Method for improving geosteering traversal rate based on deep learning

Abstract

The invention relates to the field of geological exploration, and discloses a method for improving geosteering traversal rate based on deep learning, comprising: performing stratum occurrence analysis on a target work area to obtain a first analysis result, and predicting a trajectory trend of a drilling horizontal section based on the result; geological feature data and historical while-drilling curve data in the target work area are obtained, and geological attributes of a target stratum are analyzed according to the data; constructing a deep learning algorithm learning library based on historical while-drilling curve data; based on a deep learning algorithm learning library, constructing a small-layer automatic identification model through feature optimization and algorithm training; preprocessing the forward drilling while-drilling curve data, and inputting the data into the small-layer automatic identification model to obtain small-layer probability distribution; on the basis of the small layer probability distribution, obtaining actual geological small layer classification of the position where the drill bit is located through data inverse coding; and generating a trajectory regulation and control instruction according to the actual geological small layer classification so as to improve the geosteering penetration rate of the high-quality reservoir of the horizontal well.

Description

Technical Field

[0001] This manual relates to the field of geological exploration, and in particular to a method for improving geological guidance crossing efficiency based on deep learning. Background Technology

[0002] Shale gas reservoirs are integrated source-reservoir type gas reservoirs, and horizontal wells are an effective means to increase the venting area of ​​shale gas reservoirs. Improving the penetration rate of high-quality reservoirs in the horizontal section of horizontal wells is key to high production in shale gas wells. Traditional directional drilling methods mainly rely on geological engineers to manually identify and compare small layers based on the overall shape of the gamma-ray curve during drilling, but this method has inherent defects.

[0003] On the one hand, this method struggles to perform real-time and effective feature extraction and pattern recognition on high-frequency acquired drilling data; on the other hand, under conditions where high-quality reservoirs are thin and instrument measurement blind spots exist, this method lacks the ability to accurately perceive micro-structures. These technical deficiencies lead to serious lag and misjudgment risks in horizontal well trajectory control. For example, when encountering minor anticlines or synclines during drilling, the method fails to promptly identify the changing trends of small layers, causing the trajectory to deviate from high-quality reservoirs; or, during periods of ambiguity in instrument response, the reliance on manual judgment results in missing the optimal adjustment opportunity.

[0004] Therefore, there is an urgent need for a method based on deep learning to improve the geological steering penetration rate, which can solve the core problem of precise control of horizontal well trajectory in thin reservoirs and micro-structural areas, while effectively improving the processing efficiency and identification accuracy of drilling data. Summary of the Invention

[0005] In view of this, the present invention aims to propose a method based on deep learning to improve the geological steering penetration rate, so as to solve the problem of low penetration rate of high-quality reservoirs in the current horizontal well steering process due to the lag in manual identification and the slow response of thin reservoirs and micro-structures.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for improving geological guidance crossing rate based on deep learning, the method comprising: S1. Conduct stratigraphic occurrence analysis on the target work area to obtain the first analysis result, and artificially predict the trajectory trend of the horizontal drilling section based on the first analysis result.

[0007] S2. Obtain geological feature data and historical drilling curve data within the target work area. Analyze the geological attributes of the target strata based on the geological feature data and historical drilling curve data to obtain a second analysis result. The second analysis result includes geological layer labels and multi-source input features.

[0008] S3. Based on the second analysis results, construct a deep learning algorithm learning library; the deep learning algorithm learning library consists of multi-source input features and geological layer labels.

[0009] S4. Based on the deep learning algorithm learning library, a small-layer automatic recognition model is constructed through feature optimization and algorithm training; the small-layer automatic recognition model is a BP neural network model trained to a prediction accuracy that meets a preset threshold.

[0010] S5. Preprocess the drilling curve data and input the preprocessed drilling curve data into the small layer automatic identification model to obtain the small layer probability distribution.

[0011] S6. Based on the probability distribution of the sub-layers, the actual geological sub-layer classification at the location of the drill bit is obtained through data inverse encoding.

