A Machine Learning-Based Evaluation Method for Marine-Continental Transitional Shale Gas

Abstract

This invention relates to a machine learning-based method for evaluating marine-continental transitional shale gas. The method includes acquiring well logging data, seismic migration data, and incident angle gathers for pre-stack inversion; applying cross-plot analysis to the well logging data to obtain sensitive attributes and derive petrophysical parameters of the marine-continental transitional shale; applying a pre-stack simultaneous inversion method to the seismic migration data to obtain P-wave impedance, S-wave impedance, and P-wave / S-wave velocity ratios, and calculating elastic parameters such as TOC, porosity, geostress, and brittleness index of the reservoir based on the petrophysical parameters; obtaining instantaneous frequency, coherence volume, and ant-body attributes through mathematical operations based on the seismic data; applying a greedy algorithm to the obtained elastic parameters to calculate regression coefficients after iteratively discarding feature parameters to obtain preferred attributes; and creating labeled data based on the attribute parameters and shale gas geological favorable area classification standards, and inputting the preferred attributes into a U-shaped neural network constructed using an attention mechanism and bottleneck residual structure.

Description

Technical Field

[0001] This invention relates to the field of shale gas evaluation technology, specifically a machine learning-based method for evaluating marine-continental transitional shale gas. Background Technology

[0002] my country's transitional marine-continental shale features vertically superimposed lithologies, significant variations in shale thickness, diverse organic matter composition, and well-developed porosity and microfractures, yet commercial exploitation has not yet been achieved. Conventional shale gas evaluation methods typically rely on evaluation parameters for southern marine shale gas, including gas content characteristics, total organic carbon (TOC) content, geostress, and brittleness index. These parameters are generally classified into three favorable zones using single-factor analysis and multi-factor composite methods. However, each evaluation parameter reflects structural, lithological, and fluid characteristics. Parameter selection based on interpreters' geological experience, or the use of mathematical statistical methods to calculate correlation coefficients between attributes to screen sensitive attributes, and the simple application of linear regression fitting, neglects redundant information and complex nonlinear characteristics between attributes.

[0003] Currently, the focus of shale gas evaluation methods is on fitting the relationship between well logging curves and single evaluation parameters. For example, Nie Haikuan et al. (2012) used multiple regression analysis to establish a model of organic carbon content, porosity, and shale gas content; Zhou Yexin et al. (2021) used regression equations to quantitatively calculate gas content; and Wang Meng et al. (2021) established a quantitative relationship between TOC and density, and used the density volume after pre-stack inversion to calculate TOC. Another problem in evaluating favorable shale gas areas is how to achieve attribute optimization and multi-attribute fusion. In practical applications, fuzzy mathematics or analytic hierarchy process (AHP) is mostly used to evaluate favorable shale gas areas. The solutions are complex, and the determination of weights is highly subjective, making it difficult to objectively and accurately predict favorable shale gas areas. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention provides a machine learning-based evaluation method for shale gas in the marine-continental transitional facies. By using machine learning, the method achieves intelligent fusion of multiple attributes of shale gas, solves the problems of weight allocation and empirical assignment of single attributes such as shale gas content, and better integrates multiple attributes such as shale gas content, TOC, porosity, formation pressure and brittleness index.

[0005] To achieve the above objectives, the specific process for shale gas evaluation according to the present invention includes the following steps:

[0006] Step 1: Acquire well logging data, seismic migration data, and incident angle gathers for pre-stack inversion;

[0007] Step 2: Apply cross-plot analysis to the well logging data to obtain the petrological properties of the marine-continental transitional shale.

[0008] Step 3: Apply the pre-stack simultaneous inversion method to the seismic migration data to obtain the P-wave impedance, S-wave impedance and P-wave / S-wave velocity ratio, and calculate the TOC, porosity and brittleness index of the reservoir based on the rock physics quanta.

