While-drilling lithology identification method, system and equipment and medium

Abstract

The invention provides a while-drilling lithology identification method, system, equipment and medium, and belongs to the field of petroleum and natural gas drilling engineering, and the method comprises the steps: carrying out the feature extraction and fusion of while-drilling time series data based on a multi-scale convolutional neural network of a lithology identification model, and obtaining a multi-scale fusion feature; based on a bidirectional gating circulation unit in the lithology identification model, extracting global time sequence features containing a context dependency relationship according to front-back association of a multi-scale fusion feature sequence in the forward and backward sequence directions; the multi-scale fusion feature sequence comprises a plurality of multi-scale fusion features arranged according to a time sequence; and obtaining a lithology category identification result based on the global time sequence characteristics. According to the method, while-drilling lithology identification is accurately and stably carried out in real time.

Description

Technical Field

[0001] This invention relates to the field of oil and gas drilling engineering, specifically to a method, system, equipment, and medium for identifying lithology while drilling. Background Technology

[0002] In oil and gas drilling engineering, accurate and real-time lithology identification is a crucial step in understanding formation geological characteristics, optimizing drilling parameters, and ensuring operational safety and efficiency. Faced with complex and ever-changing underground geological conditions, the ability to quickly and reliably determine the lithology encountered during drilling directly impacts the scientific nature of drilling decisions and risk control capabilities.

[0003] To address this need, the industry has developed two main types of technologies: traditional methods and data-driven intelligent methods. Traditional methods, represented by borehole coring, cuttings logging, and traditional logging (wireless / early logging-while-drilling), are either costly and inefficient, or rely on human experience and lack real-time performance. To overcome these limitations, intelligent recognition technologies based on real-time drilling data have emerged. In their early stages, they mainly relied on traditional machine learning models (such as support vector machines and random forests), while later stages have shifted to deep learning methods that can automatically extract features.

[0004] However, existing intelligent identification technologies still have significant shortcomings. Traditional machine learning methods heavily rely on manual feature engineering, making it difficult to capture complex high-dimensional nonlinear patterns in drilling time-series data. Meanwhile, current mainstream deep learning models (such as ordinary convolutional neural networks or unidirectional recurrent neural networks) often exhibit poor robustness to drilling parameter fluctuations and struggle to fully capture long-term dependencies in formation along the depth direction, especially exhibiting unstable identification accuracy in lithological gradients or thin interbedded sections. Therefore, a new method is urgently needed that can deeply integrate multi-scale features and effectively model temporal context dependencies to improve the accuracy, real-time performance, and stability of drilling lithology identification under complex geological conditions. Summary of the Invention

[0005] In view of this, it is necessary to provide a drilling lithology identification method, system, equipment and medium to solve the technical problems of high dependence on manual work, insufficient real-time performance, insufficient accuracy and stability of lithology identification in the existing technology.

[0006] To address the aforementioned technical problems, in a first aspect, the present invention provides a method for identifying lithology while drilling, comprising: A multi-scale convolutional neural network based on a lithology identification model is used to extract and fuse features from drilling time-series data to obtain multi-scale fused features. Based on the bidirectional gated recurrent unit in the lithology identification model, global temporal features containing contextual dependencies are extracted according to the sequential correlation of the multi-scale fusion feature sequence in both the forward and reverse sequence directions; the multi-scale fusion feature sequence includes multiple multi-scale fusion features arranged in chronological order. The lithology category identification result is obtained based on the global temporal features.

[0007] In one possible implementation, before the multi-scale convolutional neural network based on the lithology identification model extracts and fuses features from the drilling time-series data to obtain multi-scale fused features, it includes: Obtain raw drilling data and calculate local statistics of the raw drilling data based on a sliding window; Abnormal data in the original drilling data are identified and corrected based on the local statistics.

[0008] In one possible implementation, identifying and correcting anomalous data in the original drilling data based on the local statistics includes: If the number of consecutive abnormal data within the sliding window is less than a first threshold, the abnormal data is corrected using linear interpolation. If the number of consecutive abnormal data within the sliding window is greater than or equal to the first threshold, then cubic spline interpolation is used to correct the abnormal data.

[0009] In one possible implementation, the multi-scale convolutional neural network includes at least three sets of parallel-connected one-dimensional convolutional layers, each set of one-dimensional convolutional layers employing convolutional kernels with different time dimensions. The multi-scale convolutional neural network based on the lithology identification model extracts and fuses features from drilling time-series data to obtain multi-scale fused features, including: The drilling time-series data is input in parallel into the at least three sets of one-dimensional convolutional layers for convolution operations, and local features of different time windows are extracted to obtain corresponding feature maps; The feature maps output by each group of one-dimensional convolutional layers are sequentially subjected to non-linear activation and global average pooling. All processed features are spliced ​​together to form the multi-scale fused feature.

[0010] In one possible implementation, the bidirectional gated loop unit includes a forward gated loop unit and a reverse gated loop unit arranged in parallel and processing in opposite directions. The bidirectional gated loop unit based on the lithology identification model extracts global temporal features containing contextual dependencies based on the sequential correlation of the multi-scale fused feature sequences in the forward and reverse sequence directions, including: The multi-scale fused feature sequence is simultaneously input into the forward gated loop unit and the reverse gated loop unit, wherein the forward gated loop unit processes the multi-scale fused feature sequence in the time-increasing direction, and the reverse gated loop unit processes the multi-scale fused feature sequence in the time-decreasing direction. The forward hidden state sequence obtained by the forward gated loop unit and the reverse hidden state sequence obtained by the reverse gated loop unit are obtained respectively. The forward hidden state sequence and the reverse hidden state sequence are concatenated to obtain deep temporal features; The global time-series features are obtained by performing global average pooling on the deep temporal features.

[0011] In one possible implementation, the bidirectional gated recurrent unit is followed by a fully connected layer and an output layer; the lithology category identification result obtained based on the global temporal features includes: Based on the fully connected layer, the global temporal features are subjected to linear transformation, nonlinear activation processing, and random deactivation processing to obtain the processed temporal features. Based on the output layer and the processed temporal features, the probability distribution of generated lithology categories is predicted to obtain the identification result.

