A lithology intelligent classification system and method based on multi-modal data fusion

Abstract

The application relates to the technical field of lithology intelligent classification, in particular to a lithology intelligent classification system and method based on multi-modal data fusion, which comprises a data acquisition module, a feature extraction module, a cluster analysis module, a classification prediction module and a visualization module. Through preprocessing, sensitivity quantization and weighted feature space construction of the original drilling data of the tunnel, the improved K-means++ algorithm is combined to dynamically adjust the cluster number, generate the lithology category label, and display the distribution through the visualization interface. The application can effectively reduce noise interference, reasonably allocate feature weights, improve the lithology identification accuracy and robustness under complex geological conditions, and provide intuitive support for geological decision-making.

Description

Technical Field

[0001] This invention relates to the field of intelligent lithology classification technology, and more specifically, to an intelligent lithology classification system and method based on multimodal data fusion. Background Technology

[0002] In drilling engineering and geological exploration, lithology identification is a crucial step in ensuring construction safety and guiding the optimization of drilling parameters. Existing lithology identification methods mainly rely on well logging while drilling (MWD) data, drilling time parameters, and cuttings logging information, making judgments based on human experience or rule-based algorithms.

[0003] However, in practical applications, current technologies often suffer from the impact of operator habits, equipment performance fluctuations, and changes in the downhole environment on MWD data. The raw data contains a large amount of noise and outliers, leading to unstable subsequent analysis results. Furthermore, different measurement parameters respond differently to lithological changes, and traditional methods often fail to differentiate parameter sensitivity, resulting in low feature utilization and insufficient model discrimination capabilities. On the other hand, with increasing drilling depth and geological conditions, lithological boundaries exhibit ambiguity and diversity. Fixed cluster numbers or static classification models struggle to adapt to changing formation structures, and clustering results are prone to overfitting or misclassification, affecting the accuracy and robustness of lithology identification. Summary of the Invention

[0004] In view of this, the present invention proposes a lithology intelligent classification system and method based on multimodal data fusion, aiming to solve the problems in the current technology where MWD data is affected by multiple factors such as operating habits, equipment fluctuations and downhole environment, resulting in large data noise, ineffective differentiation of feature sensitivity differences, and clustering models with a fixed number of clusters that are difficult to adapt to complex and variable stratigraphic structures, leading to insufficient accuracy and stability in lithology identification.

[0005] This invention proposes a lithological intelligent classification method based on multimodal data fusion, comprising: The raw logging-while-drilling data is acquired and preprocessed, including eliminating dimensional differences based on normalization and removing distorted data points by combining outlier detection mechanism to form a standardized dataset. Based on principal component analysis or feature importance assessment methods, the sensitivity of multidimensional logging parameters is quantified to determine the contribution of each feature to lithological changes. Based on the contribution, higher weights are assigned to high-sensitivity features to construct a weighted feature space. An improved K-means++ algorithm is introduced into the weighted feature space, and the number of clusters is dynamically adjusted by combining the silhouette coefficient or Calinski-Harabasz index to generate lithological category labels. The real-time measurement data for the target period is input into the trained classification model, which outputs the lithology category prediction results and displays the distribution of various lithologies through a visualization interface.

[0006] Furthermore, the preprocessing of the raw logging-while-drilling data includes: Smoothing of the original signal is performed using the sliding window averaging method or wavelet transform. Normalization standardization is applied to all types of measurement parameters to eliminate biases caused by dimensional differences; Combine statistical methods such as Z-scores or box plots to remove data points that deviate from the normal range; Missing data points are filled using interpolation methods, and a standardized dataset is generated.

[0007] Furthermore, when quantifying the sensitivity of multidimensional logging parameters based on principal component analysis or characteristic importance assessment methods, the following are included: Dimensionality reduction is performed on the multidimensional logging parameters in the standardized dataset to extract the principal components and determine the variance contribution rate of the principal components; The contribution of each feature to lithological variation is determined based on the variance contribution rate; For cases where principal component analysis is not used, tree-based feature importance assessment methods, such as random forests or gradient boosting decision trees, are used to determine the importance score of each feature. Features are ranked according to their contribution or importance scores, and high-contribution features are assigned higher weights, thereby constructing a weighted feature space.

[0008] Furthermore, when introducing the improved K-means++ algorithm into the weighted feature space, it includes: When initializing cluster centers, the K-means++ algorithm is used as an optimization strategy to select the point farthest from the currently selected center as the new center; During each iteration, the distance from each sample point to the nearest cluster center is calculated, and the cluster center position is updated based on the distance. Clustering quality is assessed by combining the silhouette coefficient or the Calinski-Harabasz index, and the number of clusters is dynamically adjusted. When the clustering quality index reaches the optimal value, the iteration stops and the final clustering result is output, generating lithology category labels and associating them with the corresponding geological conditions.

[0009] Furthermore, when dynamically adjusting the number of clusters by combining the silhouette coefficient or the Calinski-Harabasz index, the following are included: The silhouette coefficient of each sample point is obtained, and the average silhouette coefficient of all sample points is determined as the evaluation index of the overall clustering quality. The acquisition of the silhouette coefficient of the sample point includes taking into account the closeness of the sample point to other sample points in its own cluster and the distance to other genera and clusters. For the Calinski-Harabasz index, obtain the ratio of the inter-cluster sum of squares to the intra-cluster sum of squares; Iterate through the clustering quality metrics under different numbers of clusters and obtain the number of clusters that reaches the maximum value as the optimal number of clusters; The optimal number of clusters is applied to the improved K-means++ algorithm to generate the final clustering results.

[0010] Furthermore, when inputting real-time measurement data for the target time period into the trained classification model, this includes: Perform the same preprocessing steps as the standardized dataset on the real-time measurement data for the target time period, including smoothing, normalization, and outlier removal. The preprocessed data is mapped to a weighted feature space to obtain its distance from each cluster center; The lithological category of the target data point is determined based on the principle of minimum distance; The lithology category labels of the target data points are bound to the corresponding timestamps to form time-series lithology identification results.

