Well logging lithology identification method, device and equipment and storage medium

Abstract

The invention discloses a well logging lithology identification method, device and equipment and a storage medium, and relates to the technical field of geological exploration, and the method comprises the steps: obtaining well logging parameters related to lithology changes to obtain initial well logging curves, and determining a target well logging curve from the initial well logging curves based on a target window length; performing fast Fourier transform on the target logging curve to obtain a frequency domain feature vector corresponding to the target logging curve, and performing depth feature extraction on the target logging curve by using a deep learning encoder to obtain a depth attention feature vector corresponding to the target logging curve; and performing feature fusion on the frequency domain feature vector and the depth attention feature vector to obtain a target feature, and performing classification identification on the target feature by using a preset lithology category classifier to obtain a lithology identification result. Therefore, lithology prediction can be refined, and a reliable basis is provided for geological interpretation, reservoir evaluation and oil and gas development.

Description

Technical Field

[0001] This invention relates to the field of geological exploration technology, and in particular to a well logging lithology identification method, apparatus, equipment, and storage medium. Background Technology

[0002] With the development of digital oil and gas fields and intelligent drilling technology, intelligent lithology identification based on well logging data has become crucial for improving exploration efficiency. Traditional methods rely on human experience and statistical models, which are highly subjective, inefficient, and ill-suited for complex formations such as thin interbedded layers and highly heterogeneous formations.

[0003] In recent years, deep learning-based identification methods have significantly improved automation levels, but they still face the following key bottlenecks under complex geological conditions: First, the mechanism for utilizing neighborhood information is rigid. Fixed-window CNNs (Convolutional Neural Networks) or RNNs (Recurrent Neural Networks) cannot adapt to the spatial context requirements of rock layers of different thicknesses (such as thin laminae and thick facies), easily introducing noise or losing sedimentary cycle features, reducing the accuracy of thin interbedded layers and interfaces. Second, the feature representation dimension is singular. Existing methods mostly rely on the original curves in the depth domain, failing to effectively integrate frequency domain and time-frequency features, weakening the ability to mine information such as sedimentary cycles and lithological "fingerprints," and lacking robustness to wellbore noise. Third, geological consistency is insufficient. Model outputs often exhibit problems such as discontinuous profiles and blurred boundaries, lacking macroscopic adherence to overall geological laws, affecting the reliability of reservoir evaluation. Fourth, generalization ability is limited. The performance of the model degrades significantly when migrating across work areas and sedimentary environments, lacking adaptive calibration capabilities, which restricts industrial applications.

[0004] Therefore, there is an urgent need to develop an intelligent lithology identification method that can adapt to neighborhood information, integrate multi-dimensional features, ensure geological consistency, and has strong generalization, so as to promote the intelligent and refined interpretation of well logging. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a well logging lithology identification method, apparatus, device, and storage medium, capable of achieving high-precision lithology identification in complex formation environments. The specific solution is as follows: In a first aspect, this application discloses a well logging lithology identification method, including: Obtain logging parameters related to lithological changes to obtain initial logging curves, and determine target logging curves from the initial logging curves based on the target window length; A fast Fourier transform is performed on the target logging curve to obtain the frequency domain feature vector corresponding to the target logging curve, and a deep learning encoder is used to extract the depth features of the target logging curve to obtain the depth attention feature vector corresponding to the target logging curve. The frequency domain feature vector and the deep attention feature vector are fused to obtain the target feature, and the target feature is classified and identified using a preset lithology classifier to obtain the lithology identification result.

[0006] Optionally, the step of acquiring logging parameters related to lithological changes to obtain an initial logging curve, and determining a target logging curve from the initial logging curve based on the target window length, includes: Several logging parameters related to lithological changes are acquired, and initial multidimensional logging curves are obtained using a preset data cleaning method, data standardization method, and feature optimization processing. The logging parameters include natural gamma, density, neutron porosity, sonic transit time, deep lateral resistivity, and shallow lateral resistivity logging data. The target logging curve is determined from the initial multidimensional logging curves based on the target window length.

[0007] Optionally, performing a fast Fourier transform on the target logging curve to obtain the frequency domain feature vector corresponding to the target logging curve includes: Perform a fast Fourier transform on the target logging curve to obtain the amplitude spectrum corresponding to the target logging curve; A learnable frequency weight matrix is ​​introduced, and the amplitude spectrum is processed using the learnable frequency weight matrix and a dot product operation to obtain the frequency domain feature vector corresponding to the target logging curve; the learnable frequency weight matrix is ​​determined based on the frequency distribution corresponding to the logging parameters.

