CN-122024077-A - Fine-grained sedimentary rock facies identification method based on deep learning

CN122024077ACN 122024077 ACN122024077 ACN 122024077ACN-122024077-A

Abstract

The invention discloses a fine-grained sedimentary rock facies recognition method based on deep learning, which relates to the technical field of deep learning and sedimentary rock facies recognition and comprises the steps of collecting lithofacies curve data and microstructure image data, screening characteristic parameters, and generating a standardized multi-mode characteristic set through a proprietary analysis engine; the method comprises the steps of calling a specific model to extract hierarchical features of a microstructure image to obtain a deep microstructure feature vector, inputting the two types of features into a dedicated network, outputting a preliminary identification result through cross-modal attention weight distribution, carrying out inversion correction and dynamic adjustment on the preliminary result by adopting a specific algorithm, and constructing a lithofacies classification system by combining feedback information to finish accurate division. According to the invention, through multi-source feature depth fusion, cross-modal association enhancement and multi-link collaborative optimization, efficient and accurate identification of fine-grained sedimentary rock facies is realized, complex lithology features are adapted, and support is provided for technical application in the related field.

Inventors

LU YUNFEI
TIAN JINGCHUN
Liang Qingshao
LIN XIAOBING

Assignees

成都理工大学

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (10)

1. The fine-grained sedimentary rock facies identification method based on deep learning is characterized by comprising the following steps of: s1, collecting lithofacies curve data and microstructure image data corresponding to a fine-grained sedimentary rock core sample, and screening the lithofacies curve multi-modal characteristic parameters and microstructure characteristic parameters; S2, constructing a lithofacies curve multi-modal intelligent analysis engine, importing the screened lithofacies curve multi-modal characteristic parameters, and generating a standardized multi-modal characteristic set through characteristic dimension mapping and modal information fusion; S3, calling a microstructure feature extraction CNN model, and carrying out feature hierarchical extraction on microstructure image data to obtain a deep microstructure feature vector comprising textures, pore distribution and mineral combination; S4, inputting the standardized multi-mode feature set and the deep microstructure feature vector into a cross-attention lithofacies recognition network, performing feature association enhancement through cross-mode attention weight distribution, and outputting a preliminary lithofacies recognition result; s5, carrying out inversion correction on the preliminary lithofacies recognition result by adopting a Gaussian constraint inversion change prediction algorithm, and optimizing the recognition result by dynamic adjustment of constraint conditions; And S6, based on the optimized recognition result, constructing a fine-grained sedimentary rock facies classification system by combining feedback information of a rock facies curve multi-mode intelligent analysis engine, and completing rock facies type division.
2. The deep learning based fine-grained sedimentary rock facies recognition method of claim 1, wherein the expression of the microstructure feature extraction CNN model is: Wherein, the Is a feature vector of a deep microstructure, In order to activate the function, The weight matrix is a convolution kernel of one to three layers of CNN models respectively, The bias terms of each layer are respectively given, conv1D, conv, D, conv D are respectively one-dimensional, two-dimensional and three-dimensional convolution operations, For a matrix of microstructure image data, Is microstructure characteristic parameter vector, pool is pooling operation, Is an element-wise product operation.
3. The deep learning based fine grain sedimentary rock facies recognition method of claim 1, wherein the cross-attention rock facies recognition network is expressed as: Wherein, the For the preliminary lithology recognition result probability distribution vectors, In order to query the matrix, In the form of a matrix of keys, In the form of a matrix of values, For the key vector dimension, For the full-connection layer weight matrix, For the full link layer bias term, cat is a feature stitching operation, For the normalized multi-modal feature set, softmax is the normalization function.
4. The depth learning based fine-grained sedimentary rock facies identification method of claim 1, wherein the gaussian constraint inversion variation prediction algorithm is expressed as: Wherein, the For the optimized lithofacies recognition result probability distribution vector, In order to constrain the intensity coefficient(s), For the purpose of the gradient operator, The standard deviation of the gaussian distribution is given, Is the mean value of the gaussian distribution, The matrix is adjusted for inversion.
5. The deep learning-based fine-grained sedimentary rock facies recognition method of claim 1, wherein the feature fusion expression of the facies curve multi-modal intelligent analysis engine is: Wherein, the In order to normalize the multi-modal feature set, Is the number of the rock phase curve modes, Is the first The weight coefficient of the seed mode, norm is the standardized operation, trans is the characteristic transformation function, Is the first Lithofacies curve data of the seed modality, Is a fusion weight matrix.
6. The deep learning-based fine-grained sedimentary rock facies recognition method of claim 1, wherein the comprehensive optimization expression of the deep learning-based fine-grained sedimentary rock facies recognition is: Wherein, the For the composite loss function, MSE is the mean square error function, Is a real label vector of the lithology, For the regularization coefficient(s), Is an L2 regularization function.