[0009] Step 4: Calculate the instantaneous frequency, coherence volume, and ant body attributes based on the seismic migration data;

[0010] Step 5: Apply a greedy algorithm to the data obtained in Step 3 and Step 4 respectively to calculate the regression coefficients after iteratively discarding feature parameters, and obtain the attribute parameters. The attribute parameters include P-wave impedance, S-wave impedance, P-wave to S-wave velocity ratio, TOC, porosity, Poisson's ratio, Young's modulus, instantaneous frequency, instantaneous phase, instantaneous amplitude, energy attribute, coherence volume, and ant body attribute data.

[0011] Step 6: Based on the attribute parameters and the classification criteria for favorable shale gas geological areas, create labeled data. Input the filtered attributes into a U-shaped neural network built with an attention mechanism and bottleneck residual structure. Use a nonlinear regression algorithm based on the U-shaped neural network to predict favorable shale gas zones.

[0012] As a preferred embodiment of the present invention, the instantaneous frequency in step four is:

[0013]

[0014] Where θ(t) is the instantaneous phase and ω(t) is the instantaneous frequency; the covariance matrix is ​​established, and the calculated coherent volume is:

[0015]

[0016] C 11 C 22 C 33 C represents the autocorrelation of the first, second, and third records. 12 C represents the cross-correlation between the first and second data sets. 13 The cross-correlation between the first and third channels is used; the maximum coherence value in each time window is taken as the final coherence value.

[0017] C1 = maxC1

[0018] As a preferred embodiment of the present invention, the working process of the U-shaped neural network in step six is ​​as follows:

[0019]

[0020]

[0021]

[0022]

[0023] Where F is the input feature, M c It is a channel attention one-dimensional convolution, M s It is a spatial attention 2D convolution, F avg It is average pooling, F max It's max pooling, σ is the sigmoid function, W1 and W0 are the hidden layer weight parameters, and f... 7×7 It's a convolution operation. It is a matrix multiplication by element, where F′ and F” represent the optimized channel attention and spatial attention feature maps, respectively.

[0024] The beneficial effects of this invention are as follows: The machine learning-based evaluation method proposed in this invention can automatically learn the characteristic weights of shale gas, thereby achieving the goal of accurately predicting favorable shale gas zones and improving the accuracy of characterizing favorable shale gas reservoirs. This allows for accurate prediction of the vertical organic-rich shale development intervals and the horizontal comprehensive sweet spot distribution characteristics of shale. This invention has important application value in studying the distribution patterns and exploration and mining of marine-continental transitional shale. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the workflow of the present invention.

[0026] Figure 2 This is the incident angle gather of the present invention;

[0027] Figure 3 The ratio of P-wave to S-wave velocity after inversion according to this invention;

[0028] Figure 4 The ant body properties, amplitude properties, and energy properties of this invention;

[0029] Figure 5 This refers to the U-shaped neural network structure involved in this invention. Detailed Implementation

[0030] Example 1

[0031] like Figure 1 As shown, a machine learning-based method for evaluating marine-continental transitional shale gas includes the following steps:

[0032] Step 1: Acquire well logging data, seismic migration data, and incident angle gathers for pre-stack inversion, where the incident angle gathers are as follows: Figure 2 As shown;

[0033] Step 2: Apply cross-plot analysis to the well logging data to obtain the petrological properties of the marine-continental transitional shale.

[0034] Step 3: Apply the pre-stack simultaneous inversion method to the seismic migration data to obtain P-wave impedance, S-wave impedance, and P-wave / S-wave velocity ratio. Calculate the reservoir's TOC, porosity, in-situ stress, and brittleness index, etc., based on rock physics metrics. The P-wave / S-wave velocity ratio image is shown below. Figure 3 As shown;

[0035] Step 4: Based on the seismic migration data, perform mathematical calculations to obtain the instantaneous frequency, coherence volume, and ant-body attributes, such as... Figure 4 As shown; the instantaneous frequency is:

[0036]

[0037] Where θ(t) is the instantaneous phase and ω(t) is the instantaneous frequency; the covariance matrix is ​​established, and the calculated coherent volume is:

[0038]

[0039] C 11 C 22 C 33 C represents the autocorrelation of the first, second, and third records. 12 C represents the cross-correlation between the first and second data sets. 13 The cross-correlation between the first and third channels is used; the maximum coherence value in each time window is taken as the final coherence value.