[0011] Furthermore, when displaying the distribution of various lithologies using a visual interface, this includes: By overlaying lithology category labels with geological profile maps, a lithology distribution map is generated; The visualization interface provides interactive functions, allowing users to adjust the display scale, switch lithology category colors, or view the lithology distribution over a specific time period, and export the lithology distribution map to various file formats; Through a real-time update mechanism, the latest lithology identification results are dynamically reflected in the visualization interface, enabling real-time formation identification and auxiliary decision support during the drilling process.

[0012] Furthermore, post-processing of lithological category labels includes: Consistency checks are performed on the lithological category labels of adjacent time steps. If frequent category switching is found in a short period of time, the majority voting method or sliding window smoothing method is used for correction. By combining prior geological knowledge, manual labeling or rule constraints are applied to areas with ambiguous lithological boundaries to improve the reliability of classification results; The post-processed lithology category labels are associated with and stored with the original measurement data to form a complete lithology identification database.

[0013] Furthermore, the maintenance and expansion of the lithology identification database includes: The measurement data and lithology category labels in the database are updated regularly, and an incremental learning mechanism is introduced to gradually integrate the new data into the training process of the classification model, thus avoiding retraining the entire model. The model performance is evaluated using cross-validation, and a model update process is triggered if a performance degradation is detected.

[0014] Compared with existing technologies, the advantages of this invention are as follows: By preprocessing the original logging-while-drilling data, denoising and smoothing algorithms effectively reduce interference caused by factors such as operating habits, equipment fluctuations, and downhole environment. Combined with an outlier detection mechanism, distorted data points are eliminated, making the input data purer and more reliable. This process significantly improves the stability of the data source, providing high-quality basic data support for subsequent feature extraction and cluster analysis. Secondly, by using principal component analysis or feature importance assessment methods to quantify the sensitivity of multidimensional logging parameters, the most representative key parameters of lithological changes can be accurately identified, and higher weights can be assigned to highly sensitive features, thereby enhancing the model's responsiveness to key geological information. This feature weighting strategy effectively solves the problem of uneven weighting of different parameters on lithology in traditional methods, improving the feature identification ability and overall accuracy of the classification model. Furthermore, an improved K-means++ algorithm is introduced in the clustering stage, and combined with the silhouette coefficient or Calinski-Harabasz index to achieve adaptive adjustment of the number of clusters, enabling the model to dynamically determine the optimal cluster structure based on the complexity of different strata and lithological boundary characteristics. This mechanism improves the model's adaptability and generalization performance under diverse geological conditions, avoiding overfitting and misclassification problems caused by a fixed number of clusters. Finally, by inputting real-time measurement data into the trained classification model and outputting the results in a visual form, dynamic display and intelligent prediction of lithological distribution are achieved. This not only improves the efficiency of real-time decision-making during drilling but also provides intuitive and reliable reference data for on-site geologists.

[0015] On the other hand, this application also provides a lithological intelligent classification system based on multimodal data fusion, including: The data acquisition module is configured to acquire raw logging data while drilling and preprocess the raw data to form a standardized dataset. The feature extraction module is configured to perform sensitivity quantification of multidimensional logging parameters through principal component analysis or feature importance assessment methods, and construct a weighted feature space. The clustering analysis module is configured to introduce an improved K-means++ algorithm into the weighted feature space and dynamically adjust the number of clusters by combining the silhouette coefficient or Calinski-Harabasz index to generate lithological category labels. The classification prediction module is configured to input real-time measurement data for the target time period into the trained classification model and output lithology category prediction results. The visualization module is configured to display the distribution of various lithologies through a visual interface, enabling real-time formation identification and decision support during the drilling process.

[0016] It is understood that the lithological intelligent classification system and method based on multimodal data fusion in the above embodiments of the present invention have the same beneficial effects, and will not be described in detail here. Attached Figure Description

[0017] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart illustrating a lithological intelligent classification method based on multimodal data fusion, provided in an embodiment of the present invention; Figure 2 A flowchart illustrating a lithological intelligent classification method based on multimodal data fusion provided in an embodiment of the present invention; Figure 3 This is a functional block diagram of a lithological intelligent classification system based on multimodal data fusion, provided for an embodiment of the present invention. Detailed Implementation

[0018] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] like Figures 1-2 As shown in some embodiments of this application, this embodiment provides a lithological intelligent classification method based on multimodal data fusion, including: Step S100: Obtain the raw logging-while-drilling data and preprocess the raw logging-while-drilling data, including eliminating dimensional differences based on normalization processing and removing distorted data points by combining an outlier detection mechanism to form a standardized dataset.

[0020] Specifically, the preprocessing of raw logging data includes: smoothing the raw signal based on the sliding window averaging method or wavelet transform; performing normalization standard processing on various measurement parameters to eliminate deviations caused by dimensional differences; removing data points that deviate from the normal range by combining statistical Z-score method or box plot method; completing missing data points using interpolation methods and generating a standardized dataset.

[0021] Understandably, multi-level signal optimization and statistical correction methods are used to eliminate noise interference and abnormal fluctuations, thereby constructing a stable and reliable standardized data foundation. First, smoothing the original signal using a sliding window averaging method or wavelet transform is a typical signal denoising and trend preservation technique. The sliding window averaging method reduces random fluctuations and preserves the main trend of formation changes by calculating local means on the time series; while wavelet transform achieves joint analysis of the frequency and time domains through multi-scale decomposition, effectively distinguishing high-frequency noise from useful geological signals, thus improving data smoothness and continuity without losing key features. Second, addressing the inconsistency in the dimensions of different logging parameters, the system performs normalization standardization processing, transforming each parameter to the same numerical range through linear mapping or standard deviation scaling. This allows the model to fairly measure the importance of each parameter during training and cluster analysis, avoiding bias caused by differences in numerical scales. This step ensures numerical consistency and feature comparability during multimodal data fusion. Furthermore, in terms of anomaly detection, this invention introduces statistical judgment mechanisms such as the Z-score method or box plot method. Z-scores calculate the standardized deviation of each sample based on the mean and standard deviation of the data. Outliers are identified when the deviation exceeds a set threshold. Box plots detect extreme values ​​using quartile ranges. Combining these two methods effectively eliminates distorted data points caused by instrument instability, signal drift, or operational abnormalities while maintaining sensitivity and robustness, thus improving the data's accuracy and representativeness. Finally, to prevent feature loss due to sensor packet loss or sampling interruptions from affecting the analysis results, the system uses interpolation algorithms (such as linear interpolation or spline interpolation) to smoothly complete missing data points, restoring the temporal continuity and spatial consistency of the data.

