time-frequency plane analysis on typical signal and the time-frequency relation curve obtained by the method are consistent with the academic results. Especially for the non-stationary signal, the method shows the sine wave relation curve of time and frequency. The method can express the seismic signal sparsely. When the iteration times go to infinity, the approaching error is going down to zero by exponential powers. In fact, including more information the seismic signal is more complex than man-made seismic signal. So it needs to be presented by more time-frequency atoms. As shown in the experiment, in order to present the exact seismic information, the synthetical record needs fifty iteration times (fifty time-frequency atoms) while practical recorder needs one hundred and fifty times (one hundred and fifty atoms).3. The time-frequency analysis can localize the spectrum character and reflect the rules of how the frequency is transformed by time changing. Two time-frequency seismic signal analysis methods of short-time FFT and ST can show reflection wave, alternating current interference, the first arrival wave and surface wave. Based on the time-changing filter of time-frequency domain, interesting signal can be picked up. It means that the signal in open-region is accepted and the others are rejected. And then the time-frequency filtering process is finished after the signal filtered through the time-frequency is rebuilt in time field. The filtering method on the time-frequency field can consider both time character and frequency character. It can filter the noise better aimed at the broad band or non-stationary signal. It has better character of denoising than traditional linear filtering methods. The denoising method on wavelet shrinkage of threshold value can eliminate the random noise. After the wavelet decomposing, the value of wavelet coefficient is bigger than the noise one. So we can consider that the bigger value of wavelet coefficient is mainly corresponding to signal and the smaller one is corresponding to noise in great degree. So the method of threshold value can save the signal coefficient and reduce the most noise coefficient to zero. The method of matching wavelet packet also can eliminate the occasional noise. From the decomposing and recomposing opinion, the noised signal has two parts, signal and noise, and the signal is the sparse elements in the noised signal. Signal has certain structure and on character it is identical to atom. Noise is random and disrelated. So it has no structure character. If the significative atom can be picked up from signal, the picked one is signal. If no significative signal can be picked up from the residuary signal more over, we can consider that there is all noise in the residua. We can construct the white noise which SNR are 20dB,15dB,10dB,5dB,0dB,-5dB,-10dB respectively. And add them to the seismic compound note. And then we can study the denoising effect of the matching wavelet pocket and wavelet shrinkage of threshold value. Both methods have certain effect on restraining white noise. And their denoising ability decline with the SNR falls. When the SNR is 0dB, both methods can not eliminate the occasional noise. As shown in the experience's result, the matching wavelet pocket method can improve the SNR about 2dB when dealing with the seismic information which SNR are 20dB,15dB,10dB,5dB respectively. So the method is better than wavelet shrinkage of threshold value. When the SNR are -5dB,-10dB, the matching wavelet pocket method is lower than the wavelet shrinkage of threshold value on the result SNR.4 After reviewing old theories of thin bed, we study spectrum's characteristic on single thin layer, explains the thin bed reconstructs the spectrum of incident wave, so thin bed is a filter, and its spectrum is of periodic, which is reciprocal of time thick. Theoretical model can see a notching from amplitude spectrum of single trace, and further expose the notching from amplitude spectrum of single frequency. Quantitatively explain that period of spectrum is reciprocal of time thick. Composite wave in the model study of one layer, formula tells us relation of peak frequency and bed's thick. If the bed is thin, formula is changed to easiness, but easy formula isn't suit to situation that two reflection coefficients is approximate equal and have the same polarities. I build four wedge models to study thin bed by some attributes. Reflection coefficients'amplitude spectrum is period, and period is reciprocal of time thick, but synthetical record's amplitude spectrum is affected by seismic wavelet, it is equal that reflection coefficient sequence is filtered by seismic wavelet, so periodic phenomena is not clear. Compare reflection coefficients'amplitude spectrum with synthetical record's amplitude spectrum, it is obvious that the same ordinal's extremum moves towards the direction of high frequency when thick of thin bed decreases. Some new conclusions are as follows: (1) value of tuning thickness is notλ/4, and less than it. (2)Change velocity in the thin bed, time thickness with tuning thickness is not affected (3) The synthetic examples of previous section show we can generate these components in the seismic response introducing the same type of components into the depth model as building blocks of impedance stratification. The variation of frequency content of a seismic trace with time carries information about the properties of the subsurface reflectivity sequence. The time-frequency representation provides a good tool to extract information on gross pattern of stratification. These gradual changes in layer thicknesses can be recognized in the time frequency representation of the seismic signal as a change of frequency content as a function of seismic travel time. (4) On base of old work, I study geological gyration by time frequency analysis. Time frequency analysis by match pursuit wavelet packet has better resolution than by short window discrete fourier transform, and the new method does not think about length of time window, but its amount of calculation is more than SDFT. The instant spectrum attributes on MPWP carry information about lithology constitute. Frequency in positive gyration moves towards low frequency from high frequency; Frequency in negative gyration moves towards high frequency from low frequency; Frequency in positive gyration added negative gyration moves from low frequency to high frequency and then do from high frequency to high frequency. Here should point out: the relation between instant spectrum attributes and geological gyration is not simple and unique, but it is useful information of geological gyration and become one important supplement of amplitude information.In the end, the dissertation summarizes the main results, and then analyzes and discusses the further issues required to study on the time-frequency analysis method for the signal in seismic prospecting and other applications of the method. The time-frequency analysis method researches the seismic wave condition changing with the time. The well corresponding relationship between the time-frequency spectrum and the geologic character provides fine explanation for the seismic geologic, and then provides rich information for geologic explanation.
|
PDF Full Text Request