7. The deep learning based fine grain sedimentary rock facies identification method of claim 1, wherein said S3 includes the sub-steps of: S31, performing pixel-level feature preliminary extraction on microstructure image data, capturing gray value change information of different spatial positions in an image through a first layer convolution operation of a microstructure feature extraction CNN model, and generating a low-level texture feature map; s32, inputting the low-level texture feature map into a second convolution layer of the model, excavating structural features of the middle layer of the pore edges and the mineral grain boundaries in the image through multi-scale convolution kernel parallel operation, and outputting a middle layer feature matrix; S33, performing space dimension expansion analysis on the middle-layer feature matrix by using a third-layer three-dimensional convolution operation of the model, and fusing feature information in different depth directions to form a three-dimensional structure feature tensor; And S34, carrying out dimension compression and calibration feature screening on the three-dimensional structural feature tensor through pooling operation, removing redundant information, and generating deep microstructure feature vectors with uniform dimensions.
8. The deep learning based fine grain sedimentary rock facies identification method of claim 1, wherein said S4 includes the sub-steps of: S41, mapping the standardized multi-modal feature set and the deep microstructure feature vector into a query matrix, a key matrix and a value matrix respectively, and determining a basic data structure of cross-modal feature interaction; s42, performing similarity operation on the query matrix and the key matrix through a cross attention computing mechanism to obtain an attention weight matrix, and quantifying the association strength between different features; s43, carrying out weighted summation on the value matrix based on the attention weight matrix, highlighting the contribution of the calibration associated features, suppressing the irrelevant feature interference, and generating a feature vector fusing attention information; s44, inputting the fused feature vectors into a full-connection layer for nonlinear transformation, mapping through an activation function to obtain preliminary probability distribution corresponding to each lithofacies type, and outputting a preliminary lithofacies identification result.
9. The deep learning based fine grain sedimentary rock facies identification method of claim 1, wherein said S5 includes the sub-steps of: s51, determining the mean value and standard deviation parameters in a Gaussian constraint inversion change prediction algorithm based on a preliminary lithology recognition result and lithology priori knowledge, and constructing a Gaussian constraint distribution model; S52, calculating a deviation vector between the preliminary lithology recognition result and the Gaussian constraint distribution model, solving the change rate of the deviation vector through a gradient operator, and determining an inversion adjustment direction; S53, adjusting inversion amplitude according to the constraint intensity coefficient, and correcting the preliminary lithology recognition result element by element to generate an intermediate correction result; and S54, coupling the intermediate correction result with characteristic feedback information output by the lithology curve multi-mode intelligent analysis engine, dynamically adjusting inversion parameters, and outputting an optimized lithology recognition result.
10. The deep learning based fine grain sedimentary rock facies identification method of any one of claims 1-9, wherein the method is implemented by a different unit, comprising: The fine-grained sedimentary rock multisource data acquisition and screening unit is used for acquiring lithofacies curve data and microstructure image data corresponding to the rock core samples, screening effective lithofacies curve multi-modal characteristic parameters and microstructure characteristic parameters, and transmitting the screened data to the lithofacies curve multi-modal intelligent analysis unit; The lithology curve multi-modal intelligent analysis unit receives the data transmitted by the multi-source data acquisition and screening unit, generates a standardized multi-modal feature set through feature dimension mapping, modal information fusion and standardization processing, and transmits the feature set to the microstructure feature extraction unit and the lithology recognition result optimization unit respectively; The microstructure feature extraction unit is used for calling a microstructure feature extraction CNN model to perform hierarchical convolution and pooling operation on microstructure image data, generating a deep microstructure feature vector and transmitting the deep microstructure feature vector to the cross attention lithofacies recognition unit; The cross-modal attention lithology recognition unit is used for receiving the standardized multi-modal feature set and the deep microstructure feature vector, outputting a preliminary lithology recognition result through cross-modal attention weight distribution and nonlinear transformation, and transmitting the preliminary lithology recognition result to the lithology recognition result optimization unit; the lithofacies recognition result optimizing unit performs inversion correction and dynamic adjustment on the preliminary lithofacies recognition result by adopting a Gaussian constraint inversion change prediction algorithm, generates an optimized recognition result and transmits the optimized recognition result to the lithofacies classification system constructing unit; And the lithofacies classification system construction unit is used for receiving the optimized recognition result and the feedback information of the lithofacies curve multi-mode intelligent analysis unit, constructing a fine-grained sedimentary lithofacies classification system, finishing the precise classification of the lithofacies types and outputting the final recognition result.