[0022] Step S200: Based on principal component analysis or feature importance assessment methods, the sensitivity of multidimensional logging parameters is quantified to determine the contribution of each feature to lithological changes, and high-sensitivity features are assigned higher weights according to their contribution to construct a weighted feature space.

[0023] Specifically, when performing sensitivity quantification of multidimensional logging parameters based on principal component analysis or feature importance assessment methods, the process includes: dimensionality reduction of multidimensional logging parameters in the standardized dataset, extraction of principal components, and determination of the variance contribution rate of the principal components; determination of the contribution of each feature to lithological variations based on the variance contribution rate; for cases where principal component analysis is not used, determination of the importance score of each feature based on tree model feature importance assessment methods, such as random forests or gradient boosting decision trees; ranking features according to contribution or importance scores, and assigning higher weights to features with high contribution, thereby constructing a weighted feature space.

[0024] Understandably, using Principal Component Analysis (PCA) or tree-based feature importance assessment methods to quantify and weight multidimensional logging parameters enhances the model's responsiveness to key geological features, reduces interference from redundant features, and optimizes the feature space. First, when using PCA, the standardized multidimensional logging parameter data undergoes linear dimensionality reduction. PCA calculates the sample covariance matrix and extracts eigenvectors, mapping the original high-dimensional data to the direction of the principal component with the largest variance, thus achieving optimal data structure representation. The variance contribution rate of each principal component reflects its proportion in the total information; a higher variance contribution rate indicates that the component carries more lithological variation information. By calculating and ranking the variance contribution rates of each principal component, the influence of each logging parameter on lithological variations can be quantified, providing a basis for subsequent feature weighting. This method effectively reduces data dimensionality, eliminates multicollinearity among features, and retains key geological information. Secondly, for cases where PCA dimensionality reduction is not used, this invention utilizes a tree-based feature importance assessment method (such as Random Forest or Gradient Boosting Decision Tree GBDT). These models repeatedly divide the sample feature space during training, statistically analyzing the contribution of each feature to the model's prediction error (e.g., information gain or Gini index) at split nodes, and calculate the feature importance score. Features with higher importance scores have a more significant impact on lithology classification. Compared to linear dimensionality reduction methods, tree models can capture the interactions between nonlinear features, making them suitable for multimodal feature coupling analysis under complex geological conditions. Finally, features are ranked and weighted based on the variance contribution rate extracted by PCA or the feature importance score calculated by the tree model. High-sensitivity or high-importance features are assigned higher weights, allowing them to have a greater impact on model decisions during subsequent clustering and classification processes, thereby improving the accuracy and stability of lithology identification. Through this weight optimization mechanism, the model can automatically strengthen the response of logging parameters sensitive to lithological changes while suppressing interference from redundant or noisy features.

[0025] Step S300: Introduce the improved K-means++ algorithm into the weighted feature space, and dynamically adjust the number of clusters by combining the silhouette coefficient or Calinski-Harabasz index to generate lithology category labels.

[0026] Specifically, when introducing the improved K-means++ algorithm into the weighted feature space, the following steps are taken: First, the K-means++ algorithm's optimization strategy is used to initialize cluster centers, selecting the point furthest from the currently selected center as the new center. Second, during each iteration, the distance from each sample point to the nearest cluster center is calculated, and the cluster center position is updated based on the distance. Third, the clustering quality is evaluated using the silhouette coefficient or Calinski-Harabasz index, and the number of clusters is dynamically adjusted. Fourth, when the clustering quality index reaches its optimal value, the iteration is stopped, the final clustering result is output, lithological category labels are generated, and these labels are associated with the corresponding geological conditions.

[0027] Specifically, when dynamically adjusting the number of clusters by combining the silhouette coefficient or the Calinski-Harabasz index, the process includes: obtaining the silhouette coefficient of each sample point and using the average silhouette coefficient of all sample points as the evaluation index for the overall clustering quality. Obtaining the silhouette coefficient of a sample point involves comprehensively considering the compactness between the sample point and other sample points within its own cluster, as well as its distance from other clusters. For the Calinski-Harabasz index, the ratio of the sum of squared deviations between clusters to the sum of squared deviations within a cluster is obtained. The clustering quality indexes under different numbers of clusters are iterated, and the number of clusters whose index reaches its maximum value is identified as the optimal number of clusters. The optimal number of clusters is then applied to the improved K-means++ algorithm to generate the final clustering result.

