CN-122017982-A - Method and device for predicting gas-containing property of tight sandstone reservoir through feature space guided modeling

CN122017982ACN 122017982 ACN122017982 ACN 122017982ACN-122017982-A

Abstract

The invention relates to the technical field of exploration geophysics and artificial intelligence deep learning, and discloses a method and a device for predicting gas content of a tight sandstone reservoir through feature space guided modeling. The method comprises the steps of establishing a general paradigm of predicting the gas content of a tight sandstone reservoir based on a pre-stack seismic trace set and a well logging gas-containing curve by a supervised learning method, establishing a feature space guiding modeling method, and utilizing a kNN method to conduct preliminary gas-containing prediction on a target work area, screening a high-probability kNN prediction result as pseudo training data, utilizing the pseudo training data to pretrain a CNN model, and utilizing an actual training sample to conduct migration learning optimization on a stable CNN model to achieve gas-containing prediction on the tight sandstone reservoir. The technical scheme of the invention is beneficial to exerting the advantage of deep learning on the depicting ability of complex nonlinear relations, and the mapping relation between the seismic response and the gas content of the tight sandstone reservoir is clear, and finally the prediction precision of the tight reservoir is improved.

Inventors

SONG CHAOHUI
SHI LEI
WANG ZHENYU

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石油勘探开发研究院

Dates

Publication Date: 20260512
Application Date: 20241112

Claims (10)

1. A method for predicting gas-bearing properties of a tight sandstone reservoir modeled by feature space guidance, comprising: establishing a general paradigm of predicting the gas content of the tight sandstone reservoir based on a prestack seismic gather and a well logging gas content curve by a supervised learning method; The method for establishing the feature space guiding modeling comprises the following steps: preliminary gas-containing prediction is carried out on the target work area by utilizing a kNN method, and a kNN prediction result with high probability is screened to be used as pseudo training data; Pre-training the CNN model by using the pseudo training data, wherein the information of the feature space of the data sample carried in the pseudo training data is used for guiding to establish an initial model of the CNN model; and performing migration learning optimization on the stable CNN model by using an actual training sample, and realizing the gas-containing prediction of the tight sandstone reservoir with the feature space guided modeling.
2. The feature space guided modeled tight sandstone reservoir gas-content prediction method of claim 1, wherein the supervised learning method is established from a generic paradigm of tight sandstone reservoir gas-content prediction based on pre-stack seismic trace sets and well-log gas-content curves, the generic paradigm being applicable to both machine learning methods and deep learning methods, And classifying the seismic sampling points into gas-containing sampling points and non-gas-containing sampling points by taking a gas-containing curve obtained by well logging interpretation as a tag and a local waveform response as a characteristic.
3. The tight sandstone reservoir gas-containing prediction method of feature space guided modeling according to claim 1, wherein the process of performing classification tasks by the kNN method is a process of performing space division on the feature space based on labeled samples, wherein the kNN method selects different k neighbor training samples for each test sample to make a decision, the number of neighbors in the decision process is k value, when a small k value is selected, the kNN method establishes a more targeted prediction model for a single test sample, and when a large k value is selected, more training samples are introduced to make a reference.
4. The method for predicting gas content of a tight sandstone reservoir modeled by feature space guidance according to claim 1, wherein in the initial model for establishing the CNN model, feature space information is transferred to the CNN modeling process in the form of pseudo training samples, and the information of the CNN modeling process is supplemented and constrained.
5. The feature space guided modeled tight sandstone reservoir gas-containing prediction method of claim 1, wherein said stable CNN model is obtained by repeating a pseudo-training data screening and CNN model training process.
6. The tight sandstone reservoir gas-containing prediction method of feature space guided modeling of claim 1, wherein said CNN model comprises an input layer, a hidden layer and an output layer, wherein said hidden layer comprises a convolutional layer, an activation function, a pooling layer and a fully connected layer, wherein said convolutional layer comprises a neuron connection specific to a convolutional neural network, wherein the reverse transmission of said convolutional neural network adjusts weights and offsets by minimizing residuals, such that said convolutional neural network is continuously updated according to training data, and wherein the reverse propagation algorithm uses the idea of random gradient descent.
7. The tight sandstone reservoir gas-containing prediction method of any of claims 1-6, wherein said kNN method and said CNN model, when applied separately to tight sandstone reservoir gas-containing prediction tasks, follow the same set of common paradigms for tight sandstone reservoir gas-containing prediction based on supervised learning methods.
8. A tight sandstone reservoir gas-containing prediction device for feature space guided modeling, comprising: a first establishing module for establishing a general paradigm of a supervised learning method for predicting gas contents of a tight sandstone reservoir based on pre-stack seismic gathers and a log gas-content curve, and The second building module is used for building a feature space guide modeling method; Wherein the second establishing module includes: The screening submodule is used for carrying out preliminary gas-containing prediction on the target work area by utilizing a kNN method, and screening a high-probability kNN prediction result as pseudo training data; a pre-training sub-module for pre-training the CNN model by using the pseudo-training data, wherein the initial model of the CNN model is built by using information guidance of a data sample feature space carried in the pseudo-training data, and And the migration learning sub-module is used for performing migration learning optimization on the stable CNN model by using an actual training sample, and realizing the gas-containing prediction of the tight sandstone reservoir with the feature space guided modeling.
9. An electronic device, comprising: Memory, and A processor; Wherein the memory is for storing one or more computer instructions for execution by the processor to implement the method of any one of claims 1 to 7.
10. A readable storage medium having stored thereon computer instructions, wherein the computer instructions, when executed by a processor, implement the method of any of claims 1 to 7.