[0012] In one possible implementation, a training step for the lithology identification model is also included: The difference between the identification result and the actual lithology label is calculated using the cross-entropy loss function; With the goal of minimizing the difference, the model parameters of the lithology identification model are iteratively updated using an adaptive moment estimation algorithm. During the training process, the identification accuracy of the validation set is monitored. If the accuracy does not improve for a predetermined number of consecutive rounds, the training is terminated.

[0013] Secondly, the present invention also provides a drilling lithology identification system, comprising: The multi-scale feature extraction module is used to extract and fuse features from drilling time-series data based on a multi-scale convolutional neural network of a lithology identification model to obtain multi-scale fused features. The global temporal extraction module is used to extract global temporal features containing contextual dependencies based on the bidirectional gated loop unit in the lithology identification model and the sequential correlation of the multi-scale fusion feature sequence in both forward and reverse sequence directions; the multi-scale fusion feature sequence includes multiple multi-scale fusion features arranged in chronological order. The classification output module is used to obtain the identification result of lithology category based on the global time series features.

[0014] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the drilling lithology identification method described in any of the above implementations.

[0015] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement the steps of the drilling lithology identification method described in any of the above implementations.

[0016] The beneficial effects of this invention are as follows: The drilling lithology identification method provided by this invention first constructs a hierarchical and complementary deep neural network architecture by organically combining a multi-scale convolutional neural network and a bidirectional gated recurrent unit. The multi-scale convolutional neural network extracts lithological features of different granularities in the spatial dimension, while the bidirectional gated recurrent unit models the contextual dependencies of the formation sequence in the temporal dimension. Their collaborative work achieves efficient parsing of drilling time-series data, significantly improving the accuracy and real-time performance of lithology identification. Furthermore, it possesses good geological adaptability and engineering practicality, providing reliable technical support for intelligent drilling and refined geological guidance. Further, by simultaneously capturing local features of different temporal granularities in the drilling time-series data through the multi-scale convolutional neural network, multi-scale local features related to lithology are comprehensively and accurately extracted. Through the bidirectional scanning of the bidirectional gated recurrent unit, sequence dependencies in the depth direction can be established, making it particularly sensitive to formation changes such as lithological transition zones, thus enhancing lithology identification capabilities. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic flowchart of an embodiment of the drilling lithology identification method provided by the present invention; Figure 2 This is a schematic flowchart of another embodiment of the drilling lithology identification method provided by the present invention; Figure 3 For the present invention Figure 1 A schematic diagram of an embodiment of S101; Figure 4 For the present invention Figure 1A schematic diagram of an embodiment of S102; Figure 5 For the present invention Figure 1 A schematic diagram of an embodiment of S103; Figure 6 This diagram illustrates the performance evolution of the lithology identification model of this invention during the training process, based on four core evaluation dimensions: accuracy, precision, F1 score, and recall. Figure 7 This is a schematic diagram of an embodiment of the drilling lithology identification system provided by the present invention; Figure 8 A schematic diagram of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0021] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.

[0022] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0023] Before demonstrating the embodiments, the following terms will be explained.

[0024] Lithology, short for rock properties, is a core term in geology, referring to the essential characteristics and physicochemical properties of rocks that can be used for identification and classification. Simply put, lithology is the basis for distinguishing different types of rocks. Lithology mainly includes the following characteristics: mineral composition (which minerals make up the rock and their proportions, such as quartz, feldspar, calcite, clay minerals, etc.), structure (the size, shape, and arrangement of mineral grains, such as coarse-grained, fine-grained, crystalline, clastic, etc.), texture (the macroscopic morphology and characteristics of the rock, such as bedding, foliation, massive, porosity, etc.), color, hardness, density, magnetism, and other physical properties.

[0025] During drilling operations, the rock type and characteristics of the formation that the drill bit is in contact with are analyzed and determined in real time using data collected in real time by the drilling measurement instrument (such as drilling pressure, rotation speed, torque, gamma rays, resistivity, etc.).

[0026] This invention provides a method, system, device, and medium for identifying lithology while drilling, which will be described below.

[0027] Figure 1 This is a schematic flowchart of an embodiment of the drilling lithology identification method provided by the present invention, as shown below. Figure 1 As shown, the drilling lithology identification method includes: S101. A multi-scale convolutional neural network based on a lithology identification model is used to extract and fuse features from drilling time-series data to obtain multi-scale fused features.

[0028] It should be noted that the lithology identification model adopts a serial structure of "Multi-scale Convolutional Neural Network (MSCNN) - Bidirectional Gated Recurrent Unit (BiGRU) - Fully Connected Layer - Output Layer". The functions of each layer work together to realize the complete process from data feature extraction and time-series dependency modeling to lithology category prediction. The drilling time-series data is input into the multi-scale convolutional neural network, which extracts multi-dimensional and multi-granular local features from the drilling time-series data and performs feature fusion to obtain multi-scale fused features.

[0029] S102. Based on the bidirectional gated loop unit in the lithology identification model, extract global temporal features containing contextual dependencies according to the sequential correlation of the multi-scale fusion feature sequence in both the forward and reverse sequence directions; the multi-scale fusion feature sequence includes multiple multi-scale fusion features arranged in chronological order. It should be noted that: the multi-scale fused features obtained from S101 are input into a bidirectional gated recurrent unit, and multiple multi-scale fused features constitute an ordered sequence in the time dimension, namely the multi-scale fused feature sequence. Each element in the multi-scale fused feature sequence... (Where t=1, 2, ..., T are time step indices) represents the multi-scale local features extracted at a specific drilling depth (or time point). The multi-scale fused feature sequence reflects the continuous variation of formation rock properties along the drilling depth direction. By using bidirectional gated cyclic units to perform bidirectional (including forward and reverse sequence directions) deep time series modeling on the multi-scale fused feature sequence, the preceding and following correlations in the bidirectional sequence directions are obtained. Furthermore, the contextual dependencies in the formation depth direction are extracted to obtain depth time series features, especially the correlation features between adjacent formations in the lithological transition zone.