[0012] S7. Based on the actual geological layer classification of the drill bit location, generate trajectory control instructions to improve the geological guidance penetration rate of high-quality reservoirs in horizontal wells.

[0013] The beneficial effects of this solution are as follows: In existing technologies, horizontal well geological steering generally relies on manual experience for small-layer identification and decision-making, which cannot meet the real-time and accuracy requirements of thin reservoirs and micro-structural areas. This application, by constructing a deep learning model to identify and classify features of real-time drilling curves and combining it with trajectory control commands, achieves a technological upgrade from manual, lagging judgment to real-time algorithmic early warning. This effectively solves the problem of trajectory loss of control caused by untimely and inaccurate identification, and significantly improves the penetration rate of high-quality reservoirs.

[0014] Furthermore, step S1 also includes: S110. Obtain seismic interpretation stratigraphic data within the work area and single-well stratigraphic data of the study area, and perform correction processing on the two types of data to obtain corrected data.

[0015] S120. Based on the correction data, an initial three-dimensional geological model is established using a spatial modeling algorithm.

[0016] S130. Based on the initial three-dimensional geological model, the first analysis result is obtained through analysis.

[0017] S140. Based on the first analysis result, the initial three-dimensional geological model is sliced ​​to generate a geological-guided two-dimensional model.

[0018] S150. Acquire the drilling-while-drilling curve data, and compare the drilling-while-drilling curve data with the corresponding predicted curve in the geological steering 2D model in real time to obtain the real-time comparison result; and update the 3D geological model based on the real-time comparison result. By establishing an initial 3D geological model and generating a geological steering 2D model, an accurate geological navigation benchmark is provided for the drilling process; by comparing the drilling data with the model prediction data in real time and dynamically updating the 3D geological model, a self-optimizing geological steering mechanism is formed, which significantly improves the control accuracy of the trajectory of horizontal wells in thin reservoirs.

[0019] Furthermore, the multi-source input features include drilling time, total hydrocarbons, methane, far-end gamma, near-bit average gamma, upward gamma value, downward gamma value, and lithological evolution indicators. By integrating engineering, chemical, and physical parameters, a multi-dimensional feature space is constructed, which can comprehensively characterize the differentiated response characteristics of different geological layers, providing high-information-density input for deep learning models and fundamentally ensuring the accuracy and reliability of the automatic layer identification model.

[0020] Furthermore, step S3 also includes: S310. Perform data cleaning on the second analysis result and extract the multi-source input features.

[0021] S320. Based on the second analysis results after cleaning, manually label the geological sub-layers.

[0022] S330. Based on the multi-source input features and the geological layer labels, a deep learning algorithm learning library is constructed through feature label matching. Through systematic data cleaning, manual annotation, and feature label matching, a high-quality deep learning algorithm learning library is built, providing a reliable data foundation for model training.

[0023] Furthermore, step S4 also includes: S410. Analyze the correlation between the multi-source input features and the geological layer labels based on the Spearman correlation coefficient method to form a correlation heatmap.

[0024] S420. Based on the correlation heatmap, determine the multi-source input features with a correlation greater than a preset threshold to obtain the preferred input features.

[0025] S430. Perform range standardization on the preferred input features, encode the geological layer labels, and construct standardized input and output matrices to obtain the model input layer and model output layer of the automatic layer recognition model.

[0026] S440. Based on the model input layer and the model output layer, a model intermediate layer is set; the model intermediate layer includes the number of intermediate layers, the number of hidden neurons, and the activation function.

[0027] S450. Based on the model input layer, the model output layer, and the model intermediate layer, an initial small-layer automatic recognition model is constructed.

[0028] S460. The initial automatic layer identification model is trained and iterated to obtain a trained automatic layer identification model. The model input is optimized through feature selection and standardization, and a high-precision identification model is constructed by combining a multi-layer network structure and iterative training. This realizes intelligent mapping from massive drilling data to geological layers, significantly improving the accuracy and timeliness of layer identification under thin reservoir conditions.