Description

Fine-grained sedimentary rock facies identification method based on deep learning Technical Field The invention relates to the technical field of deep learning and sedimentary rock facies identification, in particular to a fine sedimentary rock facies identification method based on deep learning. Background The fine-grained sedimentary rock facies identification is a core technical link in the fields of sedimentary geological analysis and resource exploration, and the accuracy of the fine-grained sedimentary rock facies identification directly influences stratum evaluation and resource development efficiency. Along with the extension of exploration targets to deep and complex stratum, fine-grained sedimentary rock presents characteristics such as various mineral compositions, complex microstructures, blurred lithofacies boundaries and the like, and the traditional identification mode is difficult to comprehensively capture effective information in multi-source data, so that the high-precision identification requirement cannot be met. In the situation, the recognition method integrating the multi-mode data and the intelligent analysis technology becomes a development trend, and a recognition flow of a system is established by deeply mining the multi-dimensional characteristics related to the lithofacies, so that a technical support is provided for solving the difficult problem of complex fine-grained sedimentary lithofacies recognition. The method has the key defects that firstly, the sufficiency of multi-source feature fusion is insufficient, the related features and microstructure features of lithofacies curves cannot be effectively related, only single type data are simply processed, so that feature expression is incomplete, lithofacies essential attributes are difficult to reflect, secondly, an optimization mechanism of an identification result lacks systematicness, a complete process of dynamic adjustment and feedback correction is lacking, deviation generated in the identification process cannot be effectively corrected, a technical path of multi-loop collaborative optimization is not formed, the identification result is easily influenced by factors such as data noise, feature redundancy and the like, and complexity and variability of fine-grained sedimentary lithofacies are difficult to adapt. Disclosure of Invention In order to overcome the defects and shortcomings in the prior art, the invention provides a fine-grained sedimentary rock facies identification method based on deep learning. The technical scheme adopted by the invention is that the fine-grain sedimentary rock facies recognition method based on deep learning is characterized by comprising the following steps: s1, collecting lithofacies curve data and microstructure image data corresponding to a fine-grained sedimentary rock core sample, and screening the lithofacies curve multi-modal characteristic parameters and microstructure characteristic parameters; S2, constructing a lithofacies curve multi-modal intelligent analysis engine, importing the screened lithofacies curve multi-modal characteristic parameters, and generating a standardized multi-modal characteristic set through characteristic dimension mapping and modal information fusion; S3, calling a microstructure feature extraction CNN model, and carrying out feature hierarchical extraction on microstructure image data to obtain a deep microstructure feature vector comprising textures, pore distribution and mineral combination; S4, inputting the standardized multi-mode feature set and the deep microstructure feature vector into a cross-attention lithofacies recognition network, performing feature association enhancement through cross-mode attention weight distribution, and outputting a preliminary lithofacies recognition result; s5, carrying out inversion correction on the preliminary lithofacies recognition result by adopting a Gaussian constraint inversion change prediction algorithm, and optimizing the recognition result by dynamic adjustment of constraint conditions; And S6, based on the optimized recognition result, constructing a fine-grained sedimentary rock facies classification system by combining feedback information of a rock facies curve multi-mode intelligent analysis engine, and completing rock facies type division. Further, the expression of the microstructure feature extraction CNN model is: Wherein, the Is a feature vector of a deep microstructure,In order to activate the function,The weight matrix is a convolution kernel of one to three layers of CNN models respectively,The bias terms of each layer are respectively given, conv1D, conv, D, conv D are respectively one-dimensional, two-dimensional and three-dimensional convolution operations,For a matrix of microstructure image data,Is microstructure characteristic parameter vector, pool is pooling operation,Is an element-wise product operation. Further, the expression of the cross-attention lithology recognition netw