[0008] Optionally, the step of extracting depth features from the target logging curve using a deep learning encoder to obtain a depth attention feature vector corresponding to the target logging curve includes: The target logging curve is transposed and linearly mapped to obtain a target sequence, and the target sequence is input into a deep learning encoder for processing to obtain initial output features; An adaptive depth weight matrix is ​​introduced, and the initial output features are weighted using the adaptive depth weight matrix to obtain the depth attention feature vector corresponding to the target logging curve; the adaptive depth weight matrix is ​​set based on the lithology category.

[0009] Optionally, the step of performing a Fast Fourier Transform on the target logging curve to obtain the frequency domain feature vector corresponding to the target logging curve, and using a deep learning encoder to extract depth features from the target logging curve to obtain the depth attention feature vector corresponding to the target logging curve, includes: Perform a fast Fourier transform on the target logging curve to obtain the weighted spectrum corresponding to the target logging curve; The weighted spectrum is nonlinearly mapped and feature compression is performed using a multilayer perceptron to obtain the frequency domain feature vector corresponding to the target logging curve. A deep learning encoder is used to extract depth features from the target logging curve to obtain the initial depth features of the target logging curve. The initial depth features are compressed using the nonlinear transformation layer in the deep learning encoder to obtain the depth attention feature vector corresponding to the target logging curve.

[0010] Optionally, the step of fusing the frequency domain feature vector and the deep attention feature vector to obtain the target features includes: The weight values ​​corresponding to the frequency domain feature vector and the deep attention feature vector are determined by using linear mapping and gating units, and the weight values ​​are used to fuse the frequency domain feature vector and the deep attention feature vector to obtain the target feature.

[0011] Optionally, the step of classifying and identifying the target features using a preset lithology classifier to obtain lithology identification results includes: The target features are compressed to obtain compressed features, and a classifier is built based on a multilayer perceptron to obtain a preset lithology classifier. The compressed features are classified and identified using a preset lithology classifier to obtain lithology identification results.

[0012] Secondly, this application discloses a well logging lithology identification device, comprising: The well logging curve acquisition module is used to acquire well logging parameters related to lithological changes to obtain an initial well logging curve, and to determine a target well logging curve from the initial well logging curve based on the target window length; The vector determination module is used to perform a fast Fourier transform on the target logging curve to obtain the frequency domain feature vector corresponding to the target logging curve, and to use a deep learning encoder to extract the depth features of the target logging curve to obtain the depth attention feature vector corresponding to the target logging curve. The feature recognition module is used to perform feature fusion on the frequency domain feature vector and the deep attention feature vector to obtain target features, and to classify and recognize the target features using a preset lithology classifier to obtain lithology recognition results.

[0013] Thirdly, this application discloses an electronic device, including: Memory, used to store computer programs; A processor is used to execute the computer program to implement the aforementioned well logging lithology identification method.

[0014] Fourthly, this application discloses a computer-readable storage medium for storing a computer program, which, when executed by a processor, implements the aforementioned well logging lithology identification method.

[0015] As can be seen, in this application, logging parameters related to lithological changes are obtained to obtain an initial logging curve, and a target logging curve is determined from the initial logging curve based on the target window length. A fast Fourier transform is performed on the target logging curve to obtain the corresponding frequency domain feature vector, and a deep learning encoder is used to extract depth features from the target logging curve to obtain the corresponding depth attention feature vector. Feature fusion is performed on the frequency domain feature vector and the depth attention feature vector to obtain the target feature, and a preset lithology classifier is used to classify and identify the target feature to obtain the lithology identification result. In this way, by fully utilizing the frequency domain-attention dual-branch fusion structure to obtain the target feature, the multi-scale feature information of the logging data can be fully utilized to achieve high-precision lithology identification in complex formation environments. Furthermore, this method still exhibits good robustness when dealing with imbalanced datasets, effectively identifying minority lithologies and avoiding model prediction bias towards the mainstream category. When faced with complex geological conditions and dramatic changes in lithology, the model can still maintain high predictive consistency and geological rationality, enabling large-scale and refined lithological prediction within the reservoir area. This provides a reliable basis for geological interpretation, reservoir evaluation, and oil and gas development, thereby effectively reducing exploration and development risks. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1 This is a flowchart of a well logging lithology identification method disclosed in this application; Figure 2 This is a flowchart of a specific well logging lithology identification method disclosed in this application; Figure 3 This is a visual comparison chart of lithology identification based on the Xushen gas field volcanic reservoir dataset disclosed in this application; Figure 4 This is a visual comparison chart of lithology identification based on the Sulige gas field carbonate reservoir dataset disclosed in this application; Figure 5 This is a schematic diagram of the structure of a well logging lithology identification device disclosed in this application; Figure 6 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] With the development of digital oil and gas fields and intelligent drilling technologies, intelligent lithology identification using well logging data has become a key technology for improving exploration and development efficiency. Lithology identification is a core component of well logging interpretation and reservoir evaluation, and its accuracy directly affects the effectiveness of oil and gas reservoir exploration and development. Therefore, this application will specifically introduce a well logging lithology identification method that can accurately identify lithology.

