Reservoir seismic information extraction method for oil and gas detection

Abstract

The invention discloses a reservoir seismic information extraction method for oil and gas detection, and relates to the technical field of reservoir prediction. The method comprises the following steps: performing high-fidelity weak-keeping signal processing on seismic data in combination with domain knowledge and seismic geological conditions of a target area; performing cepstrum feature extraction on the seismic signals by combining multivariate variational mode decomposition with cepstrum analysis; meanwhile, a shared depth self-encoder and a pulse neural network are combined to extract depth features of the seismic signals; aiming at seismic signals of different dimensions and different types, respectively designing feature extraction schemes suitable for respective data type features, and carrying out visual comparison analysis by adopting a feature visualization technology; and carrying out multi-seismic information fusion analysis by adopting a self-organizing mapping network and combining actual exploration data to obtain a reservoir oil-gas possibility detection result. Compared with the prior art, the method has the advantages that weak change or difference information of reservoir fluid seismic response can be extracted more comprehensively, and then the accuracy of reservoir oil-gas possibility detection is improved.

Description

Technical Field

[0001] The present invention relates to the technical fields of seismic data processing and reservoir prediction, and particularly relates to a method for extracting reservoir seismic information for oil and gas detection. Background Art

[0002] Seismic reservoir prediction is an important part of oil and gas exploration, and the detection of reservoir oil and gas content is the key among them. Exploration practice shows that existing reservoir detection methods such as bright spot identification, AVO anomaly analysis, and their derivative methods, etc., all have certain applicable conditions, with both successful cases and complete failure records. For example, bright spot identification is mainly applicable to shallow loose clastic reservoirs. And AVO anomaly analysis requires accurately knowing the incident angle of seismic waves. Under complex or deeply buried conditions, it is more difficult to accurately estimate the incident angle of seismic waves, and these estimation errors will cause false anomalies or submerge true anomalies, resulting in incorrect predictions.

[0003] Existing theoretical methods for seismic detection of reservoir oil and gas content based on rock physics modeling and seismic wavefield forward and inverse simulations are applicable to situations where there are large differences in physical properties between reservoirs and non-reservoirs and the signal-to-noise ratio of seismic data is relatively high. Deep oil and gas reservoir bodies have strong heterogeneity in lithology and structure. However, due to factors such as large burial depth, the physical property differences between reservoirs and non-reservoirs are very small, showing weak heterogeneity seismically. The responses of reservoir and reservoir pore fluids are both very weak, making it extremely difficult to extract reservoir fluid information from seismic data and posing many challenges to the detection of reservoir oil and gas content. Summary of the Invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned existing technologies and provide a method for extracting reservoir seismic information for oil and gas detection to solve the problem of identifying or extracting the weak seismic response signals of pore fluids in deeply buried reservoirs, and further realize the reliable detection of reservoir oil and gas content.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] A method for extracting reservoir seismic information for oil and gas detection includes the following steps:

[0007] S1. Obtain the original data of reservoirs in the target area, including drilling, seismic, and geological data, etc., and combine domain knowledge and the seismic geological conditions of the target area to analyze, evaluate, and preprocess the relevant data.

[0008] S2. Perform high-fidelity and weak-signal-preserving processing on the seismic data guided by reservoir oil and gas content detection to obtain high-fidelity, high-signal-to-noise ratio, and high-resolution seismic data.

[0009] S3. Combine multivariate variational mode decomposition with high-order spectral analysis and cepstrum analysis methods to extract cepstrum features of seismic signals.

[0010] S4. Combine the unsupervised deep learning shared deep autoencoder with the spiking neural network to perform deep feature extraction on seismic signals;

[0011] S5. For seismic signals of different dimensions and types, respectively design feature extraction schemes suitable for the characteristics of their respective data types, and use feature visualization technology for feature visualization to analyze the responses of different seismic information to reservoir targets;

[0012] S6. Use the self-organizing mapping network to detect the oil and gas bearing property of the reservoir through multi-seismic feature information fusion, and use the logging interpretation and test results of the drilled wells in the target area to analyze and evaluate the detection results, so as to realize the reliable detection of the oil and gas bearing property of the reservoir.