[0030] S103. Based on the global temporal features, the lithology category identification result is obtained.

[0031] It should be noted that: the depth time series characteristics of all time steps generated by S102 are analyzed to obtain the probability distribution of various lithologies (such as sandstone, mudstone, limestone, etc.). The lithology category with the highest probability is taken as the final identification result of the lithology identification model.

[0032] In summary, the drilling lithology identification method provided in this embodiment of the invention first constructs a hierarchical and complementary deep neural network architecture by organically combining a multi-scale convolutional neural network and a bidirectional gated recurrent unit. The multi-scale convolutional neural network extracts lithological features of different granularities in the spatial dimension, while the bidirectional gated recurrent unit models the contextual dependencies of the formation sequence in the temporal dimension. Their collaborative work enables efficient parsing of drilling time-series data, significantly improving the accuracy and real-time performance of lithology identification, and demonstrating good geological adaptability and engineering practicality, providing reliable technical support for intelligent drilling and refined geological guidance. Furthermore, the multi-scale convolutional neural network in S101 simultaneously captures local features of different temporal granularities in the drilling time-series data, comprehensively and accurately mining multi-scale local features related to lithology. Through the bidirectional scanning of the bidirectional gated recurrent unit in S102, sequence dependencies in the depth direction can be established, which is particularly sensitive to formation changes such as lithological transition zones, thus enhancing lithology identification capabilities.

[0033] To improve the accuracy of lithology identification, in some embodiments of the present invention, such as... Figure 2 As shown, before step S101, the following steps are included: S201. Obtain the raw drilling data and calculate the local statistics of the raw drilling data based on a sliding window.

[0034] It should be noted that due to various factors such as high-temperature and high-pressure downhole conditions, sensor aging and failure, drilling equipment vibration interference, signal transmission delay, and electromagnetic interference, raw drilling data inevitably contains outliers, missing values, and noise interference. Directly inputting raw drilling data into the lithology identification model will severely affect feature extraction and identification accuracy. To improve data quality and reliability, and to provide high-quality data support for subsequent lithology identification model training and inference, raw drilling data preprocessing is necessary. Raw drilling data typically includes engineering parameters such as drill pressure, rotational speed, torque, and mechanical drilling rate, which are continuously collected and uploaded by downhole sensors. Since the drilling process is non-stationary, the statistical characteristics of parameters will drift with depth due to factors such as different lithologies encountered and drilling parameter adjustments. Therefore, a sliding window approach is used to perform segmented local dynamic statistics to more accurately identify abnormal data. In specific implementation, a sliding window of fixed length is set (for example, window length W = 100–300 sampling points, the specific window length is determined according to the data sampling frequency and drilling speed). The sliding window moves point-by-point along the time series (corresponding to the depth series). At each sliding window position, local statistics are calculated for all raw drilling data within the sliding window, including the local mean. and local standard deviation .

[0035]

[0036] in, This refers to the sample sequence within the sliding window (i.e., the sequence of original drilling data). This represents the number of samples within the sliding window (i.e., the number of original drilling data).

[0037] S202. Identify and correct abnormal data in the original drilling data based on the local statistics.

[0038] It should be noted that the local mean of each sliding window is calculated based on S201. and local standard deviation Set a threshold for anomaly detection. If a sample value (raw drilling data) within the sliding window... satisfy or Then determine the original drilling data. This identifies anomalous data. This method effectively filters out anomalous data caused by sensor malfunctions, electromagnetic interference, etc., while also avoiding misjudgments due to global statistics. After identifying anomalous data in all raw drilling data within the current sliding window, the corresponding anomalous data in the current sliding window is corrected. This process is repeated, dynamically scrolling to the next sliding window and identifying and correcting anomalous data in all raw drilling data within that window. This continues until all anomalous data in the raw drilling data is corrected, resulting in drilling time-series data arranged chronologically.

[0039] In this embodiment, local statistics through a sliding window can accurately identify and locate noise points introduced by environmental interference and equipment malfunctions, avoiding misjudgments in non-stationary data by global statistics. This provides the model with high-quality, high-fidelity input data, effectively eliminating noise and anomalies in the input data, preventing these interference information from being amplified or learned in subsequent feature extraction, improving the purity and accuracy of multi-scale convolutional neural network feature extraction, and thus enhancing the model's lithology identification accuracy.

[0040] In order to adaptively select the optimal interpolation algorithm based on the different continuity characteristics of abnormal data to balance repair accuracy and computational efficiency, and to ensure the continuity and trend authenticity of the data sequence, in some embodiments of the present invention, step S202 includes: If the number of consecutive abnormal data within the sliding window is less than a first threshold, the abnormal data is corrected using linear interpolation. If the number of consecutive abnormal data within the sliding window is greater than or equal to the first threshold, then cubic spline interpolation is used to correct the abnormal data.

[0041] It should be noted that data preprocessing includes outlier handling and missing value handling. To avoid data sequence breakage caused by directly deleting outliers, after identifying outliers or missing values ​​through a sliding window and local statistics, a hierarchical interpolation strategy is used for adaptive repair based on the number of consecutive outlier data points within the sliding window, achieving an optimal balance between accuracy and efficiency. When the number of consecutive outliers identified within the current sliding window is less than a first threshold (e.g., 5), it is considered an isolated outlier or a short outlier sequence (short-term interference), and linear interpolation is used for replacement and correction. Linear interpolation is based on the assumption of continuity between adjacent normal data, and the calculation formula is:

[0042] in, This indicates the location of the abnormal data. , These are the normal samples adjacent to the anomalous data, respectively. This method can achieve batch outlier correction, meeting the engineering requirements of real-time processing during drilling.

