CN-121995470-A - Least square seismic time-frequency analysis method and device with sparse constraint

CN121995470ACN 121995470 ACN121995470 ACN 121995470ACN-121995470-A

Abstract

The embodiment of the invention relates to a sparse constraint least square seismic time-frequency analysis method and device, which comprises the steps of carrying out complex processing on a real number seismic trace to construct a complex seismic trace, constructing a diagonal matrix with data weight being a diagonal line, initializing model weight as a unit matrix, constructing Euler formula kernel matrixes formed by different cut-off frequencies and time ranges in a time domain, applying the data weight constraint to the complex seismic trace, applying the data weight constraint and the model weight constraint to the Euler formula kernel matrix, constructing an error function among the constrained Euler formula kernel matrix, model parameters and the constrained complex seismic trace, applying L1 regularized sparse constraint to the model parameters, and inverting by using a greedy iterative shrinkage threshold method to obtain a sparse constraint least square seismic time-frequency analysis result. Therefore, the earthquake signal can be efficiently and accurately decomposed to obtain a high-precision time spectrum.

Inventors

WANG YANGUANG
WEI GUOHUA
LIU HAOJIE
CAO HAIFANG
GAI PANPAN
YANG ZHAOGANG
LI JIN

Assignees

中国石油化工股份有限公司
中国石油化工股份有限公司胜利油田分公司

Dates

Publication Date: 20260508
Application Date: 20241104

Claims (10)

1. The least square seismic time-frequency analysis method of the sparse constraint is characterized by comprising the following steps of: carrying out complex processing on the real number seismic trace to construct a complex seismic trace; constructing a diagonal matrix with data weights as window functions as diagonal lines, and initializing model weights as an identity matrix; Constructing Euler formula kernel matrixes formed by different cut-off frequencies and time ranges in a time domain; applying the data weight constraint to the complex seismic trace, and applying a data weight constraint and a model weight constraint to the euler formula kernel matrix; Constructing a constrained Euler formula kernel matrix, model parameters and an error function between constrained complex seismic traces, applying L1 regularized sparse constraint to the model parameters, and inverting by using a greedy iterative shrinkage threshold method to obtain a least square seismic time-frequency analysis result of the sparse constraint.
2. The method of claim 1, wherein the complex processing of the real traces to construct complex traces comprises: Expressing a real seismic trace as a real part and an imaginary part, adding Hilbert transform as an additional constraint, and constructing a complex seismic trace by a first formula, wherein the first formula is d=d r +id i ; Where d is the windowed segment of the complex trace, d r is the windowed segment of the real trace, and d i is the windowed segment of the Hilbert transform on the trace.
3. The method of claim 2, wherein constructing the diagonal matrix with the data weights being diagonal to the window function and initializing the model weights to the identity matrix comprises: constructing a data weight through a second formula, wherein the second formula is as follows: Where W d is the data weight, nDeltat is the time relative to the window center, abs (d 0 ) is the instantaneous amplitude of the window function center position, and l is the window length; Initializing a model weight as an identity matrix through a third formula, wherein the third formula is that W m =I; wherein W m is model weight, and I is identity matrix.
4. A method according to claim 3, wherein said constructing, in the time domain, a euler formula kernel matrix consisting of different cut-off frequencies and time ranges, comprises: constructing Euler formula kernel matrixes formed by different cut-off frequencies and time ranges in a time domain through a fourth formula, wherein the fourth formula is that F (t, F) =cos (2pi k delta fm delta t) + isin (2pi k delta fm delta t); the euler formula kernel matrix F is a complex sinusoidal signal with a window length, the number k of columns in F is a frequency number, and the number m of rows in F is a sample number in a time window.
5. The method of claim 4, wherein the applying the data weight constraints on the complex seismic traces, applying data weight constraints and model weight constraints on the euler equation kernel matrix, comprises: applying a data weight constraint W d d to the complex seismic trace; Applying data weight constraint and model weight constraint to the Euler formula kernel matrix through a fifth formula, wherein the fifth formula is F w ＝W d FW m ; Wherein F w is an Euler formula kernel matrix after data weight constraint and model weight constraint are applied, W d is a data weight, W m is a model weight, and F is an Euler formula kernel matrix; Weighting the model parameter vector by a fifth formula: Where m is an unknown model parameter vector, i.e. the amplitude spectral coefficients that need to be calculated, and m w is a weighted model parameter vector.
6. The method of claim 5, wherein constructing the constrained euler equation kernel matrix, model parameters, and error functions between constrained complex seismic traces, and applying L1 regularized sparsity constraints to the model parameters, comprises: the Euler formula kernel matrix after the constraint is constructed, model parameters and an error function between complex seismic traces after the constraint are constructed, and L1 regularization sparse constraint is applied to the model parameters, as shown in a formula 7:
7. The method of claim 6, wherein applying greedy iterative shrinkage thresholding inversion to obtain sparsely constrained least squares seismic time-frequency analysis results comprises: Initializing an iterative solution x k 、y k , and enabling a step length gamma 0 =1, S >1 and xi <1 to iteratively calculate x k+1 through an eighth formula and a ninth formula, wherein the eighth formula is y k ＝x k +(x k -x k-1 ), and the ninth formula is: If (y k -x k+1 ) T (x k+1 -x k ) is not less than 0, y k ＝x k ; If x k+1 -x k ||≥S||x 1 -x 0 , γ=max (ζ, γ 1 ); if the value of the iteration is equal to the value of the iteration x k+1 -x k < epsilon, outputting an iterative solution x k+1 , otherwise repeating the iterative step.
8. A sparsely constrained least squares seismic time-frequency analysis apparatus, comprising: The construction module is used for carrying out complex processing on the real number seismic channels to construct complex seismic channels; The construction module is used for constructing a diagonal matrix with data weight being a window function being a diagonal, and initializing model weight as an identity matrix; The construction module is used for constructing Euler formula kernel matrixes formed by different cut-off frequencies and time ranges in a time domain; The constraint module is used for applying the data weight constraint to the complex seismic trace and applying the data weight constraint and the model weight constraint to the Euler formula kernel matrix; The analysis module is used for constructing an error function among the constrained Euler formula kernel matrix, the model parameters and the constrained complex seismic traces, applying L1 regularized sparse constraint to the model parameters, and inverting by using a greedy iterative shrinkage threshold method to obtain a least square seismic time-frequency analysis result of the sparse constraint.
9. An electronic device comprising a processor and a memory, wherein the processor is configured to execute a sparsely constrained least squares seismic time-frequency analysis program stored in the memory to implement the sparsely constrained least squares seismic time-frequency analysis method of any one of claims 1-7.
10. A storage medium storing one or more programs executable by one or more processors to implement the sparsely constrained least squares seismic time-frequency analysis method of any one of claims 1-7.