[0013] Furthermore, the process of performing high-fidelity and weak-signal-preserving processing on seismic data in step S2 may include steps such as static correction processing combining tomography static correction and reflection wave residual static correction, wide-band high-fidelity and weak-signal noise suppression processing, pre-stack high-fidelity amplitude compensation, and high-resolution processing, etc., to better retain azimuth information, azimuth anisotropy characteristics, as well as low-frequency and high-frequency information.

[0014] Furthermore, the multivariate variational mode decomposition adopted in step S3 is an advanced multi-channel signal processing method, which can adaptively achieve the frequency-domain dissection of signals and the effective separation of each component, and at the same time make full use of the correlation between multi-channel data.

[0015] Furthermore, the cepstrum analysis adopted in step S3 is a classic method for extracting features of speech signals. It performs spectrum analysis on seismic signals and takes the logarithm of the spectrum to obtain cepstrum coefficients.

[0016] Furthermore, the high-order spectrum analysis adopted in step S3 is the cepstrum based on high-order cumulants, which has the advantages of high-order spectrum being able to suppress additive Gaussian noise and cepstrum deconvolution, and simplifies calculations, etc.

[0017] Furthermore, the shared deep autoencoder adopted in step S4 adds multiple independent layers on the basis of the deep autoencoder to ensure the reconstruction of seismic data and various attributes, and can fuse multiple seismic attributes while extracting features, realizing seismic feature extraction guided by seismic attribute knowledge.

[0018] Furthermore, the spiking neural network adopted in step S4 is a new generation of artificial neural network with more biological interpretability, having a unique information coding and processing method, rich spatio-temporal dynamics characteristics, and a low-power event-driven working mode, and has great advantages in processing time-series signals.

[0019] Further, the different-dimensional seismic signals described in step S5 may include data of one dimension, two dimensions, three dimensions, or even higher dimensions, and different types of seismic signals may include prestack and post-stack data, etc.

[0020] Further, the self-organizing mapping network adopted in step S6 is a typical unsupervised learning algorithm, which can unify and integrate the seismic information of reservoirs obtained by using different data and different methods for the same data, and realize the comprehensive detection of hydrocarbon-bearing properties of reservoirs.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] A method for extracting reservoir seismic information for oil and gas detection provided by the present invention combines domain knowledge and seismic geological conditions of the target area, and performs high-fidelity and weak-signal-preserving processing on seismic data guided by the detection of hydrocarbon-bearing properties of reservoirs, and can obtain high-quality imaging results; by organically combining multivariate variational mode decomposition, higher-order spectral analysis, cepstrum analysis, and deep learning of seismic data, it is possible to more comprehensively mine the rich reservoir pore fluid seismic response information containing fine structures of geological media and small property changes in seismic data; adopting a self-organizing mapping network and combining actual exploration data for multi-seismic feature information fusion can realize more comprehensive and accurate detection of deep and small hydrocarbon-bearing feature anomalies. Description of the Drawings

[0023] Figure 1 is a flowchart of the method of the present invention; Detailed Embodiments

[0024] The present invention will be described in detail below with reference to the drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and gives detailed implementation manners and specific operation processes, but the protection scope of the present invention is not limited to the following embodiments.

[0025] A method for extracting reservoir seismic information for oil and gas detection provided by the present invention has a method flow as Figure 1 shown, and the specific implementation steps are as follows:

[0026] Step S1: Obtain the original data of reservoirs in the target area, including drilling, seismic, and geological data, etc., and analyze, evaluate, and preprocess the relevant data in combination with domain knowledge and seismic geological conditions of the target area;

[0027] Step S2: Perform high-fidelity and weak-signal-preserving processing on the seismic data guided by the detection of hydrocarbon-bearing properties of reservoirs to obtain high-fidelity, high-signal-to-noise ratio, and high-resolution seismic data;

[0028] Specifically, the process of performing high-fidelity and weak-signal-preserving processing on seismic data in this step may include static correction processing combining tomography static correction and reflection wave residual static correction, wide-band high-fidelity and weak-signal noise suppression processing, pre-stack high-fidelity amplitude compensation, and high-resolution processing, etc., to better retain azimuth information, azimuth anisotropy characteristics, as well as low-frequency and high-frequency information.