[0043] If the number of consecutive outliers (number of consecutive outliers or missing values) within the current sliding window reaches or exceeds a first threshold (e.g., 5), it is considered an abnormally long sequence, indicating a possible period of data acquisition interruption or continuous interference. In this case, cubic spline interpolation is used for repair. Specifically, the abnormally long sequence (i.e., the outlier or missing segment) is used as the core, and a certain number (usually 10-20 before and after) of normal data are selected at both ends as control nodes for interpolation. A smooth cubic spline curve is constructed based on these control nodes, and this curve is used to interpolate and complete all points within the abnormally long sequence. Cubic spline interpolation maintains the continuity of the first and second derivatives of the curve, thus better fitting the potential nonlinear trends of the data and reducing repair errors caused by long segments of missing data.

[0044] Of course, for missing values ​​caused by data acquisition interruptions, a linear function can be constructed using the valid data before and after the missing segment to fill in the missing samples. Let's assume the valid sample preceding the missing segment is... The corresponding time is The second valid sample is The corresponding time is The location of the missing sample is ,satisfy The linear interpolation completion formula is:

[0045]

[0046] In this embodiment, for abnormal short sequences (short-term interference), linear interpolation is used to correct the abnormal data, eliminating the sharp interference that discrete noise points may cause to subsequent feature extraction. This allows for rapid repair with minimal overhead, ensuring the system's real-time processing capability. For abnormal long sequences (continuous interference), cubic spline interpolation is used to correct the abnormal data, effectively reconstructing the reasonable evolution trajectory of long segments of missing data. This prevents information breaks and model misjudgments caused by large data gaps, ensuring high accuracy of the repair results. This hierarchical strategy achieves an optimal balance between repair accuracy and processing efficiency in the entire preprocessing stage. Furthermore, selectively using linear interpolation and cubic spline interpolation to repair abnormal data restores various types of incomplete and unreliable parts of the original data into continuous, smooth, and physically consistent high-quality drilling time-series data. This significantly improves the overall quality and reliability of the data input to the lithology identification model, thereby increasing the accuracy of the lithology identification model. Furthermore, drilling site conditions are highly variable, and data anomalies take many forms. Selective use of linear interpolation and cubic spline interpolation to repair abnormal data enables the system to adapt to various anomaly modes. Whether it is frequent instantaneous impulse noise or occasional long-term signal loss, the system can effectively tolerate and repair faults, thereby significantly reducing the sensitivity of the original data quality to the final lithology identification results and improving its practicality, stability, and overall robustness in real and harsh drilling environments.

[0047] To comprehensively and efficiently extract local features at different time scales from drilling time-series data, in some embodiments of the present invention, the multi-scale convolutional neural network includes at least three sets of parallel-connected one-dimensional convolutional layers, each set of one-dimensional convolutional layers employing convolutional kernels with different time dimensions, such as... Figure 3 As shown, step S101 includes: S301. The drilling time-series data is input in parallel into the at least three sets of one-dimensional convolutional layers for convolution operation, and local features of different time windows are extracted to obtain corresponding feature maps.

[0048] It should be noted that the dimension of the input drilling time series data is: The drilling time series data format is B represents the batch size, T represents the time step of a single sample, and F represents the feature dimension of each time step (i.e., the number of parameters for drilling and logging). The multi-scale convolutional neural network contains at least three sets of parallel connected one-dimensional convolutional layers. Each set of convolutional layers uses convolutional kernels with different time dimensions. It is preferred to use 3 sets of convolutional kernels to balance feature richness and computational efficiency, and it can also be extended to 4 sets or more (e.g., increasing the kernel size to 9, 11, etc.). For example, by setting 3 sets of convolutional layers, the 3 sets of convolutional layers use convolutional kernels of different sizes. Small-sized convolutional kernels (e.g., 3×1), medium-sized convolutional kernels (e.g., 5×1), and large-sized convolutional kernels (e.g., 7×1) can be set. In addition, the number of channels of the convolutional kernels (i.e., the number of output feature maps) is set to 32, the convolutional stride is uniformly 1, and the "same" padding method is used to keep the output time step T unchanged. The first convolutional kernel is calculated according to the following formula (5). i Feature maps output after convolution operations by convolution kernels :

[0049] in, This corresponds to 3 sets of convolutional kernels, each with 32 channels. Small-sized convolutional kernels (3...) Focusing on capturing fine features within a short time window, such as transient changes, medium-sized convolutional kernels (5 It focuses on extracting associated features within the intermediate time window, such as the periodicity of parameter fluctuations and large-size convolutional kernels. This is used to mine global and local features within a long time window, such as the long-term stable trend of parameters. Each group of convolutions is followed by a batch normalization layer to accelerate convergence and improve the robustness of the lithology identification model to drilling parameter fluctuations. Indicates the first Group convolution kernel weights, Indicates the first Group convolution bias Indicates step size, This indicates that the same filling method is used, where , This represents a 2D convolution. It is calculated according to the following formula (6):

[0050] in, Input data for the convolutional layer, index Indicates batch indexing, subscript For time step index, subscript 1 For input channel index, subscript For output channel index, This is the index for the stride of the convolution kernel. Indicates the first The temporal dimension of the convolutional kernel. Indicates the input feature dimension index. The total number of feature dimensions in the input data. This represents the element value at a specific position in the input data. This represents the weight values ​​of the convolution kernel at a specific location. For the first The bias of each output channel is determined. Drilling time-series data is fed into three parallel one-dimensional convolutional layers simultaneously. Convolution operations are performed using formulas (5) and (6) to capture fine features (i.e., short-term transients) within a short time window, correlation features within a medium time window, and global-local features (i.e., long-term trends) within a long time window, respectively. Each convolutional layer independently extracts local features at its corresponding scale from the input data, outputting three independent feature maps. The shape of each feature map is 32.

[0051] S302. The feature maps output by each group of one-dimensional convolutional layers are sequentially subjected to nonlinear activation and global average pooling.