[0020] See Figure 1 As shown in the figure, this application discloses a well logging lithology identification method, including: Step S11: Obtain logging parameters related to lithological changes to obtain initial logging curves, and determine target logging curves from the initial logging curves based on the target window length.

[0021] In this embodiment, the step of acquiring logging parameters related to lithological changes to obtain an initial logging curve, and determining a target logging curve from the initial logging curve based on a target window length, includes: acquiring several logging parameters related to lithological changes, and using a preset data cleaning method, data standardization method, and feature optimization processing to obtain an initial multidimensional logging curve; the logging parameters include natural gamma, density, neutron porosity, sonic transit time, deep lateral resistivity, and shallow lateral resistivity logging data; and determining the target logging curve from the initial multidimensional logging curve based on the target window length.

[0022] In well logging prediction, data selection and collection are crucial for ensuring model accuracy and reliability. Appropriate selection of well logging parameters not only determines the representativeness and physical meaning of the model's input features but also directly affects the model's generalization ability and the rationality of its geological interpretation. Therefore, the processing scheme in this application collects high-quality well logging data to avoid the influence of equipment performance differences or human operational factors, ensuring that the collected data accurately reflects formation characteristics and reducing noise and measurement errors. High-quality data preprocessing provides a stable and reliable input foundation for model training and inference, which is a prerequisite for achieving accurate predictions.

[0023] This application selects six logging parameters closely related to lithological variations as model inputs, including natural gamma ray (GR), density (DEN), neutron porosity (CNL), acoustic transit time (AC), deep lateral resistivity (LLD), and shallow lateral resistivity (LLS) logging data. Raw logging data is typically affected by instrument noise, sampling errors, well section differences, and human operation. Directly using this data for model training can easily lead to prediction bias and feature distortion. This application utilizes systematic data cleaning, standardization, and feature optimization processing to effectively improve data stability and information density, providing a reliable foundation for model establishment. Feature optimization processing includes, but is not limited to, feature scaling and categorical feature encoding. Due to differences in formation lithology and logging physics, logging parameters in different reservoirs exhibit significant differences in numerical range and response patterns. Furthermore, directly inputting raw data into the model can easily lead to some features dominating the training process due to their large numerical dimensions. To mitigate the impact of these discrepancies on model predictions, this study employs a standardization method to process the well logging data, as shown in the following formula, to achieve scale uniformity and distribution balance of the input data: ; in, This indicates the values ​​of the well logging data. This is the mean value of the logging curve. This represents the standard deviation of the curve.

[0024] Step S12: Perform a fast Fourier transform on the target logging curve to obtain the frequency domain feature vector corresponding to the target logging curve, and use a deep learning encoder to extract the depth features of the target logging curve to obtain the depth attention feature vector corresponding to the target logging curve.

[0025] In this embodiment, after obtaining the target logging curve, the frequency domain features and depth neighborhood attention features of the target logging curve are used, and the attention weight of each point is dynamically adjusted according to the characteristics of different lithologies to construct a dual-branch fusion structure for classifying different lithology types.

[0026] That is, the step of performing a Fast Fourier Transform on the target logging curve to obtain the frequency domain feature vector corresponding to the target logging curve, and using a deep learning encoder to extract depth features from the target logging curve to obtain the depth attention feature vector corresponding to the target logging curve, includes: performing a Fast Fourier Transform on the target logging curve to obtain the weighted spectrum corresponding to the target logging curve; performing nonlinear mapping on the weighted spectrum and using a multilayer perceptron for feature compression to obtain the frequency domain feature vector corresponding to the target logging curve; using a deep learning encoder to extract depth features from the target logging curve to obtain the initial depth features of the target logging curve; and using a nonlinear transform layer in the deep learning encoder to compress the initial depth features to obtain the depth attention feature vector corresponding to the target logging curve.

[0027] The step of performing a Fast Fourier Transform (FFT) on the target logging curve to obtain the frequency domain feature vector corresponding to the target logging curve includes: performing a FFT on the target logging curve to obtain the amplitude spectrum corresponding to the target logging curve; introducing a learnable frequency weight matrix, and processing the amplitude spectrum using the learnable frequency weight matrix and a dot product operation to obtain the frequency domain feature vector corresponding to the target logging curve; the learnable frequency weight matrix is ​​determined based on the frequency distribution corresponding to the logging parameters.