Description

Method and device for predicting gas-containing property of tight sandstone reservoir through feature space guided modeling Technical Field The invention relates to the technical field of exploration geophysics and artificial intelligence deep learning, in particular to a method and a device for predicting gas content of a tight sandstone reservoir through feature space guided modeling. Background Among existing seismic reservoir technologies, conventional seismic reservoir prediction technologies include seismic attribute analysis technology, seismic inversion technology, and AVO (Amplitude Variation with Offset, AVO) technology. Seismic attribute analysis techniques are one of the most commonly used reservoir prediction techniques. The seismic attribute refers to characteristic parameters extracted from seismic data through mathematical means, and can comprehensively reflect the changes of the seismic data in geometric aspects, statistics aspects, dynamics aspects, kinematics aspects and the like. These changes are closely related to the spatial changes in reservoir physical and fluid properties. Reservoir properties are predicted based on seismic attribute analysis techniques, and the seismic attributes need to be calibrated by using reservoir properties obtained by log interpretation. The method flow can be summarized as extracting various seismic attributes from the well side seismic data, performing correlation analysis with reservoir physical properties obtained by well logging interpretation to optimize sensitive attributes, calibrating the sensitive attributes by using the reservoir physical properties obtained by well logging, and extrapolating the calibration result from the well point position to the whole work area. Thus, the geological significance of the seismic attribute comes from the calibration. Without calibration process with logging reservoir properties, seismic attributes are purely mathematical methods to obtain geophysical parameters. However, the seismic response of subsurface reservoirs is affected by a variety of factors, such as the thickness of the reservoir, the subsurface structure, the nature of the fluid, etc., and multiple solutions are expected to exist. Obtaining subsurface elastic parameters or further fluid indicators through seismic inversion techniques is a widely used reservoir prediction technique in the industry, and seismic inversion is the process of estimating subsurface features from seismic data. Generalized seismic inversion techniques are capable of quantitatively calculating various geophysical parameters of a formation based on seismic data. Since wave impedance inversion has a definite physical meaning and has achieved a good effect in actual reservoir prediction, a narrow-sense seismic inversion technique is particularly referred to as a seismic wave impedance inversion technique. The wave impedance inversion results typically have a higher resolution than the original seismic data. The seismic inversion technology can be divided into narrow-band inversion, inversion based on well logging constraint, well logging interpolation extrapolation based on seismic constraint and other technologies according to different utilization modes of well seismic data. Depending on the type of seismic data, it can be classified into pre-stack inversion and post-stack inversion. However, the seismic inversion technology itself faces the problems of inaccurate wavelet acquisition, dependence on an initial model and the like. In practical application, the band-limited characteristics of the seismic data and the field noise interference further increase the difficulty of seismic inversion. The AVO technology refers to a technology for analyzing an underground reservoir by utilizing the change rule of seismic reflection amplitude along with offset distance, and can be divided into an AVO forward analysis technology and an AVO inversion analysis technology. The theoretical basis of AVO technology is the Zoeppritz equation. The seismic reflection coefficient is determined by the medium properties at two sides of the interface, and hydrocarbon detection or lithology distinction can be performed directly from the seismic record qualitatively by determining AVO characteristics of reservoirs under different lithology and physical properties. Based on this knowledge, AVO forward technology was developed. However, complicated underground conditions do not have a good one-to-one correspondence between AVO anomalies and lithology and physical properties of the reservoir, resulting in poor accuracy in predicting the reservoir using AVO forward technology. The AVO inversion technique establishes a direct link between certain formation parameters and reflection coefficients by approximating and simplifying the Zoeppritz equation. However, the assumptions proposed to achieve the approximation to the Zoeppritz equation limit the range of application of the gas-containing predictions, which