CN-121978759-A - Intelligent transverse wave speed prediction method and device based on channel attention mechanism

CN121978759ACN 121978759 ACN121978759 ACN 121978759ACN-121978759-A

Abstract

The invention discloses a method and a device for intelligently predicting transverse wave speed based on a channel attention mechanism, wherein the method comprises the steps of preprocessing a conventional logging curve and analyzing transverse wave speed sensitivity, and optimizing a characteristic curve sensitive to transverse wave speed; the method comprises the steps of carrying out standardization processing on a characteristic curve, constructing a training set with space structure information based on the characteristic curve after the standardization processing, constructing a transverse wave speed prediction model based on a one-dimensional deep convolutional neural network and a channel attention mechanism, utilizing the training set training model to establish a high-precision nonlinear mapping relation between the characteristic curve and the transverse wave speed, carrying out quality control evaluation on the trained transverse wave speed prediction model by adopting a test set, and carrying out transverse wave speed prediction on a well lacking the actually measured transverse wave speed by utilizing the trained transverse wave speed prediction model to obtain a corresponding transverse wave speed curve. The invention can directly utilize the conventional logging curve to realize high-efficiency and accurate transverse wave speed prediction.

Inventors

YANG LIUXIN
LV HUI
YIN LONG
MA YONGQIANG
LI YANG
DING ZHAO

Assignees

中国石油化工股份有限公司
中石化石油物探技术研究院有限公司

Dates

Publication Date: 20260505
Application Date: 20241030

Claims (10)

1. The intelligent transverse wave speed prediction method based on the channel attention mechanism is characterized by comprising the following steps of: Preprocessing a conventional logging curve and analyzing the sensitivity of the shear wave speed, wherein the logging curve sensitive to the shear wave speed is preferably used as a characteristic curve; Carrying out standardization processing on the characteristic curve; constructing a training set with spatial structure information based on the characteristic curve after the standardization processing; Constructing a transverse wave speed prediction model based on a one-dimensional depth convolution neural network and a channel attention mechanism, and training the transverse wave speed prediction model by utilizing the training set so as to establish a high-precision nonlinear mapping relation between a characteristic curve and the transverse wave speed; performing quality control evaluation on the trained transverse wave speed prediction model by adopting a test set until the prediction accuracy meets the requirement; and predicting the transverse wave speed of the well lacking the actually measured transverse wave speed by using the trained transverse wave speed prediction model, so as to obtain a corresponding transverse wave speed curve.
2. The method of claim 1, wherein preprocessing and shear wave velocity sensitivity analysis of conventional log curves comprises: Collecting well data having measured shear wave velocities; performing quality control analysis on the logging curve, and preprocessing the problem curve, including editing and correcting; Sensitivity analysis is performed on conventional log curves by using intersection analysis and pearson correlation coefficients, and log curves sensitive to transverse wave velocity are preferred as characteristic curves.
3. The method of claim 1, wherein normalizing the characteristic curve comprises: The characteristic curve is standardized by adopting a Z-score standardization method, and a standardized calculation formula is as follows: Wherein x is the data to be normalized, x mean is the mean, x std is the variance, and x norm is the normalization And (5) the subsequent data.
4. The method of claim 1, wherein the dimension of the input data of each sample in the training set is m×n, where M is a step length, i.e., the number of the sample points and the adjacent sample points in the longitudinal direction thereof, N is a number of characteristic curves, and the label data corresponding to a sample is a measured transverse wave velocity of the sample.
5. The method of claim 4, wherein the setting mode of the tag data corresponding to the sample includes: the top layer mode is that the tag data is the transverse wave speed corresponding to the first sampling point in the input sampling points of the current sample; the bottom layer mode is that the label data is the transverse wave speed corresponding to the last sampling point in the input sampling points of the current sample; In the middle mode, the tag data is the transverse wave speed corresponding to the middle sampling point in the input sampling points of the current sample.
6. The method of claim 1, wherein the shear wave velocity prediction model comprises a plurality of attention mechanism modules and two fully connected layers connected in sequence, wherein each attention mechanism module comprises a convolution layer, a batch normalization layer, relu function nonlinear transformation, a fully connected layer and Softmax function nonlinear transformation, and the number of convolution kernels of the plurality of attention mechanism modules is gradually increased.
7. The method of claim 6, wherein the attention mechanism module is configured to extract the characteristic data related to the shear wave velocity from the training set during the training process, and wherein the overall characteristic extraction process comprises convolution operations, batch normalization, nonlinear mapping, channel attention weight calculation, and characteristic weight assignment, and wherein the process is represented as follows: z (l) ＝Conv1(W i (l) ,a (l-1) )+b i (2) p (l) ＝BN(z (l) ) (3) q (l) ＝f(p (l) ) (4) r (l) ＝f(W j (l) ,q (l) )+b j (5) a (l) ＝M(q (l) ,s (l) ) (7) Wherein a (l-1) is the output of the previous layer, a (l) is the output of the current layer, conv1 represents the execution of one-dimensional convolution operation, the edge filling is carried out on input data during the operation, W i (l) is the weight matrix of the ith convolution kernel of the current layer, b i is the ith bias term, BN is a batch normalization function, p (l) is the result after batch normalization, f is an activation function for nonlinear transformation, reLU is adopted as the activation function, W j (l) is the weight matrix of the 1 st fully connected layer, and b j is the jth neuron bias term; the method comprises the steps of obtaining a weight matrix of a 2 nd full-connection layer, obtaining a k-th neuron bias item by b k , carrying out nonlinear transformation by g as an activation function, learning weight coefficients for different features by adopting Softmax as the activation function, obtaining a result of a current attention mechanism module after the corresponding weight coefficient is given to each feature by M as a matrix multiplication operation.
8. The method of claim 7, wherein the error between the model predictions and the real labels is calculated during training using a mean square error function, the objective function being: Wherein E is an objective function, y i ' is a predicted transverse wave speed, and y i is an actual transverse wave speed; And the residual error value calculated by the objective function is reversely propagated by adopting a gradient descent algorithm so as to update network parameters, thereby realizing the training of the model.
9. An intelligent transverse wave speed prediction device based on a channel attention mechanism is characterized by comprising: the preprocessing and sensitivity analysis module is used for preprocessing a conventional logging curve and analyzing the sensitivity of the transverse wave speed, and the logging curve sensitive to the transverse wave speed is preferably used as a characteristic curve; The standardized processing module is used for carrying out standardized processing on the characteristic curve; the training set construction module is used for constructing a training set with space structure information based on the characteristic curve after the standardization processing; The prediction model construction module is used for constructing a transverse wave speed prediction model based on a one-dimensional depth convolutional neural network and a channel attention mechanism, and training the transverse wave speed prediction model by utilizing the training set so as to establish a high-precision nonlinear mapping relation between a characteristic curve and the transverse wave speed; The test module is used for carrying out quality control evaluation on the trained transverse wave speed prediction model by adopting a test set until the prediction accuracy meets the requirement; and the prediction module is used for predicting the transverse wave speed of the well lacking the actually measured transverse wave speed by using the trained transverse wave speed prediction model, so as to obtain a corresponding transverse wave speed curve.
10. An electronic device, the electronic device comprising: at least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the channel attention mechanism based shear wave velocity intelligent prediction method of any of claims 1-8.