[0028] The step of extracting depth features from the target logging curve using a deep learning encoder to obtain a depth attention feature vector corresponding to the target logging curve includes: transposing and linearly mapping the target logging curve to obtain a target sequence, and inputting the target sequence into the deep learning encoder for processing to obtain initial output features; introducing an adaptive depth weight matrix, and using the adaptive depth weight matrix to weight the initial output features to obtain a depth attention feature vector corresponding to the target logging curve; the adaptive depth weight matrix is ​​set based on lithology category.

[0029] Specifically, such as Figure 2 As shown, in well logging data, the curve morphology corresponding to different lithologies exhibits significant differences in frequency distribution. For example, thick mudstone layers show low-frequency smooth characteristics, while thin sandstone or interlayers show high-frequency oscillating characteristics. Therefore, this invention addresses this by processing the input multidimensional well logging curves... C represents the number of logging parameters. Given the depth window length, perform a Fast Fourier Transform (FFT) to obtain its amplitude spectrum in the frequency domain: ; The window length L determines the number of frequency bands in the frequency domain FFT. and frequency resolution , The sampling interval, L, is used to determine the appropriate sliding window size for different lithological layers. A larger L results in higher frequency resolution and a more detailed spectrum, suitable for low-frequency sedimentary cycles and thick lithologies. Conversely, a smaller L results in more frequency bands, making it more suitable for high-frequency abrupt changes in strata and for identifying rapidly changing characteristics. In practical applications, a default window size can be used as a baseline, and L can be gradually increased to observe the changes in recognition performance and computational cost, thereby selecting the optimal sliding window size that matches the research objective. Subsequently, a learnable frequency weight matrix is ​​introduced. Adaptive learning of the importance of different frequency components is achieved through dot product operations: .

[0030] The weighted spectrum is then nonlinearly mapped and compressed using a multilayer perceptron (MLP) to obtain the frequency domain feature vector. This process enables the automatic extraction of the periodic variation characteristics of strata and can identify the response patterns of different lithologies in the frequency domain.

[0031] The depth feature extraction module utilizes Transformer to model the complex dependencies between neighborhood points in the depth sequence and dynamically adjusts the weights of different depth locations through a lithology-adaptive attention mechanism. This accurately captures the periodic and spatial variation characteristics of strata, improving the accuracy and robustness of lithology identification. This module includes: a linear mapping embedding module, an improved Transformer submodule, a lithology-adaptive depth-weighted submodule, and a spatial feature compression submodule. For the input neighborhood matrix... Similar to the frequency domain window, the depth domain window improves the Transformer model to capture the complex dependencies between neighborhood feature depth points. It is implemented by transposing the depth dimension... The sequence embedding is obtained through linear mapping: ; Where D is the feature dimension. Then, it is input into a multi-layer Transformer encoder, where each layer contains a multi-head self-attention mechanism and a feedforward network structure, capable of capturing complex dependencies between neighboring depth points, such as bedding variations, phase boundary transitions, and sedimentary sequence features. The Transformer computation process uses a depth of... The encoders are stacked, with each layer containing both a self-attention network and a feedforward network, and layer normalization is performed before the residual connections: ; ; Where, Q,K , V These are query, key, and value matrices, respectively. After multi-head parallel processing, they are concatenated and projected back. The dimension is then connected to the residual via a feedforward network to obtain the updated dimension. Furthermore, this invention introduces a lithology-adaptive depth weight matrix based on the Transformer output features. ,in This represents the number of lithology categories. After Softmax normalization, this matrix dynamically assigns depth-oriented weights to different lithology categories, thereby achieving adaptive weighting of neighborhood information during feature aggregation. ; Finally, the depth characteristics weighted by various lithologies are averaged and compressed to a nonlinear transformation layer. Dimension, to obtain deep attention features This branch enables differentiated utilization of depth neighborhood information by different lithologies, maintaining identification accuracy and stability in thick, interbedded, and abrupt layer environments.

[0032] In summary, this application performs a Fast Fourier Transform on the logging curves in the frequency domain branch and introduces a learnable frequency weight matrix to automatically learn the importance of different frequency components in lithology identification, thereby effectively extracting the periodicity and texture features of the formation. In the attention branch, a Transformer network is used to model the multidimensional logging curves in the deep neighborhood, and a multi-head attention mechanism is used to capture the complex relationships between neighboring points, combined with a lithology adaptive weight matrix to form a dynamic context representation. Then, the two branches are processed by a multilayer perceptron for feature compression, and information is integrated in the fusion layer, ultimately achieving complementary fusion of frequency information and spatial dependency information to obtain high-precision lithology classification results.

[0033] Step S13: Perform feature fusion on the frequency domain feature vector and the deep attention feature vector to obtain target features, and use a preset lithology classifier to classify and identify the target features to obtain lithology identification results.

