CN-122017945-A - Seismic wave impedance inversion method and related device

CN122017945ACN 122017945 ACN122017945 ACN 122017945ACN-122017945-A

Abstract

The invention discloses a seismic wave impedance inversion method and a related device, which are used for acquiring seismic data to be inverted, inputting the seismic data to be inverted into a pre-trained bidirectional time sequence convolution neural network model to obtain predicted wave impedance data, wherein the bidirectional time sequence convolution neural network model comprises a forward time sequence convolution branch, a reverse time sequence convolution branch and a feature merging layer, the forward time sequence convolution branch and the reverse time sequence convolution branch comprise a plurality of residual blocks, each residual block uses expansion convolution, the forward time sequence convolution branch is used for processing an input seismic data sequence according to a positive sequence, the reverse time sequence convolution branch is used for processing the input seismic data sequence according to a reverse sequence, and the feature merging layer is used for merging outputs of the forward time sequence convolution branch and the reverse time sequence convolution branch. The invention aims to solve the problems that in the existing inversion method based on deep learning, a convolution neural network is difficult to capture a long-range dependency relationship, the calculation force requirement of a circulation neural network is high, gradient problems are easy to occur, and unidirectional sequence modeling does not fully utilize bi-directional context information so as to restrict inversion accuracy.

Inventors

ZHOU LIYE
HE XIAOYAN
HE MEI
ZHAO DONGFENG
LIANG SHASHA
LI HAO
WANG LEI
LI XIAOYUE

Assignees

地球脉动(宁波)科技有限公司
宁波东方理工产业技术研究有限公司

Dates

Publication Date: 20260512
Application Date: 20260123

Claims (10)

1. A method of seismic wave impedance inversion, comprising: acquiring seismic data to be inverted; inputting the seismic data to be inverted into a pre-trained bidirectional time sequence convolutional neural network model to obtain predicted wave impedance data; the bidirectional time sequence convolution neural network model comprises a forward time sequence convolution branch, a reverse time sequence convolution branch and a feature merging layer, wherein the forward time sequence convolution branch and the reverse time sequence convolution branch comprise a plurality of residual blocks, expansion convolution is used in each residual block, the forward time sequence convolution branch is used for processing an input seismic data sequence according to a positive sequence, the reverse time sequence convolution branch is used for processing the input seismic data sequence according to a reverse sequence, and the feature merging layer is used for merging outputs of the forward time sequence convolution branch and the reverse time sequence convolution branch.
2. A method of seismic wave impedance inversion according to claim 1 wherein the seismic data to be inverted is normalized prior to being input into the two-way time series convolutional neural network model, the normalization comprising converting the seismic data to be inverted into data conforming to a standard normal distribution.
3. A seismic wave impedance inversion method according to claim 1, wherein the expansion rate of said expansion convolution increases exponentially with increasing network depth.
4. A seismic wave impedance inversion method according to claim 1, wherein said feature combination layer is followed by an attention mechanism for weighting the combined features.
5. The seismic wave impedance inversion method of claim 1, wherein the two-way time series convolutional neural network model is trained by using training sample pairs, the training sample pairs comprising a seismic data sequence and corresponding real wave impedance data, and the seismic data sequence and the real wave impedance data are normalized.
6. A seismic wave impedance inversion method according to claim 5 wherein the seismic data sequence and the true wave impedance data in the training sample pair are windowed centered at the well location and the sequence lengths are the same.
7. A seismic wave impedance inversion method according to claim 5 wherein said training process uses an early stop mechanism and a learning rate decay strategy.
8. A seismic wave impedance inversion apparatus, comprising: The acquisition module is used for acquiring the seismic data to be inverted; The prediction module is used for inputting the seismic data to be inverted into a pre-trained bidirectional time sequence convolutional neural network model to obtain predicted wave impedance data; the bidirectional time sequence convolution neural network model comprises a forward time sequence convolution branch, a reverse time sequence convolution branch and a feature merging layer, wherein the forward time sequence convolution branch and the reverse time sequence convolution branch comprise a plurality of residual blocks, expansion convolution is used in each residual block, the forward time sequence convolution branch is used for processing an input seismic data sequence according to a positive sequence, the reverse time sequence convolution branch is used for processing the input seismic data sequence according to a reverse sequence, and the feature merging layer is used for merging outputs of the forward time sequence convolution branch and the reverse time sequence convolution branch.
9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements a seismic wave impedance inversion method as claimed in any one of claims 1 to 7 when the computer program is executed.
10. A computer readable storage medium storing a computer program, which when executed by a processor implements a seismic wave impedance inversion method according to any one of claims 1 to 7.