[0028] Specifically, firstly, the K-means++ algorithm, as an existing clustering initialization method, selects initial cluster centers based on a "distance-based probability distribution" to improve the representativeness of the cluster centers and avoid the local optima problem caused by random initialization in the traditional K-means algorithm. Specifically, after selecting the first center, K-means++ successively selects subsequent centers according to the principle that "the squared distance is positively correlated with the probability of a sample being selected as a center," ensuring that the new centers are located as far away from the existing centers as possible, thereby enhancing the dispersion of the initial centers and improving the stability of the clustering results. Building on this, the improved K-means++ method used in this invention further incorporates strategies such as feature weights and weighted distance metrics, ensuring that the center initialization not only meets the dispersion requirements but also better reflects the importance of different logging features. For example, the improved algorithm introduces a weighted feature space in the distance calculation, allowing highly sensitive features to contribute more to the distance, thus enhancing the physical significance of clustering in lithology identification tasks. Furthermore, for logging data with high noise levels and uneven feature distribution, this technique can also employ density constraints or local noise suppression strategies to further enhance the stability of cluster center selection. To quantitatively evaluate clustering results, this technique introduces existing clustering effectiveness metrics, including the silhouette coefficient and the Calinski-Harabasz (CH) index. The silhouette coefficient measures the tightness between samples of the same class and the separation between samples of different classes, ranging from -1 to 1. A value closer to 1 indicates better clustering performance, and it is a widely used evaluation method in existing cluster analysis. The Calinski-Harabasz index, on the other hand, is a variance-based evaluation metric. It calculates the ratio of inter-class dispersion to intra-class dispersion; a higher CH value indicates greater separation between classes and tighter intra-class clustering, resulting in a more reasonable cluster structure. In this technical solution, after generating preliminary clustering results using the improved K-means++ algorithm, the silhouette coefficient and the Calinski-Harabasz index are used as dual evaluation criteria to evaluate the cluster structure from the perspectives of "tightness-separation" and "reasonableness of variance distribution," respectively. By using a comprehensive scoring system with multiple indicators, the instability caused by the bias of a single indicator can be avoided, ensuring that the final selected clustering results have high reliability and interpretability.

[0029] It is understandable that by optimizing the initialization of cluster centers and the self-adjustment strategy of cluster numbers, efficient partitioning and stable identification of complex geological data distributions can be achieved, thereby generating more accurate and robust lithological classification results in the weighted feature space. First, in the cluster initialization stage, this invention adopts the optimization strategy of the K-means++ algorithm. Compared to the traditional K-means algorithm's random selection of initial center points, K-means++ calculates the distance distribution between samples and selected centers, prioritizing the sample furthest from the current center as the new center, thus maximizing the dispersion among initial cluster centers. This initialization method effectively avoids the shortcomings of traditional algorithms, such as getting trapped in local optima and clustering results depending on the initial point selection, ensuring good global distribution characteristics from the beginning of the clustering process. Second, in the clustering iteration stage, the Euclidean distance between each sample point and the nearest cluster center is calculated, and the center position is continuously updated, iteratively converging to the optimal state. This process is based on the weighted feature space, meaning that different logging parameters participate in the distance measurement according to their importance weights, making the model pay more attention to feature dimensions sensitive to lithological changes when calculating similarity. Therefore, the cluster boundaries better reflect real geological differences, achieving cluster optimization from "feature equivalence" to "feature-oriented". Furthermore, this invention introduces a dynamic cluster quality evaluation mechanism, using the Silhouette Coefficient or Calinski-Harabasz Index (CH Index) as evaluation indicators for cluster performance. The Silhouette Coefficient measures the rationality of the clustering results by combining the density of sample points within clusters with the separation between clusters; while the CH Index reflects the compactness and discriminative power of the overall cluster structure by calculating the ratio of the sum of squared deviations between clusters to the sum of squared deviations within clusters. By traversing the clustering results with different numbers of clusters, the above indicators are calculated in real time, and the cluster number with the highest evaluation value is selected as the optimal number of clusters. This adaptive mechanism breaks the limitation of traditional K-means clustering requiring manual pre-setting of the number of clusters, enabling the algorithm to automatically adjust the clustering complexity according to stratigraphic characteristics. Finally, when the clustering quality indicators reach their optimal values, iteration stops and the final clustering results are output, generating corresponding lithological category labels and associating them with geological conditions, thereby achieving automatic identification and intelligent classification of lithological characteristics. This mechanism can not only dynamically adapt to changes in stratigraphy and differences in parameter distribution, but also ensure the consistency and interpretability of clustering results in a geological sense.

[0030] Step S400: Input the real-time measurement data of the target time period into the trained classification model, output the lithology category prediction results, and display the distribution of various lithologies through a visualization interface.

[0031] Specifically, when inputting real-time measurement data of the target time period into the trained classification model, the process includes: performing the same preprocessing steps on the real-time measurement data of the target time period as on the standardized dataset, including smoothing, normalization, and outlier removal; mapping the preprocessed data to a weighted feature space and obtaining its distance from each cluster center; determining the lithology category of the target data point according to the minimum distance principle; and binding the lithology category label of the target data point with the corresponding timestamp to form a time-series lithology identification result.

[0032] Specifically, when displaying the distribution of various lithologies based on the visualization interface, the following features are included: overlaying lithology category labels with geological profile maps to generate lithology distribution maps; providing interactive functions in the visualization interface, allowing users to adjust the display scale, switch lithology category colors, or view lithology distribution over a specific time period, and exporting the lithology distribution maps to various file formats; and dynamically reflecting the latest lithology identification results in the visualization interface through a real-time update mechanism, thereby achieving real-time stratigraphic identification and auxiliary decision support during the drilling process.

[0033] Specifically, the post-processing of lithology category labels includes: performing consistency checks on lithology category labels at adjacent time steps; if frequent category switching is found within a short period of time, corrections are made using majority voting or sliding window smoothing methods; combining prior geological knowledge, manually labeling or constraining ambiguous lithology boundary areas to improve the reliability of classification results; and associating and storing the post-processed lithology category labels with the original measurement data to form a complete lithology identification database.

[0034] Specifically, the maintenance and expansion of the lithology identification database includes: regularly updating the measurement data and lithology category labels in the database, and introducing an incremental learning mechanism to gradually integrate new data into the training process of the classification model to avoid retraining the entire model; evaluating model performance based on cross-validation methods, and triggering the model update process if a performance degradation is found.