[0034] In this embodiment, the feature fusion of the frequency domain feature vector and the deep attention feature vector to obtain the target feature includes: determining the weight values ​​corresponding to the frequency domain feature vector and the deep attention feature vector using a linear mapping and gating unit, and using the weight values ​​to fuse the frequency domain feature vector and the deep attention feature vector to obtain the target feature. The classification and identification of the target feature using a preset lithology classifier to obtain a lithology identification result includes: compressing the target feature to obtain compressed features, constructing a classifier based on a multilayer perceptron to obtain a preset lithology classifier; and using the preset lithology classifier to classify and identify the compressed features to obtain the lithology identification result.

[0035] Specifically, in order to comprehensively utilize the complementary information of frequency domain and depth features, this application fuses the feature vectors output from the two branches: .

[0036] Then, the weights are obtained adaptively through linear mapping and gating units. and utilize weights The dual-domain features are weighted and fused to obtain the final fused feature Z: ; ; in, For the Sigmoid function, This represents element-wise product. The fused features are further compressed through a linear layer before being input into the classifier module. The classifier uses a multilayer perceptron (MLP) structure, and after ReLU activation, it outputs the probability distribution of lithology categories: .

[0037] The final classification results can include various lithological types such as mudstone, sandstone, siltstone, limestone, and interbedded sand and mudstone.

[0038] Furthermore, to verify the feasibility of the method, this application selects the volcanic rock reservoir of the Xushen gas field in the Xujiaweizi structure of the Songliao Basin and the carbonate rock reservoir of the Sulige gas field in the Ordos Basin as research objects. Both of these structural blocks have complex geological structures, which can effectively verify the feasibility of the model. Furthermore, this example selected seven mainstream comparison models—CNN, GRU (Gated Recurrent Unit), LSTM (Long Short-Term Memory), TCN (Temporal Convolutional Network), ViT (Vision Transformer), N-BEATs (Neural Basis Extension Analysis), and MLP-Mixer (Multilayer Perceptron Mixer)—and conducted a systematic comparative experiment with the method presented in this example. The predictive performance of each model was comprehensively evaluated based on multiple dimensions, including classification accuracy for siltstone, conglomerate, rhyolite, basalt, and andesite, overall accuracy (OA(%)), average accuracy (AA(%)), and Kappa coefficient (Kappa(×100)). The results are as follows: A significant breakthrough was achieved in the overall accuracy of lithological identification of volcanic reservoirs in the Xushen gas field of the Xujiaweizi structure in the Liao Basin (Table 1). Firstly, the well-logging lithological identification method (ANW-DFNet model) proposed in this invention achieved an overall accuracy (OA) of 96.50%, approximately 6 percentage points higher than the best-performing MLP-Mixer algorithm (90.48%). More importantly, the Kappa score reached 94.31, far exceeding all comparative models. Kappa eliminates the influence of random consistency, and this absolute lead fully demonstrates the higher reliability and robustness of the classification results of this invention; its superior performance is not accidental. The comparison results are shown in Table 1 below: Table 1 Comparison of lithology identification results based on the Xushen Gas Field volcanic reservoir dataset

[0039] Visualized comparison results of lithology identification based on the Xushen gas field volcanic reservoir dataset are as follows: Figure 3 As shown.

[0040] Secondly, it demonstrates stronger ability to distinguish complex lithologies, solving the problem of missed and misidentified identification of specific lithologies: In terms of identification accuracy across different categories, ANW-DFNet exhibits superior targeted identification capabilities. For conglomerate and basalt, other models generally have lower identification accuracy; for example, the classification accuracy for rhyolite is below 80% for most models, and for andesite, some models like ViT have an accuracy as low as 41.24%. This indicates that these lithologies are easily confused with other lithologies in terms of logging response characteristics, making them a challenge in volcanic rock identification. In contrast, the model of this invention achieves identification accuracies of 97.62% and 95.88% for conglomerate and basalt, respectively, accurately capturing the characteristics of difficult samples and effectively solving the identification bottleneck of traditional models for these lithologies. Although its accuracy in andesite identification is not the highest, this invention maintains extremely high accuracy in other categories, far exceeding other algorithms. Furthermore, the classification accuracy for basalt in this lithology identification remains at an acceptable level, demonstrating the balanced discrimination capability of the ANW-DFNet model across different lithologies. In addition, the AA value of this invention reaches 91.56%, significantly higher than other models. This demonstrates that ANW-DFNet effectively overcomes model bias that may result from data class imbalance, and does not sacrifice the recognition performance of a few classes in pursuit of high overall accuracy, thus possessing stronger generalization ability and practical application value.