[0029] Step S3: Combine multivariate variational mode decomposition with high-order spectrum analysis and cepstrum analysis methods to extract cepstrum features of seismic signals;

[0030] Specifically, the process of seismic feature extraction by combining multivariate variational mode decomposition and cepstrum analysis in this step is as follows. First, perform multivariate variational mode decomposition on the seismic signals processed by high-fidelity and weak-signal-preserving processing to obtain a series of intrinsic mode function components. Calculate time-domain and frequency-domain features for the first few components respectively according to reservoir characteristics, and form a one-dimensional vector representing the features of the high-frequency part of the seismic signals with the calculated feature values. Then divide each component into different principal and secondary component components, and perform cepstrum analysis to extract seismic information respectively. Finally, based on the seismic information extracted from the original seismic signals and the principal and secondary component components, and the calculated time-frequency domain features, use the correlation weighting method for fusion to obtain the cepstrum features characterizing reservoir information.

[0031] Specifically, the multivariate variational mode decomposition used in this step is an advanced multi-channel signal processing method, which can adaptively achieve frequency-domain dissection of signals and effective separation of each component, and at the same time make full use of the correlation between multi-channel seismic data.

[0032] Specifically, the cepstrum analysis used in this step is a classic method for extracting features of speech signals. It performs spectrum analysis on seismic signals and takes the logarithm of the spectrum to obtain cepstrum coefficients.

[0033] Specifically, the high-order spectrum analysis used in this step is the cepstrum based on high-order cumulants, which has the advantages of high-order spectrum being able to suppress additive Gaussian noise and cepstrum deconvolution, and simplifying calculations.

[0034] Step S4: Combine unsupervised deep learning shared deep autoencoder with pulsed neural network to extract deep features of seismic signals;

[0035] Specifically, the seismic feature extraction using the shared deep autoencoder in this step includes two parts: an encoder and a decoder. First, input the seismic data processed by high-fidelity and weak-signal-preserving processing into the encoder for encoding to obtain reservoir seismic features. Subsequently, the decoder restores the seismic features obtained by the encoder to output data, and design multiple independent layers before the output layer to avoid the series or parallel connection of different seismic attribute data, so that they can independently affect the extraction of deep features.

[0036] Specifically, in this step, the network model training realizes the update of the entire network structure by minimizing the error between the input data and the output data, making the output data as similar as possible to the input seismic data, and thus obtaining the seismic depth features representing reservoir information.

[0037] Specifically, the spiking neural network adopted in this step is a new generation of artificial neural network with more biological interpretability, having a unique information encoding and processing method, rich spatio-temporal dynamics characteristics, and a low-power event-driven working mode, and having great advantages in time series signal processing.

[0038] Step S5: For seismic signals of different dimensions and different types, respectively design feature extraction schemes suitable for the characteristics of their respective data types, and use feature visualization technology for feature visualization to analyze the response of different seismic information to reservoir targets.

[0039] Specifically, the different-dimensional seismic signals described in this step may include data of one dimension, two dimensions, three dimensions or even higher dimensions, and different types of seismic signals may include pre-stack and post-stack data, etc.

[0040] Step S6: On this basis, use a self-organizing mapping network for reservoir oil and gas bearing detection by fusing multi-seismic feature information, and use the logging interpretation and test results of the drilled wells in the target area to analyze and evaluate the detection results, so as to realize reliable detection of the reservoir oil and gas bearing property.

