CN-122014239-A - Shale gas content prediction method based on logging curve

CN122014239ACN 122014239 ACN122014239 ACN 122014239ACN-122014239-A

Abstract

The invention relates to the technical field of logging, and discloses a shale gas content prediction method based on a logging curve, which comprises the steps of 1, preprocessing logging data, 2, carrying out self-adaptive multi-scale decomposition on the logging curve, 3, screening IMF components, 4, constructing a low-frequency background component and a medium-high frequency modulation component, 5, screening features, 6, constructing a background gas content target, 7, constructing a gas content prediction model, 8, training and restraining the model, and solving the problems that in the prior art, long-scale information reflecting regional geological background in a logging signal and medium-high frequency information depicting local structures of thin layers and layers are difficult to effectively separate, feature aliasing and physical significance are easy to cause, geological scale constraint and longitudinal continuity are easy to ignore, non-physical fluctuation is easy to generate, and the generalization capability of the model is limited.

Inventors

LI YUEXIN
YANG RUI
ZHANG WENJIE
CHENG SIJUN
XU JIANTING
CHEN KEFEI

Assignees

中国地质大学（武汉）

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (6)

1. A shale gas content prediction method based on a log, the method comprising: S1, acquiring at least one conventional logging curve of a target well and measured gas content data of a corresponding well section, and performing depth registration to perform pretreatment on the conventional logging curve; s2, sequentially decomposing each preprocessed conventional logging curve along the depth direction by adopting a complete set empirical mode decomposition method to obtain a plurality of intrinsic mode function IMF components; s3, calculating the energy ratio of the IMF components of the eigenmode function, reserving effective IMF components with energy larger than a preset threshold value, and removing noise components with the lowest energy ratio; S4, superposing at least two effective IMF components with the lowest frequency in the effective IMF components and residual errors to construct a low-frequency background component reflecting the overall change trend of the stratum, and respectively carrying out linear superposition and absolute value superposition treatment on the rest effective IMF components to obtain a medium-high frequency modulation component used for representing the local physical property change characteristics and the layer development intensity in the reservoir; s5, respectively calculating correlation coefficients of the acquired low-frequency background component and the acquired medium-high frequency modulation component, and eliminating characteristic components with high similarity; S6, selecting a corresponding background window thickness according to the interval scale based on the actually measured gas content curve, and carrying out sliding window average treatment on the actually measured gas content along the depth direction to obtain a background gas content target curve representing long-scale change characteristics, wherein the background gas content target curve is used as a supervision target and geological constraint of a model background branch; S7, constructing a dual-branch gas content prediction model based on the preprocessed background characteristic input and the medium-high frequency modulation characteristic input, wherein the dual-branch gas content prediction model comprises a background branch and a modulation branch, the background branch is used for establishing a mapping relation between logging background characteristics and gas content background components and outputting a background gas content predicted value which is not negatively restrained, the modulation branch takes the medium-high frequency modulation characteristic as input and introduces the background gas content predicted value as a conditional restraint for representing a gas content residual error predicted value caused by local change of a reservoir and growth of a tattooing layer; and S8, performing supervision training on the dual-branch gas content prediction model based on measured gas content data, introducing main prediction loss, background gas content constraint loss and residual supervision loss in the training process, and constructing a joint loss function through the multi-objective joint optimization and constraint training to obtain a stable dual-branch gas content prediction model with geological significance.
2. The method of logging-based shale gas content prediction as claimed in claim 1, wherein said step of pre-treating said conventional log comprises: Uniformly carrying out quantile truncation processing on the conventional logging curve to remove abnormal values, and adopting a wavelet denoising method to inhibit high-frequency noise interference so as to improve the curve signal-to-noise ratio; Setting sampling intervals, and carrying out resampling and interpolation processing on all the conventional logging curves along the depth direction so that the curves are uniformly mapped to the same depth sampling grid.
3. The method for predicting shale gas content based on a well logging curve according to claim 1, wherein the step of sequentially decomposing each pretreated conventional well logging curve along the depth direction by adopting a complete set empirical mode decomposition method to obtain a plurality of intrinsic mode function IMF components comprises the following steps: Applying a plurality of groups of Gaussian white noise with controlled amplitude to the preprocessed conventional log signals to form a noise auxiliary signal set: (1) In the formula, For noise-assisted signals, x (z) is the conventional log signal, z is depth, White noise sequence added for the kth time; performing empirical mode decomposition on each noise auxiliary signal, extracting a first-order eigenmode function of the noise auxiliary signal, and performing collective average on all test results to obtain a first-order CEEMDAN eigenmode component The expression is: (2) On the basis, the first-order eigenmode component is removed from the original conventional log curve signal, and a residual signal is obtained: (3) Wherein, the For the residual signal, then for the residual signal Repeating the noise auxiliary decomposition and the set averaging process, and extracting the ith-order eigenmode component step by step: (4) And updates the residual signal: (5) finally, the conventional log signal is expressed as the sum of a plurality of eigenvalues and residual terms, namely: (6)。
4. the method for predicting shale gas content based on a logging curve according to claim 1, wherein the method comprises the following steps: the formula for calculating the energy duty ratio of the IMF component of the eigenmode function is as follows: (7) Wherein, the Representing the energy of the i-th order eigenmode component for the sum of squares of the i-th order eigenmode component along the depth direction; the sum of squares of the corresponding log along the depth direction is indicative of the total energy of the log, and the IMF component is considered as an ineffective noise component when the ratio is less than a preset threshold.
5. The shale gas content prediction method based on the logging curve according to claim 1, wherein the pre-processed background feature input and medium-high frequency modulation feature input are used for constructing a dual-branch gas content prediction model, the dual-branch gas content prediction model comprises a background branch and a modulation branch, the background branch is used for establishing a mapping relation between logging background features and gas content background components and outputting a background gas content prediction value which is not negative constraint, the modulation branch takes the medium-high frequency modulation features as input and introduces the background gas content prediction value as a conditional constraint and is used for representing a gas content residual prediction value caused by local change of a reservoir and layer development, and the background gas content prediction value and the residual prediction value are overlapped to obtain a gas content prediction result of a target well section, so that the cooperative modeling of the gas content background trend and the local change features is realized, and the method comprises the steps of: the background branch in the dual-branch gas content prediction model has a low-frequency background component For input, calculating a background air content predicted value through a feedforward neural network, wherein the calculation process is expressed as follows: (8) Wherein, the For training the resulting weights and bias terms, Representing a non-linear activation function, Is a non-negative constraint function; at the same time, the modulation branch modulates the characteristic vector at a medium-high frequency For input, extracting local variation features through a nonlinear function: (9) Will be With a corresponding depth Splicing to form an extended feature vector, and further performing nonlinear mapping to obtain a residual prediction value: (10) finally, the air content predicted value is obtained by superposing a background air content predicted value and the residual error predicted value, and the overall predicted model can be expressed as follows: (11)。
6. The shale gas content prediction method based on a logging curve according to claim 1, wherein the step of performing supervised training on the dual-branch gas content prediction model based on measured gas content data, introducing main prediction loss, background gas content constraint loss and residual error supervision loss simultaneously in the training process, constructing a joint loss function through the multi-objective joint optimization and constraint training, and obtaining a stable and geologically significant dual-branch gas content prediction model comprises the following steps: the construction form of the joint loss function is as follows: (12) Wherein, the The mean square error between the predicted value and the measured value of the air content is used for restraining the overall prediction precision; For the background to be close to the long-scale gas content constraint, the background branch is made to approach to the smooth measured gas content, For the context constraint weight coefficient, To measure the resulting value of the air content smoothed along the background window, Can be expressed as: (13) the average value constraint is used for limiting the overall average value of the tattooing correction value to be close to 0, so that the average level and the long-scale change of the total air content are mainly borne by background branches, the tattooing branches are only subjected to local positive and negative small-amplitude correction near the background, The weight coefficient is constrained for the layers of the grain, Can be expressed as: (14)。