[0052] It should be noted that the feature maps output by each convolutional layer introduce non-linearity through the ReLU activation function. The ReLU function, by setting negative values ​​to zero, preserves positive feature information while effectively mitigating the gradient vanishing problem, thus improving the stability of the lithology identification model training. After non-linear activation using the ReLU activation function, global average pooling is used to compress the spatial dimension of the feature maps in each channel, transforming the T time-step features of each channel into one global statistical feature. This reduces subsequent computational complexity and increases the robustness of the features. The activation function and global average pooling formulas are as follows:

[0053] In formula (7) Activation function By setting the negative values ​​to zero, nonlinearity is introduced, thus solving the gradient vanishing problem. In formula (8)... Global average pooling means averaging the features of each channel over T time steps to obtain a single scalar feature.

[0054] S303. All processed features are spliced ​​together to form the multi-scale fused features.

[0055] It should be noted that concatenating all processed features (e.g., features after 3 pooling operations) along the channel dimension captures features at a wider time scale, resulting in the multi-scale local features output by MSCNN. The concatenation formula is as follows:

[0056] In this embodiment, by setting convolutional kernels of different sizes in parallel, the lithology identification model can simultaneously scan short, medium, and long time windows of the input data. This allows the model to adaptively capture multi-scale patterns related to different lithologies (such as thin interlayers and thick blocky rock masses), resulting in more comprehensive feature extraction. Furthermore, batch normalization ensures training stability, and ReLU activation introduces necessary nonlinear expressive power. Global average pooling effectively suppresses the interference of random noise and significantly reduces the dimensionality of features by performing global statistics on each channel, thereby improving feature robustness and generalization ability. The final multi-scale fusion feature is a compact and powerful feature representation that integrates multi-source and multi-scale information, laying a solid foundation for the entire lithology identification model. Since different lithological strata often exhibit characteristic combinations at different time scales in drilling data (for example, hard interlayers may cause short-term spikes in drilling pressure, while thick mudstone may exhibit long-term high gamma values), the multi-scale fusion features obtained by splicing can simultaneously provide these key clues, enhance the richness and discriminativeness of feature expression, and improve the accuracy and reliability of lithology identification models in the face of complex strata (such as thin interbedded layers and lithological gradient zones).

[0057] In some embodiments of the present invention, the bidirectional gated loop unit includes a forward gated loop unit and a reverse gated loop unit arranged in parallel and processing in opposite directions, such as... Figure 4 As shown, step S102 includes: S401. The multi-scale fused feature sequence is simultaneously input into the forward gated loop unit and the reverse gated loop unit, wherein the forward gated loop unit processes the multi-scale fused feature sequence in the time-increasing direction, and the reverse gated loop unit processes the multi-scale fused feature sequence in the time-decreasing direction.

[0058] It should be noted that the bidirectional gated loop unit includes two independent GRU units: a forward gated loop unit (i.e., forward GRU) and a reverse gated loop unit (i.e., reverse GRU). The forward GRU processes the multi-scale fused feature sequence along the time-increasing direction (corresponding to the drilling from shallow to deep layers, i.e., the direction from time step 0 to time step T-1) to capture forward temporal dependencies and obtain the forward hidden state sequence. The reverse GRU processes the multi-scale fused feature sequence in the reverse direction (corresponding to the drilling from deep to shallow layers, i.e., the direction from time step T-1 to time step 0) to capture backward temporal dependencies and obtain the reverse hidden state sequence. By fusing bidirectional temporal features, the contextual information of the formation sequence is fully mined.

[0059] S402. Obtain the forward hidden state sequence processed by the forward gated loop unit and the reverse hidden state sequence processed by the reverse gated loop unit, respectively.

[0060] It should be noted that bidirectional GRU units (including forward GRU and reverse GRU) achieve selective forgetting of historical information and selective updating of current information through dynamic adjustment of update and reset gates, effectively solving the gradient vanishing problem of traditional RNNs. Let the hidden layer dimension of the unidirectional GRU be H=64 (this value was obtained through cross-validation parameter tuning and performed best on multiple drilling time-series datasets; the empirical range is 32–256), inputting the multi-scale fusion features output by MSCNN. The core formula is as follows:

[0061] These represent the updated door, reset door, candidate hidden state, and final hidden state, respectively. For time indexing, Indicates the first Input features at time steps Indicates the first The hidden state of the time step. This is the weight matrix. For bias vectors, It is the Sigmoid activation function. The hyperbolic tangent activation function is used. This represents element-wise multiplication. For each time step t, the input of the forward GRU is the feature vector xt of the current step (i.e., ...). ), and combine with the data from the previous time step t 1 (representing a shallower stratum) is in a hidden state. According to the GRU gating mechanism (update gate) shown in formulas (10)-(13), Reset door Perform the calculation and output the positive hidden state at the current time step t. The inverse GRU takes xt as input to the same time step t, but incorporates the hidden state from the next time step t+1 (representing a deeper stratum). Perform the same GRU calculation and output the reverse hidden state at the current time step t. .

[0062] The forward GRU processes the multi-scale fused feature sequence sequentially from time step t=0 to t=T-1, capturing the lithological influence of shallow strata on deep strata. The reverse GRU processes the multi-scale fused feature sequence in reverse from time step t=T-1 to t=0, capturing the geological genetic correlation between deep strata and shallow strata. According to formulas (14) and (15), after the forward and reverse GRUs complete the bidirectional scan of the entire multi-scale fused feature sequence, the forward GRU outputs a forward hidden state sequence. The inverse GRU outputs a reverse hidden state sequence. .

[0063]

[0064] S403. Concatenate the forward hidden state sequence with the reverse hidden state sequence to obtain deep temporal features; S404. Perform global average pooling on the deep temporal features to obtain global temporal features.