[0041] Specifically, the self-organizing mapping network described in this step is a typical unsupervised learning algorithm, which can fuse and unify the reservoir seismic information obtained by using different data and different methods of the same data, and realize comprehensive detection of the reservoir oil and gas bearing property.

[0042] Those skilled in the art should understand that the above embodiments are only preferred embodiments of the present invention, and are only used to exemplarily illustrate the beneficial effects of the present invention, and are not used to limit the present invention. Therefore, any simple modifications, equivalent changes and improvements made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention should not be excluded from the protection scope of the present invention.

Claims

1. A reservoir seismic information extraction method for oil and gas detection, characterized in that: The following steps are involved: (1) Obtain the original reservoir data of the target area, including drilling, seismic and geological data, and analyze, evaluate and pre-process the relevant data in combination with domain knowledge and seismic and geological conditions of the target area; (2) Perform high-fidelity weak signal processing on seismic data with the goal of detecting the oil and gas content of the reservoir to obtain high-fidelity, high signal-to-noise ratio, and high-resolution seismic data; (3) Combine multivariate variational mode decomposition with high-order spectrum analysis and cepstrum analysis to extract cepstrum features of seismic signals; (4) Combine unsupervised deep learning shared deep autoencoders with spiking neural networks to extract deep features from seismic signals; (5) For seismic signals of different dimensions and types, feature extraction schemes suitable for the characteristics of their respective data types are designed, and feature visualization technology is used to visualize the features and analyze the response of different seismic information to reservoir targets; (6) A self-organizing mapping network is used to detect the oil and gas content of the reservoir by fusing multiple seismic feature information. The detection results are analyzed and evaluated using the logging interpretation and test results of the wells drilled in the target area, thereby achieving reliable detection of the oil and gas content of the reservoir.

2. A reservoir seismic information extraction method for oil and gas detection according to claim 1, characterized in that: The process of performing high-fidelity weak signal processing on seismic data described in step (2) may include static correction processing combining tomographic static correction and reflection wave residual static correction, broadband high-fidelity weak signal noise suppression processing, pre-stack high-fidelity amplitude compensation and high-resolution processing, so as to better preserve azimuth information, azimuthal anisotropy characteristics, and low-frequency and high-frequency information.

3. The reservoir seismic information extraction method for oil and gas detection according to claim 1, characterized in that: The multivariate variational mode decomposition used in step (3) is an advanced multi-channel signal processing method that can adaptively realize the frequency domain decomposition of the signal and the effective separation of each component, while making full use of the correlation between multi-channel data; cepstrum analysis is a classic method for extracting speech signal features. It performs spectral analysis on the seismic signal and obtains the cepstrum coefficients by taking the logarithm of the spectrum; high-order spectrum analysis is based on the cepstrum of high-order cumulants, and has the advantages of high-order spectrum being able to suppress additive Gaussian noise and cepstrum deconvolution and simplify calculation.

4. The reservoir seismic information extraction method for oil and gas detection according to claim 1, characterized in that: The shared deep autoencoder used in step (4) adds multiple independent layers on the basis of the deep autoencoder to ensure the reconstruction of seismic data and multiple attributes. It can fuse multiple seismic attributes while extracting features, and realize seismic feature extraction guided by seismic attribute knowledge. The pulse neural network is a new generation of artificial neural network with more biological interpretability. It has a unique information encoding and processing method, rich spatiotemporal dynamic characteristics, and a low-power event-driven working mode. It has great advantages in time series signal processing.

5. The reservoir seismic information extraction method for oil and gas detection according to claim 1, characterized in that: The seismic signals of different dimensions described in step (5) may include one-dimensional, two-dimensional, three-dimensional or even higher-dimensional data, and the seismic signals of different types may include pre-stack and post-stack data, etc.

6. The reservoir seismic information extraction method for oil and gas detection according to claim 1, characterized in that: The fusion of multiple seismic characteristic information in step (6) refers to fusing and unifying reservoir seismic information obtained using different data or different methods for the same data to achieve comprehensive detection of reservoir oil and gas content.

Images