Description

Shale gas content prediction method based on logging curve Technical Field The invention relates to the technical field of logging, in particular to a shale gas content prediction method based on a logging curve. Background The gas content is a key parameter in unconventional oil and gas reservoir evaluation and favorable region optimization, and the accurate prediction of the gas content has important significance for reservoir fine characterization and development scheme formulation. The existing gas content prediction method mainly depends on core experimental analysis or a statistical regression and machine learning model based on a conventional logging curve, but is limited by factors such as scarcity of core data, strong non-stationarity of logging signals, remarkable reservoir heterogeneity and the like, and prediction precision and stability are difficult to consider. The traditional well logging curve processing method mostly adopts fixed scale filtering or experience feature extraction means, and is difficult to effectively separate long scale information reflecting regional geological background from medium-high frequency information depicting local structures such as thin layers, tattoos and the like in well logging signals, so that feature aliasing and unclear physical significance are easily caused. In recent years, although partial data driving models achieve a certain effect in gas content prediction, geological scale constraint and longitudinal continuity are often ignored, non-physical fluctuation is easy to generate, and the generalization capability of the models is limited. Therefore, there is a need for an air content prediction method that can fully mine multi-scale information of a logging curve, combine geological background and local heterogeneous characteristics, and have good stability and interpretability, so as to meet the actual requirements of fine evaluation of a complex reservoir. Disclosure of Invention The invention aims to provide a shale gas content prediction method based on a logging curve, which aims to solve the problems that in the prior art, long-scale information reflecting regional geological background and medium-high frequency information depicting local structures of thin layers and layers are difficult to effectively separate in logging signals, characteristic aliasing and unclear physical significance are easy to cause, geological scale constraint and longitudinal continuity are easy to ignore, non-physical fluctuation is easy to generate, and model generalization capability is limited. In order to achieve the above object, the present invention provides the following method: the shale gas content prediction method based on the logging curve provided by the invention comprises the following steps: S1, acquiring at least one conventional logging curve of a target well and measured gas content data of a corresponding well section, and performing depth registration to perform pretreatment on the conventional logging curve; s2, sequentially decomposing each preprocessed conventional logging curve along the depth direction by adopting a complete set empirical mode decomposition method to obtain a plurality of intrinsic mode function IMF components; s3, calculating the energy ratio of the IMF components of the eigenmode function, reserving effective IMF components with energy larger than a preset threshold value, and removing noise components with the lowest energy ratio; S4, superposing at least two effective IMF components with the lowest frequency in the effective IMF components and residual errors to construct a low-frequency background component reflecting the overall change trend of the stratum, and respectively carrying out linear superposition and absolute value superposition treatment on the rest effective IMF components to obtain a medium-high frequency modulation component used for representing the local physical property change characteristics and the layer development intensity in the reservoir; s5, respectively calculating correlation coefficients of the acquired low-frequency background component and the acquired medium-high frequency modulation component, and eliminating characteristic components with high similarity; S6, selecting a corresponding background window thickness according to the interval scale based on the actually measured gas content curve, and carrying out sliding window average treatment on the actually measured gas content along the depth direction to obtain a background gas content target curve representing long-scale change characteristics, wherein the background gas content target curve is used as a supervision target and geological constraint of a model background branch; S7, constructing a dual-branch gas content prediction model based on the preprocessed background characteristic input and the medium-high frequency modulation characteristic input, wherein the dual-branch gas content prediction model comprises a background b