[0041] To further verify the generalization ability of the proposed ANW-DFNet model in different types of complex reservoirs, we conducted comparative experiments in the JT1 carbonate block of the Sulige gas field in the Ordos Basin. Carbonate reservoirs exhibit strong heterogeneity and complex pore, fracture, and vault systems, thus placing extremely high demands on the robustness of lithology identification algorithms. The experimental results are shown in Table 2 below. It was found that the method in this application achieves superior overall performance compared to other algorithms.

[0042] Table 2 Comparison of lithology identification results based on the Sulige Gas Field carbonate reservoir dataset

[0043] A visual comparison of lithology identification based on the Sulige gas field carbonate reservoir dataset is shown below. Figure 4As shown in the figure, firstly, several advanced models (such as GRU, ViT, and N-BEATS) have demonstrated high recognition levels, with OA exceeding 92.5%. In this competitive environment with high benchmarks, the ANW-DFNet model proposed in this invention goes a step further, improving OA to 94.10% and achieving the highest Kappa coefficient (92.49). This proves that the method of this invention not only has advantages in absolute accuracy but also produces the most consistent and reliable classification results, and its robustness is fully demonstrated in complex carbonate rock environments. Secondly, ANW-DFNet exhibits comprehensive and balanced lithological identification capabilities, with an AA value of 92.93%, the highest among all models. Most importantly, this invention achieves the highest identification accuracy for dolomite and argillaceous dolomite, at 95.03% and 97.07% respectively, demonstrating its precise ability to capture specific carbonate rock microfacies. For other categories, the identification accuracy of this invention is at a leading or first-tier level, with no category showing a sharp drop in identification accuracy. This ensures that the model can provide consistently high-performance output when facing complex and varied carbonate rock profiles in practical applications, avoiding overall interpretation errors caused by the failure to identify a certain type of lithology. Furthermore, the core challenge of carbonate reservoirs lies in their strong multi-scale heterogeneity, with extremely large variations in characteristic scales, from microscopic intergranular pores to macroscopic vugs and fractures. The core mechanism of this invention, ANW-DFNet, is key to addressing this challenge: the adaptive neighborhood sensing mechanism dynamically adjusts its sensing field. When dealing with homogeneous and dense carbonate rock sections, the model expands its sensing range to acquire background trends; when encountering thin interlayers or fractured sections, the model can quickly focus on local abrupt changes. This capability ensures that the model can grasp the overall characteristics of thick lithology while accurately locating thin layers and lithological interfaces, thus achieving high accuracy across all categories. The frequency sensing mechanism reveals sedimentary and diagenetic information of carbonate rocks from another dimension. The cyclical deposition of carbonate rocks and the pore system created by later dissolution leave unique imprints in the frequency domain of logging signals. This mechanism, by mining these frequency domain features, provides a crucial basis for distinguishing carbonate rocks with similar mineral compositions but different structures (such as dense limestone and porous limestone).

[0044] In summary, the ANW-DFNet algorithm exhibits comprehensive superiority over other algorithms, stemming from its core adaptive neighborhood perception and frequency perception mechanisms. The adaptive neighborhood perception mechanism, through dynamic adjustment, accurately captures multi-scale spatial features of volcanic and carbonate rocks, ranging from microscopic mineral fabrication to macroscopic eruption cycles. This allows the model to effectively focus on key discriminative features when dealing with structurally complex lithologies, avoiding insufficient information utilization or overfitting issues caused by fixed-scale convolution kernels or loop steps. The frequency perception mechanism deeply mines the frequency domain "fingerprint" information contained in well logging signals, related to the rhythm and pore structure of volcanic rocks. This information is easily overlooked in traditional time-domain models, but provides crucial supplementary discriminative criteria for distinguishing between volcanic and carbonate rocks with similar mineral compositions but different structures.

[0045] As can be seen, in this application, logging parameters related to lithological changes are obtained to obtain an initial logging curve, and a target logging curve is determined from the initial logging curve based on the target window length. A fast Fourier transform is performed on the target logging curve to obtain the corresponding frequency domain feature vector, and a deep learning encoder is used to extract depth features from the target logging curve to obtain the corresponding depth attention feature vector. Feature fusion is performed on the frequency domain feature vector and the depth attention feature vector to obtain the target feature, and a preset lithology classifier is used to classify and identify the target feature to obtain the lithology identification result. In this way, by fully utilizing the frequency domain-attention dual-branch fusion structure to obtain the target feature, the multi-scale feature information of the logging data can be fully utilized to achieve high-precision lithology identification in complex formation environments. Furthermore, this method still exhibits good robustness when dealing with imbalanced datasets, effectively identifying minority lithologies and avoiding model prediction bias towards the mainstream category. When faced with complex geological conditions and dramatic changes in lithology, the model can still maintain high predictive consistency and geological rationality, enabling large-scale and refined lithological prediction within the reservoir area. This provides a reliable basis for geological interpretation, reservoir evaluation, and oil and gas development, thereby effectively reducing exploration and development risks.