[0035] It is understandable that an adaptive and evolving lithology identification and decision support system is constructed through real-time data mapping, category prediction, visualization, result post-processing, and model incremental learning and maintenance mechanisms. Its technical principle lies in achieving high-precision prediction of lithology identification, improved result stability, and continuous model optimization through a consistent data processing flow and closed-loop feedback mechanism. First, in the real-time identification stage, this invention follows the same data preprocessing and feature mapping mechanism as the training stage. Standardization processes such as smoothing, normalization, and outlier removal are performed on the real-time measurement data for the target period to ensure consistency between the real-time data and training samples in statistical distribution and feature scale. Subsequently, the preprocessed data is mapped to a weighted feature space, and the lithology category of the sample is determined based on the minimum distance principle by calculating its distance to each cluster center. This process essentially utilizes the geological representativeness of cluster centers in the multimodal feature space, achieving rapid matching and intelligent discrimination between real-time data and lithology categories. Second, to achieve visualized monitoring and decision support for the drilling process, this invention constructs a lithology distribution visualization interface. The identified lithology category labels are spatially overlaid with geological profile maps to generate an intuitive lithology distribution map. Interactive functions are provided to allow users to freely adjust the display scale, filter specific time periods, switch colors, or export data. This visualization technology based on geological profile overlay not only improves data interpretability but also dynamically presents the latest identification results through a real-time refresh mechanism. This enables continuous stratigraphic monitoring and risk warning during drilling, providing data-driven decision support for geologists. Furthermore, this invention introduces a lithology category label post-processing mechanism to improve the temporal stability and geological consistency of the results. By verifying the consistency of lithology prediction results at adjacent time steps, if frequent category switching occurs within a short period, a majority voting method or sliding window smoothing method is used for correction, thereby eliminating misjudgments caused by local noise or instantaneous measurement fluctuations. In addition, artificial or rule-based constraints are applied to correct ambiguous lithology boundary areas using prior geological knowledge, making the identification results more consistent with geological laws and bedding continuity characteristics. This post-processing mechanism is essentially a time-series constraint and knowledge fusion strategy, making the prediction results spatially coherent and geologically interpretable. Finally, to ensure long-term stable operation and adaptive expansion, this invention constructs an incremental learning and model maintenance mechanism at the database management level. The measurement data and label samples in the lithology identification database are updated regularly, and new data is gradually integrated into the model training process based on an incremental learning algorithm, avoiding the computational burden of retraining the entire model. Simultaneously, cross-validation is used to periodically evaluate model performance; if a decrease in identification accuracy is detected, the model update process is automatically triggered, thereby maintaining the model's continuous adaptability to new geological environments and equipment characteristics.

[0036] In the above embodiments, by preprocessing the original logging-while-drilling data, denoising and smoothing algorithms effectively reduce interference caused by factors such as operating habits, equipment fluctuations, and downhole environment. Combined with an outlier detection mechanism, distorted data points are eliminated, making the input data purer and more reliable. This process significantly improves the stability of the data source, providing high-quality basic data support for subsequent feature extraction and cluster analysis. Secondly, by using principal component analysis or feature importance assessment methods to quantify the sensitivity of multidimensional logging parameters, the most representative key parameters of lithological changes can be accurately identified, and higher weights can be assigned to highly sensitive features, thereby enhancing the model's responsiveness to key geological information. This feature weighting strategy effectively solves the problem of uneven weighting of different parameters on lithology in traditional methods, improving the feature identification ability and overall accuracy of the classification model. Furthermore, an improved K-means++ algorithm is introduced in the clustering stage, and combined with the silhouette coefficient or Calinski-Harabasz index to achieve adaptive adjustment of the number of clusters, enabling the model to dynamically determine the optimal cluster structure based on the complexity of different strata and lithological boundary characteristics. This mechanism improves the model's adaptability and generalization performance under diverse geological conditions, avoiding overfitting and misclassification problems caused by a fixed number of clusters. Finally, by inputting real-time measurement data into the trained classification model and outputting the results in a visual form, dynamic display and intelligent prediction of lithological distribution are achieved. This not only improves the efficiency of real-time decision-making during drilling but also provides intuitive and reliable reference data for on-site geologists.

[0037] In another preferred embodiment based on the above embodiments, such as Figure 3 As shown, this embodiment provides a lithological intelligent classification system based on multimodal data fusion, including: a data acquisition module, a feature extraction module, a cluster analysis module, a classification prediction module, and a visualization module.

[0038] Specifically, the data acquisition module is configured to acquire raw logging data while drilling and preprocess the raw data to form a standardized dataset; the feature extraction module is configured to perform sensitivity quantification of multidimensional logging parameters through principal component analysis or feature importance assessment methods to construct a weighted feature space; the clustering analysis module is configured to introduce an improved K-means++ algorithm into the weighted feature space and dynamically adjust the number of clusters by combining the silhouette coefficient or Calinski-Harabasz index to generate lithology category labels; the classification prediction module is configured to input real-time measurement data of the target time period into the trained classification model and output lithology category prediction results; and the visualization module is configured to display the distribution of various lithologies through a visualization interface, realizing real-time formation identification and auxiliary decision support during the drilling process.

[0039] Specifically, the data acquisition module, feature extraction module, cluster analysis module, classification prediction module, and visualization module are logically connected to enable data flow and processing, thereby completing the entire process from raw well logging while drilling (MWD) data to the generation of lithology category labels.

[0040] In practical applications, the data acquisition module first acquires raw logging-while-drilling data through sensors or other measuring devices. This data typically contains multiple parameters, such as natural gamma, resistivity, density, and sonic transit time. To reduce the impact of noise and outliers on subsequent processing, the data acquisition module performs preprocessing operations on the raw data. Specifically, a sliding window averaging method is used to smooth the raw signal to reduce the impact of short-term random disturbances; simultaneously, wavelet transform is used to further remove high-frequency noise, ensuring data stability. Subsequently, normalization standardization processing is performed on various measurement parameters to unify data of different dimensions to the same numerical range, avoiding deviations caused by differences in dimensions. In addition, statistical outlier detection mechanisms are introduced, such as Z-scores or box plots, to remove data points that deviate from the normal range. For missing data points, interpolation methods are used to complete them, ensuring data integrity. After the above steps, a standardized dataset is formed, laying the foundation for subsequent feature extraction and modeling analysis.