Description

Intelligent transverse wave speed prediction method and device based on channel attention mechanism Technical Field The invention relates to the technical field of oil and gas geophysics, in particular to a transverse wave speed intelligent prediction method and device based on a channel attention mechanism. Background Transverse wave velocity is one of the key parameters in the aspects of pre-stack inversion, reservoir prediction, fluid identification, etc. The measured shear wave velocity of the subsurface formation may typically be obtained by coring laboratory analysis or dipole shear logging. However, most wells lack measured shear wave data due to high cost and other factors. In practical application, a petrophysical modeling method is generally adopted to predict the transverse wave speed. The method uses a conventional logging curve to describe the characteristics of the pore structure, mineral components, elastic modulus and the like of the rock from a microscopic angle under a certain assumption condition, and further calculates and obtains the transverse wave velocity information of the rock. However, in practical application, the prediction accuracy and efficiency of the transverse wave speed still face a great challenge. Because of the complexity of the deposition environment, diagenetic effect and post-transformation, the pore structure, fluid characteristics and the influence of the pore structure, the fluid characteristics and the influence on the elastic modulus of a reservoir are difficult to accurately describe by a petrophysical model based on certain assumption conditions, and the accurate reservoir transverse wave speed is difficult to obtain by adopting a conventional petrophysical model. In addition, the method for predicting the transverse wave speed by using the petrophysical modeling method involves a very complex calculation process, a plurality of parameters and low prediction efficiency, and the petrophysical model has obvious difference on the adaptability of different reservoirs. In recent years, artificial intelligence techniques typified by deep learning have been rapidly developed, and remarkable progress has been made in the fields of image recognition, speech recognition, large models, and the like. The deep learning technology is used for completing various tasks such as classification segmentation, regression fitting and the like by constructing a deep neural network model, is also introduced into the field of geophysical exploration, and has good application prospect in the aspect of transverse wave speed prediction. When the deep learning technology is adopted to predict the transverse wave speed, the actually measured transverse wave speed is required to be collected as the label data to train the deep neural network model, but as the actually measured transverse wave speed data which can be used as the label data is generally less, and the reservoir layer is always provided with space structural characteristics such as transverse sedimentary phase change, longitudinal lithology interbedding, thin layer or multi-period superposition and the like, the problems of inaccurate precision, low model generalization, low noise resistance and the like still exist when the conventional deep neural network is adopted to predict the transverse wave speed. Disclosure of Invention The invention aims to provide a channel attention mechanism-based intelligent transverse wave speed prediction method and device, which can directly utilize a conventional logging curve to realize efficient and accurate transverse wave speed prediction, and aims to solve the problems that in the prior art, a rock physical modeling method is complex in calculation process, multiple in parameters, and insufficient in transverse wave speed prediction precision and efficiency, and a conventional deep learning model is difficult to consider the influence of stratum space structure characteristics on rock transverse wave speed prediction, and model generalization is not high under the condition of a small sample. In order to achieve the above object, in a first aspect, the present invention provides a method for intelligently predicting a transverse wave velocity based on a channel attention mechanism, including: Preprocessing a conventional logging curve and analyzing the sensitivity of the shear wave speed, wherein the logging curve sensitive to the shear wave speed is preferably used as a characteristic curve; Carrying out standardization processing on the characteristic curve; constructing a training set with spatial structure information based on the characteristic curve after the standardization processing; Constructing a transverse wave speed prediction model based on a one-dimensional depth convolution neural network and a channel attention mechanism, and training the transverse wave speed prediction model by utilizing the training set so as to establish a high-precision nonlinear mapping relation bet