[0046] refer to Figure 5 The present application also discloses a well logging lithology identification device, comprising: The logging curve acquisition module 11 is used to acquire logging parameters related to lithological changes to obtain an initial logging curve, and to determine a target logging curve from the initial logging curve based on the target window length; The vector determination module 12 is used to perform a fast Fourier transform on the target logging curve to obtain the frequency domain feature vector corresponding to the target logging curve, and to use a deep learning encoder to extract the depth features of the target logging curve to obtain the depth attention feature vector corresponding to the target logging curve. The feature recognition module 13 is used to perform feature fusion on the frequency domain feature vector and the deep attention feature vector to obtain target features, and to classify and recognize the target features using a preset lithology classifier to obtain lithology recognition results.

[0047] As can be seen, in this embodiment, by fully utilizing the frequency domain-attention dual-branch fusion structure to obtain target features, it is possible to fully leverage the multi-scale feature information of well logging data to achieve high-precision lithology identification in complex formation environments. Furthermore, this method maintains good robustness when dealing with datasets with imbalanced sample distributions, effectively identifying minority lithology classes and preventing model predictions from biasing towards the mainstream category. Even when facing geological conditions with complex structures and drastic lithological variations, the model can still maintain high predictive consistency and geological rationality, achieving large-scale, refined lithology prediction within the reservoir area. This provides a reliable basis for geological interpretation, reservoir evaluation, and oil and gas development, thereby effectively reducing exploration and development risks.

[0048] In some specific embodiments, the well logging curve acquisition module 11 may specifically include: The data acquisition unit is used to acquire several logging parameters related to lithological changes, and to obtain initial multidimensional logging curves by using preset data cleaning methods, data standardization methods, and feature optimization processing; the logging parameters include natural gamma, density, neutron porosity, sonic transit time, deep lateral resistivity, and shallow lateral resistivity logging data. The curve generation unit is used to determine the target logging curve from the initial multidimensional logging curves based on the target window length.

[0049] In some specific embodiments, the vector determination module 12 may specifically include: An amplitude spectrum determination unit is used to perform a fast Fourier transform on the target logging curve to obtain the amplitude spectrum corresponding to the target logging curve. The frequency distribution determination unit is used to introduce a learnable frequency weight matrix and process the amplitude spectrum using the learnable frequency weight matrix and a dot product operation to obtain the frequency domain feature vector corresponding to the target logging curve; the learnable frequency weight matrix is ​​determined based on the frequency distribution corresponding to the logging parameters.

[0050] In some specific embodiments, the vector determination module 12 may specifically include: An initial feature determination unit is used to transpose and linearly map the target logging curve to obtain a target sequence, and input the target sequence into a deep learning encoder for processing to obtain initial output features; A depth feature determination unit is used to introduce an adaptive depth weight matrix and use the adaptive depth weight matrix to weight the initial output features to obtain the depth attention feature vector corresponding to the target logging curve; the adaptive depth weight matrix is ​​set based on the lithology category.

[0051] In some specific embodiments, the vector determination module 12 may specifically include: The weighted spectrum determination unit is used to perform a fast Fourier transform on the target logging curve to obtain the weighted spectrum corresponding to the target logging curve. A frequency domain feature vector determination unit is used to perform nonlinear mapping on the weighted spectrum and use a multilayer perceptron for feature compression to obtain the frequency domain feature vector corresponding to the target logging curve. An initial depth feature determination unit is used to extract depth features from the target logging curve using a deep learning encoder to obtain the initial depth features of the target logging curve. The deep attention feature vector determination unit is used to perform feature compression on the initial depth features using the nonlinear transformation layer in the deep learning encoder to obtain the deep attention feature vector corresponding to the target logging curve.

[0052] In some specific embodiments, the feature recognition module 13 may specifically include: The target feature determination unit is used to determine the weight values ​​corresponding to the frequency domain feature vector and the deep attention feature vector using the linear mapping and gating unit, and to perform feature fusion on the frequency domain feature vector and the deep attention feature vector using the weight values ​​to obtain the target features.

[0053] In some specific embodiments, the feature recognition module 13 may specifically include: The classifier construction unit is used to compress the target features to obtain compressed features, and to construct a classifier based on a multilayer perceptron to obtain a preset lithology classifier. The lithology identification unit is used to classify and identify the compressed features using a preset lithology category classifier to obtain lithology identification results.

[0054] Furthermore, embodiments of this application also disclose an electronic device, Figure 6 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0055] Figure 6 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the well logging lithology identification method disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0056] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0057] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0058] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the well logging lithology identification method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include a computer program capable of performing other specific tasks.