[0041] The feature extraction module receives a standardized dataset from the data acquisition module and performs sensitivity-based quantification. Specifically, the feature extraction module first performs dimensionality reduction on the multidimensional logging parameters, then uses principal component analysis (PCA) to extract the principal components and calculate their variance contribution rates. The variance contribution rate reflects the degree to which each feature contributes to lithological changes, and based on this, the importance ranking of each feature is determined. For cases where PCA is not used, the feature extraction module uses tree-based feature importance assessment methods, such as random forests or gradient boosting decision trees, to calculate the importance score of each feature. After ranking the features according to their contribution or importance scores, higher-contribution features are assigned higher weights, thus constructing a weighted feature space. This process ensures the dominant role of key features in subsequent cluster analysis, improving the model's ability to capture lithological change signals.

[0042] The clustering analysis module receives data from the weighted feature space and incorporates an improved K-means++ algorithm for clustering analysis. During initialization, the selection of cluster centers employs the optimization strategy of the K-means++ algorithm, choosing the point furthest from the currently selected centers as the new centers to avoid over-concentration of initial centers. In each iteration, the distance from each sample point to the nearest cluster center is calculated, and the cluster center positions are updated based on this distance. To dynamically adjust the number of clusters, the clustering analysis module combines the silhouette coefficient or the Calinski-Harabasz index to evaluate clustering quality. Specifically, the silhouette coefficient of each sample point is calculated, which comprehensively considers the compactness of the sample point with other points within its own cluster and its distance from other clusters; simultaneously, the average silhouette coefficient of all sample points is calculated as an evaluation index of overall clustering quality. For the Calinski-Harabasz index, the ratio of the inter-cluster sum of squares to the intra-cluster sum of squares is calculated; a higher ratio indicates better clustering performance. By iterating through clustering quality indices under different cluster numbers, the optimal cluster number is selected based on the cluster number that maximizes the indices. This optimal cluster number is then applied to an improved K-means++ algorithm to generate the final clustering results. The lithology category labels output by the clustering analysis module are correlated with the corresponding geological conditions to support subsequent classification predictions.

[0043] The classification and prediction module receives real-time measurement data for the target time period and performs the same preprocessing steps as the standardized dataset, including smoothing, normalization, and outlier removal. The preprocessed data is mapped to a weighted feature space, and its distance to each cluster center is calculated. Based on the minimum distance principle, the lithology category of the target data point is determined, and the lithology category label of the target data point is bound to the corresponding timestamp, forming a time-series lithology identification result. The output of the classification and prediction module not only reflects the current lithological distribution of the strata but also provides support for real-time strata identification and auxiliary decision-making during drilling.

[0044] The visualization module receives lithology category labels output by the classification and prediction module and overlays them onto the geological profile to generate a lithology distribution map. The visualization interface offers rich interactive functions, allowing users to adjust the display scale, switch lithology category colors, or view lithology distribution over a specific time period. Furthermore, the visualization module supports exporting lithology distribution maps to various file formats, such as PDF or SVG, facilitating subsequent analysis and report writing. Through a real-time update mechanism, the visualization module dynamically reflects the latest lithology identification results on the interface, providing geological engineers with intuitive data support.

[0045] In actual operation, the collaboration between the above modules is as follows: The data acquisition module is responsible for acquiring and preprocessing the raw logging data while drilling, forming a standardized dataset, and then passing it to the feature extraction module; the feature extraction module constructs a weighted feature space by performing sensitivity quantification on multidimensional logging parameters, and passes the data in the weighted feature space to the clustering analysis module; the clustering analysis module generates lithology category labels using an improved K-means++ algorithm and passes them to the classification prediction module; the classification prediction module maps the real-time measurement data of the target period to the weighted feature space, outputs the lithology category prediction results, and passes the results to the visualization module; the visualization module displays the lithology distribution through an interface, providing support for geological decision-making.

[0046] Furthermore, to improve the reliability of the classification results, the system also performs post-processing on the lithology category labels. Specifically, consistency checks are performed on the lithology category labels of adjacent time steps. If frequent category switching is detected within a short period, a majority voting method or a sliding window smoothing method is used for correction. Combined with prior geological knowledge, areas with ambiguous lithology boundaries are manually labeled or subject to rule constraints to further improve the accuracy of the classification results. The post-processed lithology category labels are then linked and stored with the original measurement data to form a complete lithology identification database.

[0047] Maintaining and expanding the lithology identification database is crucial for the long-term operation of the system. Regularly updating the measurement data and lithology category labels in the database ensures data timeliness. An incremental learning mechanism is introduced, gradually incorporating new data into the training process of the classification model, avoiding the need to retrain the entire model. Model performance is evaluated using cross-validation; if performance degradation is detected, a model update process is triggered. Multi-user collaboration is supported, allowing multiple geological engineers to simultaneously access and edit the lithology identification database, improving work efficiency.

[0048] Through the above specific implementation methods, the present invention achieves adaptive classification of complex geological conditions, improves the accuracy and robustness of lithology identification, and provides intuitive data support for geological decision-making.

[0049] To enable those skilled in the art to fully understand and implement this invention, the specific implementation principle of this invention is further explained below in conjunction with a specific application scenario.

[0050] In actual drilling operations, suppose a geological exploration team is performing lithology identification on a deep well. This well is located in a complex geological structure area with frequent changes in lithology and blurred boundaries, making it difficult for traditional methods to meet the high-precision identification requirements. In this case, a lithology intelligent classification system and method based on multimodal data fusion can be applied.

[0051] First, the data acquisition module acquires raw logging-while-drilling (MWD) data through sensors installed on the drilling rig. This data includes parameters such as natural gamma ray, resistivity, density, and sonic transit time. Due to the complex downhole environment, noise and outliers are unavoidable in the raw data. To reduce the impact of these interference factors, the data acquisition module uses a sliding window averaging method to smooth the signal. For example, for natural gamma ray data, a time window of length 5 is selected, and the mean value of the data within the window is calculated to replace the value of the center point, thereby reducing the impact of short-term random disturbances. Simultaneously, wavelet transform is used to decompose and remove high-frequency noise, ensuring data stability. Subsequently, normalization processing is performed on various measurement parameters, unifying data of different dimensions into the range [0,1]. For example, for resistivity data, a minimum-maximum normalization formula is used to convert it into a standardized value. Furthermore, outliers are detected based on the Z-score method; if the absolute value of the Z-score of a data point is greater than 3, it is identified as an outlier and removed. For missing data points, linear interpolation is used to complete the data, ensuring data integrity. After the above steps, a standardized dataset is formed, laying the foundation for subsequent analysis.