[0040] C1 = maxC1

[0041] Step 5: Apply a greedy algorithm to the obtained elastic parameters to calculate the regression coefficients after iteratively discarding the feature parameters, and obtain the attribute parameters. The attribute parameters include longitudinal wave impedance, transverse wave impedance, longitudinal wave to transverse wave velocity ratio, TOC, porosity, Poisson's ratio, Young's modulus, instantaneous frequency, instantaneous phase, instantaneous amplitude, energy attribute, coherence volume, and ant body attribute data.

[0042] Step Six: Based on the attribute parameters and the classification criteria for favorable shale gas geological areas, create labeled data. Input the filtered attributes into a U-shaped neural network constructed using an attention mechanism and bottleneck residual structure. Then, use a nonlinear regression algorithm based on the U-shaped neural network to predict favorable shale gas zones. The attention mechanism and bottleneck residual are specifically as follows:

[0043]

[0044]

[0045]

[0046]

[0047] Where F is the input feature, M c It is a channel attention one-dimensional convolution, M s It is a spatial attention 2D convolution, F avg It is average pooling, F max It's max pooling, σ is the sigmoid function, W1 and W0 are the hidden layer weight parameters, and f... 7×7 It's a convolution operation. It is a matrix multiplication by element, where F′ and F” represent the optimized channel attention and spatial attention feature maps, respectively.

[0048] The parts not described in detail in this article are existing technologies.

[0049] While the specific embodiments of the present invention have been described in detail above, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention, and modifications or variations without creative effort are still within the protection scope of the present invention.

Claims

1. A machine learning-based method for evaluating marine-continental transitional shale gas, characterized in that, The steps include the following: Step 1: Acquire well logging data, seismic migration data, and incident angle gathers for pre-stack inversion; Step 2: Apply cross-plot analysis to the well logging data to obtain the petrological properties of the marine-continental transitional shale. Step 3: Apply the pre-stack simultaneous inversion method to the seismic migration data to obtain the P-wave impedance, S-wave impedance and P-wave / S-wave velocity ratio, and calculate the TOC, porosity and brittleness index of the reservoir based on the rock physics quanta. Step 4: Calculate the instantaneous frequency, coherence volume, and ant body attributes based on the seismic migration data; Step 5: Apply a greedy algorithm to the data obtained in Step 3 and Step 4 respectively to calculate the regression coefficients after iteratively discarding feature parameters, and obtain the attribute parameters. The attribute parameters include P-wave impedance, S-wave impedance, P-wave to S-wave velocity ratio, TOC, porosity, Poisson's ratio, Young's modulus, instantaneous frequency, instantaneous phase, instantaneous amplitude, energy attribute, coherence volume, and ant body attribute data. Step 6: Based on the attribute parameters and the classification criteria for favorable shale gas geological areas, create labeled data. Input the filtered attributes into a U-shaped neural network built with an attention mechanism and bottleneck residual structure. Use a nonlinear regression algorithm based on the U-shaped neural network to predict favorable shale gas zones.

2. The machine learning-based evaluation method for marine-continental transitional shale gas according to claim 1, characterized in that: The instantaneous frequency in step four is: ， in, For instantaneous phase, Given the instantaneous frequency; establish the covariance matrix, and calculate the coherence volume as follows: ， C 11 C 22 C 33 C represents the autocorrelation of the first, second, and third records. 12 C represents the cross-correlation between the first and second data sets. 13 The cross-correlation between the first and third channels is used; the maximum coherence value in each time window is taken as the final coherence value. 。 3. The machine learning-based evaluation method for marine-continental transitional shale gas according to claim 1, characterized in that: The working process of the U-shaped neural network described in step six is ​​as follows: ， ， ， ， in, These are input features. It is a one-dimensional convolution with channel attention. It is a spatial attention 2D convolution. It is average pooling. It is max pooling. It is the sigmoid function. and These are the hidden layer weight parameters. It's a convolution operation. It is matrix element-wise multiplication. and These represent the optimized channel attention and spatial attention feature maps, respectively.

Images