[0059] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned well logging lithology identification method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0060] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0061] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0062] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0063] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0064] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A well logging lithology identification method, characterized in that, include: Obtain logging parameters related to lithological changes to obtain initial logging curves, and determine target logging curves from the initial logging curves based on the target window length; A fast Fourier transform is performed on the target logging curve to obtain the frequency domain feature vector corresponding to the target logging curve, and a deep learning encoder is used to extract the depth features of the target logging curve to obtain the depth attention feature vector corresponding to the target logging curve. The frequency domain feature vector and the deep attention feature vector are fused to obtain the target feature, and the target feature is classified and identified using a preset lithology classifier to obtain the lithology identification result.

2. The well logging lithology identification method according to claim 1, characterized in that, The process of acquiring logging parameters related to lithological changes to obtain an initial logging curve, and determining a target logging curve from the initial logging curve based on the target window length, includes: Several logging parameters related to lithological changes are acquired, and initial multidimensional logging curves are obtained using a preset data cleaning method, data standardization method, and feature optimization processing. The logging parameters include natural gamma, density, neutron porosity, sonic transit time, deep lateral resistivity, and shallow lateral resistivity logging data. The target logging curve is determined from the initial multidimensional logging curves based on the target window length.

3. The well logging lithology identification method according to claim 1, characterized in that, The step of performing a Fast Fourier Transform on the target logging curve to obtain the frequency domain feature vector corresponding to the target logging curve includes: Perform a fast Fourier transform on the target logging curve to obtain the amplitude spectrum corresponding to the target logging curve; A learnable frequency weight matrix is ​​introduced, and the amplitude spectrum is processed using the learnable frequency weight matrix and a dot product operation to obtain the frequency domain feature vector corresponding to the target logging curve; the learnable frequency weight matrix is ​​determined based on the frequency distribution corresponding to the logging parameters.

4. The well logging lithology identification method according to claim 1, characterized in that, The step of extracting depth features from the target logging curve using a deep learning encoder to obtain the depth attention feature vector corresponding to the target logging curve includes: The target logging curve is transposed and linearly mapped to obtain a target sequence, and the target sequence is input into a deep learning encoder for processing to obtain initial output features; An adaptive depth weight matrix is ​​introduced, and the initial output features are weighted using the adaptive depth weight matrix to obtain the depth attention feature vector corresponding to the target logging curve; the adaptive depth weight matrix is ​​set based on the lithology category.

5. The well logging lithology identification method according to claim 1, characterized in that, The step of performing a Fast Fourier Transform on the target logging curve to obtain the frequency domain feature vector corresponding to the target logging curve, and extracting depth features from the target logging curve using a deep learning encoder to obtain the depth attention feature vector corresponding to the target logging curve, includes: Perform a fast Fourier transform on the target logging curve to obtain the weighted spectrum corresponding to the target logging curve; The weighted spectrum is nonlinearly mapped and feature compression is performed using a multilayer perceptron to obtain the frequency domain feature vector corresponding to the target logging curve. A deep learning encoder is used to extract depth features from the target logging curve to obtain the initial depth features of the target logging curve. The initial depth features are compressed using the nonlinear transformation layer in the deep learning encoder to obtain the depth attention feature vector corresponding to the target logging curve.

6. The well logging lithology identification method according to claim 1, characterized in that, The feature fusion of the frequency domain feature vector and the deep attention feature vector to obtain the target features includes: The weight values ​​corresponding to the frequency domain feature vector and the deep attention feature vector are determined by using linear mapping and gating units, and the weight values ​​are used to fuse the frequency domain feature vector and the deep attention feature vector to obtain the target feature.

7. The well logging lithology identification method according to claim 1, characterized in that, The step of classifying and identifying the target features using a preset lithology classifier to obtain lithology identification results includes: The target features are compressed to obtain compressed features, and a classifier is built based on a multilayer perceptron to obtain a preset lithology classifier. The compressed features are classified and identified using a preset lithology classifier to obtain lithology identification results.

8. A well logging lithology identification device, characterized in that, include: The well logging curve acquisition module is used to acquire well logging parameters related to lithological changes to obtain an initial well logging curve, and to determine a target well logging curve from the initial well logging curve based on the target window length; The vector determination module is used to perform a fast Fourier transform on the target logging curve to obtain the frequency domain feature vector corresponding to the target logging curve, and to use a deep learning encoder to extract the depth features of the target logging curve to obtain the depth attention feature vector corresponding to the target logging curve. The feature recognition module is used to perform feature fusion on the frequency domain feature vector and the deep attention feature vector to obtain target features, and to classify and recognize the target features using a preset lithology classifier to obtain lithology recognition results.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the logging lithology identification method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store a computer program, which, when executed by a processor, implements the logging lithology identification method as described in any one of claims 1 to 7.

Images