Search

CN-122020145-A - Intelligent frequency expansion method for post-stack seismic data

CN122020145ACN 122020145 ACN122020145 ACN 122020145ACN-122020145-A

Abstract

The invention provides an intelligent frequency expansion method for post-stack seismic data, which relates to the field of seismic exploration and development and comprises the following steps of (1) building a network, namely building a deep neural network by adopting a coding and decoding framework combining a convolutional neural network and a cyclic neural network; training and predicting, namely training the deep neural network obtained in the step (1) by using the synthetic data set obtained in the step (2) to obtain a trained network model, inputting the seismic data to be developed into the trained network model to obtain the seismic data with development frequency, and widening the frequency band of the seismic data predicted by the trained network model by 10% -15%. The intelligent frequency extension method for post-stack seismic data combines the seismic data and the logging data to extend the frequency, broadband information of the logging data is introduced into the seismic data, and the well-shock correlation of the seismic data after the frequency extension is high and the reliability is high.

Inventors

  • CAO FEI
  • CHEN ZHIMING
  • LU ZHIQIANG
  • HE XINMING
  • BAO DIAN
  • Nigati Abby Bra

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西北油田分公司

Dates

Publication Date
20260512
Application Date
20241111

Claims (10)

  1. 1. A post-stack seismic data intelligent frequency expansion method is characterized by comprising the following steps: (1) Constructing a network, namely constructing a deep neural network by adopting an encoding and decoding architecture combining a convolutional neural network and a cyclic neural network; (2) Generating a synthetic training data set comprising input data and tag data, generating a data set from convolutions of reflection coefficients and seismic wavelet data, adding noise, generating input data related to actual data features; (3) Training and predicting, namely training the deep neural network obtained in the step (1) by using the synthetic data obtained in the step (2) to obtain a trained network model, inputting the to-be-developed seismic data into the trained network model to obtain the developed seismic data, and widening the frequency band of the seismic data predicted by the trained network model by 10% -15%.
  2. 2. The intelligent frequency extension method for post-stack seismic data according to claim 1, wherein the encoding part in the step (1) is composed of two convolution layers, a maximum pooling layer is arranged behind the convolution layers, downsampling is carried out through the maximum pooling layer, and the decoding part comprises upsampling layers, and two layers of bidirectional convolution LSTM are arranged behind each upsampling layer.
  3. 3. The intelligent frequency extension method for post-stack seismic data according to claim 2, wherein the up-sampling layer adopts a difference value plus convolution method, and the post-maximum pooling layer adopts batch standardization to accelerate the network convergence speed.
  4. 4. The intelligent frequency expansion method for post-stack seismic data according to claim 1, wherein the step (2) comprises the following steps: Analyzing the actual data characteristics, and extracting wavelet information from the seismic data to obtain seismic wavelet data; generating a reflection coefficient, namely performing time-depth conversion on logging data through well shock calibration, resampling a P-wave speed and density curve, and calculating to obtain the reflection coefficient; Where r is the reflection coefficient, ρ is the density, v is the speed, i=1, 2, 3; Generating a dataset from convolutions of the reflection coefficients and the seismic wavelet data, adding noise to generate input data related to the actual data features, and generating trained tag data from convolutions of the reflection coefficients and the broadband B-spline wavelet data.
  5. 5. The intelligent frequency extension method for post-stack seismic data according to claim 4, wherein the sampling rate of the P-wave velocity and density curve is the same as the sampling rate of the seismic data.
  6. 6. The method of intelligent frequency extension of post-stack seismic data of claim 5, wherein the actual data characteristics comprise frequency content of the seismic wavelet, frequency shape of the seismic wavelet, content of reflection coefficient and content of random noise.
  7. 7. The intelligent frequency extension method for post-stack seismic data according to claim 1, wherein the method is characterized by involution The training data set and the actual data are normalized, and the normalization formula is as follows: Wherein x min and x max are the minimum and maximum values of the synthetic training data set or the actual data, respectively, x is the original input synthetic training data set or the original input actual data, and x' is the synthetic training data set or the actual data after normalization.
  8. 8. The intelligent frequency extension method for post-stack seismic data according to claim 1, wherein the training process in the step (3) is characterized in that a gradient descent algorithm in a back propagation algorithm is adopted to update and learn a deep neural network, the back propagation algorithm is that initial weight values of the deep neural network are random initially, input data are input into the deep neural network and then are calculated with the weight values of the deep neural network to obtain a deep neural network output value, the deep neural network output value and trained tag data are calculated through a loss function to obtain an error, the error is back propagated to the deep neural network, the weight values of the deep neural network are iteratively updated according to the gradient descent algorithm, and the above processes are circulated until the error meets the accuracy requirement.
  9. 9. The intelligent frequency extension method for post-stack seismic data according to claim 8, wherein the loss function is calculated by using a symmetrical average absolute percentage error SMAPE, and the formula is: Wherein X i is the output value of the deep neural network, Y i is the trained label data, n is the number of labels, and the value range of i is 1 to n.
  10. 10. The intelligent frequency extension method for post-stack seismic data according to claim 9, wherein the weight value of the deep neural network is iteratively updated according to a gradient descent algorithm by adopting an Adam optimizer algorithm.