Description

Least square seismic time-frequency analysis method and device with sparse constraint Technical Field The embodiment of the invention relates to the technical field of seismic exploration and signal processing, in particular to a sparse constraint least square seismic time-frequency analysis method and device. Background As exploration targets gradually turn to thinner, smaller, unconventional and other complex reservoirs, it is necessary to extract attributes from complex seismic data wave packet features that reflect the changes in the tiny structural and lithologic properties. However, the conventional seismic data in the existing time domain has low resolution, is insufficient in capability for solving finer and more complex oil reservoir descriptions, is difficult to realize the fine description of a target layer, and is one of the basic problems for restricting the fine description of the current oil and gas reservoirs. Aiming at the problem that the resolution of the time domain seismic data is low and the complex fine structure is difficult to explain, the frequency-time-space characteristics of the seismic signals can be analyzed through a spectrum decomposition technology, and the amplitude anomaly attribute is extracted to indicate a thin layer or attenuate an anomaly geologic body. However, the conventional method is limited by the length of a time window, generates window tailing effect or is interfered by surrounding seismic event energy, has the defects of lower resolution or larger calculated amount, and cannot meet the current high-precision reservoir identification requirement. Therefore, how to more accurately invert the amplitude spectrum of the seismic reflection wave, and simultaneously maintain higher calculation efficiency, and realizing stable decomposition of the seismic data is one of the core problems for improving the interpretability of the seismic data. The existing similar time-frequency decomposition method directly calculates the Fourier series coefficient as a function of time by inverting the basis of a truncated Euler formula core of a moving time window. The method can obtain a time spectrum with higher time and frequency resolution, can improve the limitation of the traditional short-time Fourier transform by the length of a time window, weaken the tailing effect of the window or the energy interference of surrounding seismic events, and can also improve the time resolution problem of continuous wavelet transform at low frequency. However, the noise immunity of the method is weak, and the Gaussian elimination method is adopted to perform large matrix calculation, so that the calculation efficiency under large-scale seismic data is required to be improved. In the existing Wigner-Violet time-frequency decomposition method based on energy separation, a similar time-frequency decomposition method is mentioned, wherein an original one-dimensional signal x (t) is subjected to Fourier-Bezier series expansion, zero-order Fourier-Bezier series expansion and Fourier-Bezier coefficient sequences of the original one-dimensional signal x (t) are obtained, an amplitude envelope of the Fourier-Bezier coefficient sequences is calculated by an energy separation algorithm, M groups of Fourier-Bezier coefficient subsequences are obtained, fourier-Bezier transformation is performed on the Fourier-Bezier coefficient subsequences, a time domain sub-signal xi (t) of each subsequence is obtained, wigner-Violet distribution of the time domain sub-signal xi (t) of each subsequence is obtained, and Wigner-Violet distribution of all the sub-signals is overlapped, so that Wigner-Violet time spectrum of the one-dimensional signal x (t) without cross terms is obtained. The invention can inhibit the generation of cross terms, so that the obtained time spectrum is more accurate. However, the method uses the amplitude envelope in the energy separation algorithm to calculate the subsequence of the fourier-bessel coefficient sequence, which introduces a certain error, especially for low energy signals or noise signals, with the risk of information loss. In addition, the generation of cross terms cannot be completely accurately suppressed for non-stationary signals or signals with abrupt changes. The conventional multi-impact vibration signal time-frequency decomposition method comprises the steps of performing short-time Fourier transform on a multi-impact vibration signal, transforming the multi-impact vibration signal into a time-frequency domain signal, establishing a quadratic convex optimization model with the square of the 2 norm of a time-frequency two-dimensional amplitude moment of the time-frequency signal as an optimization target, converting a multi-impact signal time-frequency decomposition problem into a variation problem of an impact time-frequency dimensional center and a decomposition time-frequency sub-signal, then performing iterative solution on the solution problem into the decomposi