[0065] It should be noted that, in order to integrate the most complete context information at each time step t, the forward hidden state sequence corresponding to the same time step t is used. and reverse hidden state sequence The feature sequence is concatenated according to the feature dimension, i.e., according to formula (16) to obtain the final deep temporal feature sequence. To obtain the global temporal features of the entire sequence, the bidirectional fusion features (i.e., the deep temporal feature sequence) are processed according to formula (17). Perform global average pooling along the time dimension to obtain global temporal features:

[0066]

[0067] In this embodiment, through the collaborative work of forward and reverse GRUs, the lithology identification model overcomes the limitation of unidirectional recurrent neural networks, which can only capture single-directional dependencies. This allows the generated deep temporal features to contain the most complete sequence context for each depth point, enabling more accurate and stable identification of lithological interfaces and transitional zones. This significantly enhances the ability to identify complex geological sequences, especially lithological transition zones. Furthermore, through global average pooling, the variable-length deep temporal feature sequences are aggregated into a fixed-length global feature vector, reducing the number of parameters in subsequent fully connected layers, improving computational efficiency, and meeting real-time requirements. Moreover, the pooling operation itself smooths noise, enhancing the robustness of feature representation. This makes the final decision less susceptible to interference from local anomalies in the sequence, reducing the risk of misjudgment due to local data fluctuations or noise. This makes the lithology identification model more robust to local data anomalies and improves the reliability of the overall lithology identification results.

[0068] High-dimensional, long-sequence features derived from deep time-series modeling are efficiently and robustly mapped to intuitive lithology category probability distributions, thereby achieving real-time lithology identification. In some embodiments of this invention, a fully connected layer and an output layer are connected after the bidirectional gated recurrent unit; for example... Figure 5 As shown, step S103 includes: S501. Based on the fully connected layer, the global temporal features are subjected to linear transformation, nonlinear activation processing and random deactivation processing to obtain the processed temporal features. S502. Based on the output layer and the processed temporal characteristics, predict the probability distribution of generated lithology categories to obtain the identification result.

[0069] It should be noted that after multi-scale feature extraction by MSCNN and temporal dependency modeling by BiGRU, the global temporal features... The core features required for lithology identification are already included, but the global temporal features are still in a high-dimensional feature space. They need to be mapped and converted into a probability distribution in the lithology category space before they can be directly used for lithology judgment. The fully connected layer and the output layer together realize the mapping from the feature space to the category space, and output the intuitive identification results through probability normalization. Specifically, the number of neurons in the fully connected layer is 128. The fully connected layer receives the global temporal feature vector as input, and then connects to the ReLU activation function and the Dropout regularization layer to perform linear transformation and nonlinear activation processing. Let the number of lithology categories be C. The calculation of linear transformation and nonlinear activation processing is shown in the following formula (18):

[0070] in, For the weights of the fully connected layer, This is used for biasing the fully connected layer.

[0071] The number of neurons in the output layer is equal to the number of lithology categories, C. The number of lithology categories, C, is determined based on the lithology types in the actual drilling area. In this invention, C ≥ 4, covering common lithologies such as sandstone, mudstone, limestone, and dolomite. The activation function of the output layer is the Softmax function, with Dropout regularization applied to the output layer. As input, the normalized probability distribution is calculated based on the following formula (19), and the class scores output by the fully connected layer are converted into the probability distributions corresponding to each class. This is used to directly determine the most likely lithology category to obtain the identification result, thus completing the final lithology identification task:

[0072] in, For output layer weights, This is the output layer bias.

[0073] In this embodiment, a fully connected layer combined with the ReLU activation function provides a powerful ability to combine nonlinear features, enabling the learning of highly abstract and complex patterns related to lithology identification. The introduced Dropout regularization layer randomly deactivates some neurons during training, improving the model's generalization performance and stability on unseen complex geological data. Finally, the probability distribution output by the output layer and the Softmax function provides an intuitive and quantifiable probability distribution as the identification result for lithology identification. Through aggregation, transformation, and regularization, a stable, reliable, and interpretable lithology identification result is obtained, which can directly and effectively serve real-time decision-making and safe and efficient production at the drilling site.

[0074] The lithology identification model is trained efficiently and stably, enabling it to accurately learn the complex mapping relationship from data to lithology, while avoiding overfitting and ensuring that the model achieves optimal generalization ability. In some embodiments of the present invention, a training step for the lithology identification model is also included: The difference between the identification result and the actual lithology label is calculated using the cross-entropy loss function; With the goal of minimizing the difference, the model parameters of the lithology identification model are iteratively updated using an adaptive moment estimation algorithm. During the training process, the identification accuracy of the validation set is monitored. If the accuracy does not improve for a predetermined number of consecutive rounds, the training is terminated.

[0075] It should be noted that: before training begins, drilling time-series data with labeled real lithology categories are prepared and divided into training and validation sets. In each training round, the model performs forward propagation on a batch of data to obtain the predicted lithology probability distribution (i.e., the identification result). Subsequently, the cross-entropy loss function is used as the optimization objective to quantify the difference between the predicted result and the real label. The cross-entropy loss function is shown in the following formula (20):

[0076] To minimize the aforementioned cross-entropy loss function, this invention employs the Adaptive Moment Estimation (Adam) algorithm as the optimizer to iteratively update all model parameters in the lithology identification model (including the convolutional kernel weights of MSCNN, the weights and biases of BiGRU, and the weights and biases of fully connected layers). An initial learning rate (e.g., η = 0.001) and an exponential decay rate for its internal first-moment and second-moment estimates are set for the Adam optimizer. In each training batch, the gradient of the loss function with respect to the model parameters is calculated. The Adam algorithm uses the first-moment estimate (mean) and second-moment estimate (uncentered variance) of the gradient to calculate an adaptive, independent learning rate for each model parameter. A learning rate decay strategy is employed to further optimize the training process. For example, after each preset number of training rounds (e.g., 200 rounds), the learning rate is multiplied by a decay factor (e.g., 0.9), and the number of training rounds is set to a preset total number of rounds (e.g., 1000 rounds). Furthermore, to prevent the model from overfitting on the training set and causing a decrease in generalization ability, this invention employs an early stopping strategy. That is, after each training round, the current model is evaluated using an independent validation set, and its recognition accuracy is calculated (precision, recall, etc. can also be monitored simultaneously). A consecutive number of rounds is set (e.g., 50 rounds). During training, the best accuracy on the validation set is continuously recorded. If, after 50 consecutive training rounds, the validation set accuracy does not exceed the historical best record, it is determined that the model performance has no longer improved and may have begun to overfit. At this point, the training process is immediately terminated. After training terminates, instead of selecting the model parameters from the last round, the model parameters corresponding to the round with the highest validation set accuracy are rolled back and loaded as the final trained lithology recognition model.