Description

Intelligent frequency expansion method for post-stack seismic data Technical Field The invention relates to the field of seismic exploration and development, in particular to an intelligent frequency expansion method for post-stack seismic data. Background The vertical resolution of the seismic data is determined by the bandwidth and dominant frequency of the data. Because of the limited width of the source spectrum, the seismic signal is also band limited. During the propagation of the seismic wave, high-frequency components are severely attenuated due to the absorption of the stratum. Thus, conventionally acquired seismic data lacks efficient high or low frequency information and is not sufficiently resolved. The resolution of conventionally acquired seismic data is insufficient for seismic interpretation purposes. Typically, one defines a horizon as a thin layer that is less than a quarter wavelength of the dominant frequency of the data. For such horizons, the top and bottom of the reflected wave will be tuned and therefore indistinguishable. In practical seismic data, it is often desirable to improve seismic data resolution due to the presence of many thin reservoirs or micro-formations. Conventional seismic data frequency expansion methods generally require that the seismic data meet different assumption conditions to achieve a good frequency expansion effect. However, the actual seismic data in different working areas has large difference, and various noises, complex underground structures and the like often cause that the seismic data are difficult to meet the assumed conditions of the conventional frequency expansion method, so that the frequency expansion precision is affected. In addition, most conventional frequency extension methods only use seismic data, often introduce various artifacts, and the frequency extension result has lower precision. The deep learning algorithm has received much attention from geophysicists in recent years, and is applied to various geophysical problems requiring generalization because it is a data-driven method. The Chinese patent with publication number CN116643310A discloses a method, a device, an electronic device and a computer readable storage medium for seismic frequency extension, wherein the method comprises the steps of obtaining first intermediate frequency data and first full frequency data in seismic sample data; the first intermediate frequency data is used for representing data in a first frequency range in the seismic sample data, the first full frequency data is used for representing data in a second frequency range in the seismic sample data, a first neural network is trained according to the first intermediate frequency data and the first full frequency data, and seismic frequency expansion is conducted on the second intermediate frequency data according to the trained first neural network to obtain predicted second full frequency data. The invention provides the full-band data of the sample seismic data as the training data to perform seismic frequency expansion, and the low-frequency information and the high-frequency information of the seismic data are recovered with lower cost, so that the full-waveform inversion precision is improved, and the resolution of the seismic data is improved. The Chinese patent with publication number CN113721294A discloses a complex domain least square constraint spectrum bluing frequency-expanding method, and proposes to design a broadband target spectrum to widen the spectrum from the frequency domain. In particular to a complex domain least square constrained spectrum bluing frequency-expanding method. The method comprises the steps of establishing an objective function of frequency extension based on a complex domain least square method, solving the objective function to obtain a spectrum bluing frequency extension operator, multiplying the spectrum bluing frequency extension operator with an original seismic spectrum, and then performing Fourier inverse transformation to realize frequency extension processing of seismic data. The method can effectively compensate lost low-frequency information and enhance attenuated high-frequency information under the condition of not changing the phase spectrum of the seismic data, and can adjust the distance between the frequency spectrum of the seismic data after frequency expansion and the broadband target spectrum by setting different constraint parameters so as to obtain frequency expansion results with different main frequencies and bandwidths. For seismic data frequency development processing, the key problem is the reliability of the frequency development data. However, the method only uses the seismic data, various false images are often introduced, well seismic information is not matched, and aiming at the problems, the method for intelligently expanding the frequency of the post-stack seismic data, which has high well seismic correlation and high reliability, is ne