[0052] Next, the feature extraction module receives the standardized dataset and performs sensitivity quantification on it. Specifically, the feature extraction module first uses principal component analysis (PCA) to reduce the dimensionality of the multidimensional logging parameters. For example, the first three principal components extracted from the original dataset have a cumulative variance contribution rate of over 85%, indicating that these principal components can effectively reflect the main information of lithological changes. Based on this, the importance ranking of each feature is determined, and features with high contribution are assigned higher weights. For example, natural gamma and resistivity have variance contribution rates of 40% and 30%, respectively, and are therefore assigned higher weight values. For cases where PCA is not used, the feature extraction module uses a random forest model to calculate the importance score of each feature. For example, by constructing decision trees 100 times on the training dataset and counting the frequency of use of each feature in the split nodes, the final importance score of natural gamma is 0.35, and that of resistivity is 0.28. After ranking the features according to their contribution or importance scores, a weighted feature space is constructed to ensure the dominant role of key features in subsequent clustering analysis.

[0053] Subsequently, the clustering analysis module receives data from the weighted feature space and introduces an improved K-means++ algorithm for clustering analysis. During the initialization phase, the selection of cluster centers employs the optimization strategy of the K-means++ algorithm. For example, a sample point is first randomly selected as the initial cluster center. Then, the distances from the remaining sample points to the selected center are calculated, and the point furthest away is selected as the second center, and so on, until all initial centers are selected. In each iteration, the distance from each sample point to the nearest cluster center is calculated, and the cluster center position is updated based on the distance. To dynamically adjust the number of clusters, the clustering analysis module combines silhouette coefficients to evaluate clustering quality. For example, the silhouette coefficient of each sample point is calculated, comprehensively considering its compactness with other points within its own cluster and its distance from other clusters. Assuming the average silhouette coefficient of a cluster is 0.7, it indicates that the clustering effect of that cluster is good. By traversing the clustering quality indicators under different cluster numbers, the number of clusters that maximizes the average silhouette coefficient is selected as the optimal number of clusters. For example, when the number of clusters is 5, the average silhouette coefficient reaches a peak of 0.72. Therefore, 5 is used as the optimal number of clusters in the improved K-means++ algorithm to generate the final clustering result.

[0054] The classification and prediction module receives real-time measurement data for the target time period and performs the same preprocessing steps as on the standardized dataset. For example, it performs sliding window averaging and normalization on the real-time acquired natural gamma and resistivity data to ensure data format consistency. The preprocessed data is mapped to a weighted feature space, and its distance to each cluster center is calculated. For example, assuming that the distances of a real-time data point to the five cluster centers are 1.2, 2.5, 0.8, 3.1, and 1.9, the data point is determined to belong to the lithology category corresponding to the third cluster center based on the minimum distance principle. The lithology category label of the target data point is bound to the corresponding timestamp to form a time-series lithology identification result. For example, the data point recorded at the 10th minute is labeled as sandstone and stored in the database.

[0055] The visualization module receives lithology category labels output by the classification and prediction module and overlays them onto the geological profile to generate a lithology distribution map. The visualization interface offers rich interactive features. For example, users can adjust the display scale to view the lithology distribution at different depths or switch lithology category colors for more intuitive differentiation. Furthermore, the visualization module supports exporting lithology distribution maps to PDF or SVG format files, facilitating subsequent analysis and report writing. Through a real-time update mechanism, the visualization module dynamically reflects the latest lithology identification results on the interface. For example, when a new data point at the 15th minute is classified as mudstone, the visualization interface immediately updates to display the lithology distribution for that time period, providing geological engineers with intuitive data support.

[0056] To improve the reliability of classification results, the system also performs post-processing on lithological category labels. For example, consistency checks are performed on lithological category labels between adjacent time steps. If frequent category changes are detected within a short period, a sliding window smoothing method is used for correction. For instance, if the data points from minute 10 to minute 14 are classified as sandstone, mudstone, sandstone, mudstone, and sandstone respectively, the sliding window smoothing method will uniformly correct them to the sandstone category. Furthermore, based on prior geological knowledge, areas with ambiguous lithological boundaries are manually labeled or subject to rule constraints. For example, within the depth range of 2000 to 2010 meters, geological data indicates that this area is primarily a transition zone; therefore, the lithological category labels for this range are manually labeled to further improve the accuracy of the classification results.

[0057] Maintaining and expanding the lithology identification database is crucial for the long-term operation of the system. For example, regularly updating the measurement data and lithology category labels in the database ensures data timeliness. An incremental learning mechanism is introduced to gradually integrate new data into the training process of the classification model. For instance, when 100 new data points are added, only these data points are updated in the model, without retraining the entire model. Model performance is evaluated using cross-validation methods, and a model update process is triggered if performance deteriorates. For example, a five-fold cross-validation method is used to evaluate model accuracy; if the accuracy falls below 90%, a model update process is triggered. Multi-user collaboration is supported, allowing multiple geological engineers to access and edit the lithology identification database simultaneously, improving work efficiency.

[0058] Through the implementation steps of the above specific application scenarios, this invention achieves adaptive classification of complex geological conditions, improves the accuracy and robustness of lithology identification, and provides intuitive data support for geological decision-making. The above scenarios are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0059] It is understood that the lithological intelligent classification system and method based on multimodal data fusion in the above embodiments of the present invention have the same beneficial effects, and will not be described in detail here.