[0077] In this embodiment, the cross-entropy loss function drives the model parameters to update in a direction that maximizes the prediction likelihood (i.e., minimizes the difference), ensuring that the lithology identification model can learn a true and effective mapping relationship between the data and lithology categories. The adaptive moment estimation algorithm (Adam) combines the advantages of momentum and adaptive learning rate, providing personalized update step sizes for different parameters, accelerating the convergence speed of the lithology identification model in the early stages of training, reducing the workload of parameter tuning, and enabling smoother and more direct optimization in the parameter space, effectively avoiding violent oscillations during training and improving the stability of the model training process. In addition, by continuously monitoring the identification accuracy of the validation set, the lithology identification model stops learning at the moment when its generalization ability is optimal according to the early stopping strategy, avoiding the overfitting trap commonly found in models with limited data, and ensuring that the finally deployed model can maintain high identification accuracy and reliability when facing unknown and complex real drilling data. Furthermore, by combining learning rate decay with an early stop strategy, the lithology identification model is trained, reducing excessive reliance on hyperparameters such as the number of training rounds. Ultimately, a high-performance, high-reliability drilling-while-drilling lithology identification model is trained, providing a foundation for real-time intelligent decision-making at the drilling site.

[0078] like Figure 6 As shown, the performance evolution of the lithology identification model of this invention during the training process is dynamically recorded from four core evaluation dimensions: accuracy, precision, F1 score, and recall. Figure 6 (a) Show the accuracy curves of the lithology identification model on the training and validation sets as a function of training epochs. This is used to observe whether the lithology identification model can learn effectively (accuracy on the training set increases) and whether it is overfitting (whether the accuracy on the validation set increases synchronously or stagnates prematurely). Figure 6 (b) Shows the precision curves of the lithology identification model on the training and validation sets as a function of training epochs. Precision is an important metric in lithology identification, and this figure is used to demonstrate the reliability of the lithology identification model in identifying specific lithologies (such as oil and gas reservoirs). Figure 6 (c) Show the F1 score of the lithology identification model on the training and validation sets as a function of training epochs. The F1 score is the harmonic mean of precision and recall, used to comprehensively evaluate the classification performance of the lithology identification model and demonstrate the balanced identification ability of the lithology identification model across various types of lithology. Figure 6 (d) Show the recall curves of the lithology identification model on the training and validation sets as a function of training epochs. Recall measures the ability of the lithology identification model to detect various types of lithology. This figure is used to demonstrate that the lithology identification model is less likely to miss any cases.

[0079] This invention ensures the quality of input data through data preprocessing, achieving high computational efficiency. The lithology identification model boasts a simple structure and fast inference speed, meeting the real-time identification requirements during drilling. Simultaneously, the lithology identification model exhibits strong robustness to drilling parameter fluctuations and data noise, ensuring the stability and reliability of the identification results. MSCNN extracts features in parallel using multiple sets of convolutional kernels of different sizes, effectively covering lithology-related features at different scales, solving the problem of insufficient feature extraction in traditional single-scale convolution, and accurately mining high-order nonlinear features from drilling time-series data. BiGRU captures contextual information of formation sequences from both forward and reverse directions, particularly enhancing the identification capability of lithological transition zones, overcoming the deficiency of existing lithology identification models in capturing formation depth-direction sequence dependencies. Through structural optimization and training strategy adjustments, the lithology identification model outperforms traditional drilling lithology identification technologies and existing single machine learning and deep learning methods in core evaluation metrics such as accuracy, precision, and F1 score. It can adapt to drilling operation scenarios under complex geological conditions, providing strong technical support for the safe and efficient implementation of oil and gas drilling projects.

[0080] To better implement the drilling lithology identification method in the embodiments of the present invention, based on the drilling lithology identification method, correspondingly, as follows: Figure 7 As shown, this embodiment of the invention also provides a drilling lithology identification system 700, which includes: The multi-scale feature extraction module 701 is used to extract and fuse features from drilling time-series data based on a multi-scale convolutional neural network of a lithology identification model to obtain multi-scale fused features. The global temporal extraction module 702 is used to extract global temporal features containing contextual dependencies based on the bidirectional gated loop unit in the lithology identification model and the sequential correlation of the multi-scale fusion feature sequence in the forward and reverse sequence directions; the multi-scale fusion feature sequence includes multiple multi-scale fusion features arranged in chronological order. The classification output module 703 is used to obtain the identification result of lithology category based on the global time series features.

[0081] The drilling lithology identification system 700 provided in the above embodiments can realize the technical solutions described in the above drilling lithology identification method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content of the above drilling lithology identification method embodiments, which will not be repeated here.

[0082] like Figure 8 As shown, the present invention also provides an electronic device 800. The electronic device 800 includes a processor 801, a memory 802, and a display 803. Figure 8Only some components of the electronic device 800 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

[0083] In some embodiments, processor 801 may be a central processing unit (CPU), microprocessor or other data processing chip, used to run program code stored in memory 802 or process data, such as the drilling lithology identification method of the present invention.

[0084] In some embodiments, processor 801 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 801 may be local or remote. In some embodiments, processor 801 may be implemented on a cloud platform. In one embodiment, the cloud platform may include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, intranet, multi-cloud, etc., or any combination thereof.

[0085] In some embodiments, memory 802 may be an internal storage unit of electronic device 800, such as a hard disk or memory of electronic device 800. In other embodiments, memory 802 may also be an external storage device of electronic device 800, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 800.

[0086] Furthermore, the memory 802 may include both internal storage units of the electronic device 800 and external storage devices. The memory 802 is used to store application software and various types of data installed on the electronic device 800.