Description

Seismic wave impedance inversion method and related device Technical Field The invention belongs to the technical field of geophysical exploration and artificial intelligence intersection, and particularly relates to a seismic wave impedance inversion method and a related device. Background Wave impedance inversion is one of the key technologies in seismic exploration, and its task is to convert seismic recordings into wave impedance data, revealing subsurface lithology and fluid distribution characteristics. At present, the traditional wave impedance inversion method mainly comprises channel integral inversion, sparse pulse inversion, model-based inversion and the like. The method retains reflection coefficient information by eliminating wavelet influence in seismic data, and further obtains physical parameters representing stratum physical property change, namely wave impedance through reflection coefficient calculation. However, the method is often dependent on theoretical approximation in the implementation process, is sensitive to an initial model and an assumed condition, and has the problems of high computational complexity and low efficiency. In recent years, with the development of artificial intelligence technology, deep learning is gradually introduced into seismic wave impedance inversion. For example, convolutional neural networks, cyclic neural networks and other methods have been initially applied to achieve certain effects. However, the existing inversion method based on deep learning still has the following limitations that firstly, the receptive field of a convolutional neural network is limited, long-range dependency relationship in seismic data is difficult to capture effectively, secondly, the cyclic neural network has high calculation force requirements in the training process, and the problem of gradient disappearance or gradient explosion is easy to occur, and furthermore, the convolutional neural network and the cyclic neural network only adopt a unidirectional sequence modeling mode, bidirectional context information contained in the seismic data cannot be fully utilized, and inversion accuracy is restricted. Disclosure of Invention Aiming at the problems existing in the prior art, the invention provides a seismic wave impedance inversion method and a related device, and aims to solve the problems that a convolutional neural network is difficult to capture a long-range dependency relationship, a cyclic neural network has high calculation force requirement and is easy to cause gradient problems, and unidirectional sequence modeling does not fully utilize bi-directional context information so as to restrict inversion accuracy in the existing inversion method based on deep learning, so that the accuracy and efficiency of seismic wave impedance inversion are improved. In order to solve the technical problems, the invention is realized by the following technical scheme: according to a first aspect of the present invention there is provided a seismic wave impedance inversion method comprising: acquiring seismic data to be inverted; inputting the seismic data to be inverted into a pre-trained bidirectional time sequence convolutional neural network model to obtain predicted wave impedance data; the bidirectional time sequence convolution neural network model comprises a forward time sequence convolution branch, a reverse time sequence convolution branch and a feature merging layer, wherein the forward time sequence convolution branch and the reverse time sequence convolution branch comprise a plurality of residual blocks, expansion convolution is used in each residual block, the forward time sequence convolution branch is used for processing an input seismic data sequence according to a positive sequence, the reverse time sequence convolution branch is used for processing the input seismic data sequence according to a reverse sequence, and the feature merging layer is used for merging outputs of the forward time sequence convolution branch and the reverse time sequence convolution branch. In a possible implementation manner of the first aspect, the seismic data to be inverted is subjected to a normalization process before being input into the bidirectional time sequence convolutional neural network model, and the normalization process includes converting the seismic data to be inverted into data conforming to a standard normal distribution. In a possible implementation manner of the first aspect, the expansion rate of the expansion convolution increases exponentially with the network depth. In a possible implementation manner of the first aspect, the feature merging layer further introduces an attention mechanism for weighting the merged features. In a possible implementation manner of the first aspect, the bidirectional time sequence convolutional neural network model is obtained by training using a training sample pair, wherein the training sample pair comprises a seismic data sequence