[0060] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0061] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0062] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0063] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A lithological intelligent classification method based on multimodal data fusion, characterized in that, include: The raw logging-while-drilling data is acquired and preprocessed. The preprocessing includes eliminating dimensional differences based on normalization and removing distorted data points by combining an outlier detection mechanism to form a standardized dataset. Based on principal component analysis or feature importance assessment methods, the sensitivity of multidimensional logging parameters is quantified to determine the contribution of each feature to lithological changes. Based on the contribution, higher weights are assigned to high-sensitivity features to construct a weighted feature space. An improved K-means++ algorithm is introduced into the weighted feature space, and the number of clusters is dynamically adjusted by combining the silhouette coefficient or Calinski-Harabasz index to generate lithological category labels. The real-time measurement data for the target period is input into the trained classification model, which outputs the lithology category prediction results and displays the distribution of various lithologies through a visualization interface.

2. The lithological intelligent classification method based on multimodal data fusion as described in claim 1, characterized in that, Preprocessing of raw logging data during drilling includes: Smoothing of the original signal is performed using the sliding window averaging method or wavelet transform. Normalization standardization is applied to all types of measurement parameters to eliminate biases caused by dimensional differences; Combine statistical methods such as Z-scores or box plots to remove data points that deviate from the normal range; Missing data points are filled using interpolation methods, and a standardized dataset is generated.

3. The lithological intelligent classification method based on multimodal data fusion as described in claim 1, characterized in that, When quantifying the sensitivity of multidimensional logging parameters based on principal component analysis or eigenvalue importance assessment methods, the following methods are included: Dimensionality reduction is performed on the multidimensional logging parameters in the standardized dataset to extract the principal components and determine the variance contribution rate of the principal components; The contribution of each feature to lithological variation is determined based on the variance contribution rate; For cases where principal component analysis is not used, tree-based feature importance assessment methods, such as random forests or gradient boosting decision trees, are used to determine the importance score of each feature. Features are ranked according to their contribution or importance scores, and high-contribution features are assigned higher weights, thereby constructing a weighted feature space.

4. The lithological intelligent classification method based on multimodal data fusion as described in claim 1, characterized in that, When introducing the improved K-means++ algorithm into the weighted feature space, it includes: When initializing cluster centers, the K-means++ algorithm is used as an optimization strategy to select the point farthest from the currently selected center as the new center; During each iteration, the distance from each sample point to the nearest cluster center is calculated, and the cluster center position is updated based on the distance. Clustering quality is assessed by combining the silhouette coefficient or the Calinski-Harabasz index, and the number of clusters is dynamically adjusted. When the clustering quality index reaches the optimal value, the iteration stops and the final clustering result is output, generating lithology category labels and associating them with the corresponding geological conditions.

5. The lithological intelligent classification method based on multimodal data fusion as described in claim 4, characterized in that, When dynamically adjusting the number of clusters by combining the silhouette coefficient or the Calinski-Harabasz index, the following applies: The silhouette coefficient of each sample point is obtained, and the average silhouette coefficient of all sample points is determined as the evaluation index of the overall clustering quality. The acquisition of the silhouette coefficient of the sample point includes taking into account the closeness of the sample point to other sample points in its own cluster and the distance to other genera and clusters. For the Calinski-Harabasz index, obtain the ratio of the inter-cluster sum of squares to the intra-cluster sum of squares; Iterate through the clustering quality metrics under different numbers of clusters and obtain the number of clusters that reaches the maximum value as the optimal number of clusters; The optimal number of clusters is applied to the improved K-means++ algorithm to generate the final clustering results.

6. The lithological intelligent classification method based on multimodal data fusion as described in claim 1, characterized in that, When inputting real-time measurement data for the target time period into the trained classification model, the following is included: Perform the same preprocessing steps as the standardized dataset on the real-time measurement data for the target time period, including smoothing, normalization, and outlier removal. The preprocessed data is mapped to a weighted feature space to obtain its distance from each cluster center; The lithological category of the target data point is determined based on the principle of minimum distance; The lithology category labels of the target data points are bound to the corresponding timestamps to form time-series lithology identification results.

7. The lithological intelligent classification method based on multimodal data fusion as described in claim 1, characterized in that, When displaying the distribution of various lithologies using a visual interface, including: By overlaying lithology category labels with geological profile maps, a lithology distribution map is generated; The visualization interface provides interactive functions, allowing users to adjust the display scale, switch lithology category colors, or view the lithology distribution over a specific time period, and export the lithology distribution map to various file formats; Through a real-time update mechanism, the latest lithology identification results are dynamically reflected in the visualization interface, enabling real-time formation identification and auxiliary decision support during the drilling process.

8. The lithological intelligent classification method based on multimodal data fusion as described in claim 1, characterized in that, Post-processing of lithological category labels includes: Consistency checks are performed on the lithological category labels of adjacent time steps. If frequent category switching is found in a short period of time, the majority voting method or sliding window smoothing method is used for correction. By combining prior geological knowledge, manual labeling or rule constraints are applied to areas with ambiguous lithological boundaries to improve the reliability of classification results; The post-processed lithology category labels are associated with and stored with the original measurement data to form a complete lithology identification database.

9. The lithological intelligent classification method based on multimodal data fusion as described in claim 8, characterized in that, Maintaining and expanding the lithology identification database includes: The measurement data and lithology category labels in the database are updated regularly, and an incremental learning mechanism is introduced to gradually integrate the new data into the training process of the classification model, thus avoiding retraining the entire model. The model performance is evaluated using cross-validation, and a model update process is triggered if a performance degradation is detected.

10. A lithological intelligent classification system based on multimodal data fusion, applicable to the lithological intelligent classification method based on multimodal data fusion as described in any one of claims 1-9, characterized in that, include: The data acquisition module is configured to acquire raw logging data while drilling and preprocess the raw data to form a standardized dataset. The feature extraction module is configured to perform sensitivity quantification of multidimensional logging parameters through principal component analysis or feature importance assessment methods, and construct a weighted feature space. The clustering analysis module is configured to introduce an improved K-means++ algorithm into the weighted feature space and dynamically adjust the number of clusters by combining the silhouette coefficient or Calinski-Harabasz index to generate lithological category labels. The classification prediction module is configured to input real-time measurement data for the target time period into the trained classification model and output lithology category prediction results. The visualization module is configured to display the distribution of various lithologies through a visual interface, enabling real-time formation identification and decision support during the drilling process.

Images