[0087] In some embodiments, display 803 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 803 is used to display information from electronic device 800 and to display a visual user interface. Components 801-803 of electronic device 800 communicate with each other via a system bus.

[0088] In one embodiment, when the processor 801 executes the drilling lithology identification program in the memory 802, the following steps can be implemented: A multi-scale convolutional neural network based on a lithology identification model is used to extract and fuse features from drilling time-series data to obtain multi-scale fused features. Based on the bidirectional gated recurrent unit in the lithology identification model, global temporal features containing contextual dependencies are extracted according to the sequential correlation of the multi-scale fusion feature sequence in both the forward and reverse sequence directions; the multi-scale fusion feature sequence includes multiple multi-scale fusion features arranged in chronological order. The lithology category identification result is obtained based on the global temporal features.

[0089] It should be understood that when the processor 801 executes the drilling lithology identification program in the memory 802, in addition to the functions mentioned above, it can also perform other functions, as detailed in the description of the corresponding method embodiments above.

[0090] Furthermore, this embodiment of the invention does not specifically limit the type of electronic device 800 mentioned. Electronic device 800 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the invention, electronic device 800 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0091] Accordingly, this application also provides a computer-readable storage medium for storing computer-readable programs or instructions. When the programs or instructions are executed by a processor, they can implement the steps or functions of the drilling lithology identification methods provided in the above-described method embodiments.

[0092] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0093] The above provides a detailed description of the drilling lithology identification method, system, equipment, and medium provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for identifying lithology while drilling, characterized in that, include: A multi-scale convolutional neural network based on a lithology identification model is used to extract and fuse features from drilling time-series data to obtain multi-scale fused features. Based on the bidirectional gated recurrent unit in the lithology identification model, global temporal features containing contextual dependencies are extracted according to the sequential correlation of the multi-scale fusion feature sequence in both the forward and reverse sequence directions; the multi-scale fusion feature sequence includes multiple multi-scale fusion features arranged in chronological order. The lithology category identification result is obtained based on the global temporal features.

2. The drilling lithology identification method according to claim 1, characterized in that, Before the multi-scale convolutional neural network based on the lithology identification model extracts and fuses features from drilling time-series data to obtain multi-scale fused features, it includes: Obtain raw drilling data and calculate local statistics of the raw drilling data based on a sliding window; Abnormal data in the original drilling data are identified and corrected based on the local statistics.

3. The drilling lithology identification method according to claim 2, characterized in that, The step of identifying and correcting anomalous data in the original drilling data based on the local statistics includes: If the number of consecutive abnormal data within the sliding window is less than a first threshold, the abnormal data is corrected using linear interpolation. If the number of consecutive abnormal data within the sliding window is greater than or equal to the first threshold, then cubic spline interpolation is used to correct the abnormal data.

4. The drilling lithology identification method according to claim 1, characterized in that, The multi-scale convolutional neural network includes at least three sets of parallel-connected one-dimensional convolutional layers, each set of one-dimensional convolutional layers employing convolutional kernels with different time dimensions. The multi-scale convolutional neural network based on the lithology identification model extracts and fuses features from drilling time-series data to obtain multi-scale fused features, including: The drilling time-series data is input in parallel into the at least three sets of one-dimensional convolutional layers for convolution operations, and local features of different time windows are extracted to obtain corresponding feature maps; The feature maps output by each group of one-dimensional convolutional layers are sequentially subjected to non-linear activation and global average pooling. All processed features are spliced ​​together to form the multi-scale fused feature.

5. The drilling lithology identification method according to claim 1, characterized in that, The bidirectional gated loop unit includes a forward gated loop unit and a reverse gated loop unit arranged in parallel and processing in opposite directions. The bidirectional gated loop unit based on the lithology identification model extracts global temporal features containing contextual dependencies based on the sequential correlation of the multi-scale fused feature sequences in the forward and reverse sequence directions, including: The multi-scale fused feature sequence is simultaneously input into the forward gated loop unit and the reverse gated loop unit, wherein the forward gated loop unit processes the multi-scale fused feature sequence in the time-increasing direction, and the reverse gated loop unit processes the multi-scale fused feature sequence in the time-decreasing direction. The forward hidden state sequence obtained by the forward gated loop unit and the reverse hidden state sequence obtained by the reverse gated loop unit are obtained respectively. The forward hidden state sequence and the reverse hidden state sequence are concatenated to obtain deep temporal features; The global time-series features are obtained by performing global average pooling on the deep temporal features.

6. The drilling lithology identification method according to claim 1, characterized in that, The bidirectional gated recurrent unit is followed by a fully connected layer and an output layer; the lithology category identification result obtained based on the global temporal features includes: Based on the fully connected layer, the global temporal features are subjected to linear transformation, nonlinear activation processing, and random deactivation processing to obtain the processed temporal features. Based on the output layer and the processed temporal features, the probability distribution of generated lithology categories is predicted to obtain the identification result.

7. The drilling lithology identification method according to claim 1, characterized in that, It also includes the training steps of the lithology identification model: The difference between the identification result and the actual lithology label is calculated using the cross-entropy loss function; With the goal of minimizing the difference, the model parameters of the lithology identification model are iteratively updated using an adaptive moment estimation algorithm. During the training process, the identification accuracy of the validation set is monitored. If the accuracy does not improve for a predetermined number of consecutive rounds, the training is terminated.

8. A drilling lithology identification system, characterized in that, include: The multi-scale feature extraction module is used to extract and fuse features from drilling time-series data based on a multi-scale convolutional neural network of a lithology identification model to obtain multi-scale fused features. The global temporal extraction module is used to extract global temporal features containing contextual dependencies based on the bidirectional gated loop unit in the lithology identification model and the sequential correlation of the multi-scale fusion feature sequence in both forward and reverse sequence directions; the multi-scale fusion feature sequence includes multiple multi-scale fusion features arranged in chronological order. The classification output module is used to obtain the identification result of lithology category based on the global time series features.

9. An electronic device, characterized in that, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the drilling lithology identification method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in the drilling lithology identification method according to any one of claims 1 to 7.

Images