[0029] Furthermore, step S460 also includes: S461. Initialize the initial small-layer automatic recognition model, randomly assign initial weights, and obtain the prediction result of the first model. S462. Based on the prediction results of the first model and the geological layer labels, calculate the training error of the initial layer automatic identification model using a preset loss function; S463. Based on the training error, the initial small-layer automatic identification model is adjusted in reverse weights, and the trained small-layer automatic identification model is obtained through iterative optimization. The automatic optimization and iteration of model parameters is achieved through the error backpropagation mechanism, enabling the small-layer automatic identification model to have adaptive learning capabilities, effectively improving the fitting accuracy and generalization performance for complex geological features.

[0030] Furthermore, step S7 also includes: S710. Based on the actual geological layer classification of the drill bit's location, determine the relative position of the drill bit in the target window and generate a trajectory control command; the target window is a three-dimensional spatial target body preset based on the location of a high-quality reservoir.

[0031] in: S711. If the actual geological layer classification shows that the drill bit is located below the target window, a descent command is generated.

[0032] S712. If the actual geological layer classification shows that the drill bit is located in the middle of the target window, then a stabilization command is generated.

[0033] S713. If the actual geological layer classification shows that the drill bit is located above the target window, an azimuth increase command is generated. By establishing a real-time mapping relationship between geological layer classification and trajectory control commands, adaptive and precise control of the drill bit trajectory within the target window is achieved, effectively overcoming the trajectory deviation problem caused by delays in manual judgment in traditional methods.

[0034] Furthermore, the horizontal slope adjustment angle does not exceed 1.5° / 10m. By limiting the horizontal slope adjustment angle to within the range of 1.5° / 10m, it ensures that the trajectory adjustment has sufficient response sensitivity to thin reservoirs, and avoids drastic fluctuations in the wellbore trajectory caused by excessive build-up rate, thereby achieving stable and precise drilling through high-quality reservoirs. Attached Figure Description

[0035] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein: Figure 1 This is an exemplary flowchart of a method for improving geological guidance crossing rate based on deep learning according to the present invention; Figure 2 This is an exemplary flowchart for predicting the trajectory trend of the horizontal segment of a well. Figure 3 This is an exemplary flowchart for building a deep learning algorithm learning library; Figure 4 This is an exemplary flowchart for training a small-layer automatic recognition model; Figure 5 This is an exemplary flowchart of the training and iteration process of the initial small-layer automatic recognition model; Figure 6 This is an exemplary flowchart for generating trajectory control instructions; Figure 7 This is a Spearman correlation heatmap; Figure 8 This is a schematic diagram of intelligent stratified geological guidance. Detailed Implementation

[0036] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some examples or embodiments of this specification. For those skilled in the art, these drawings can be applied to other similar scenarios without creative effort. Unless obvious from the context or otherwise specified, the same reference numerals in the drawings represent the same structures or operations.

[0037] As indicated in this specification and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of expressly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0038] Flowcharts are used in this specification to illustrate the operations performed by the system according to embodiments of this specification. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0039] The following detailed explanation illustrates the specific implementation methods: Example 1: Figure 1 This is an exemplary flowchart of a method for improving geological guidance crossing rate based on deep learning according to the present invention.

[0040] In this embodiment, as Figure 1 As shown, the process includes the following steps. In this embodiment, the process can be executed by a processor.

[0041] Step S1: Perform stratigraphic occurrence analysis on the target work area to obtain the first analysis result, and manually predict the trajectory trend of the horizontal drilling section based on the first analysis result.

[0042] Stratigraphic attitude refers to the orientation of shale layers in three-dimensional space, determined by three elements: strike, dip, and dip angle. Strike is the direction of the line of intersection between the shale bedding plane and the horizontal plane; dip is the actual direction of the shale layer's inclination, perpendicular to the strike line and pointing downwards; and dip angle is the maximum angle between the shale bedding plane and the horizontal plane.

[0043] For example, if the shale layer in the target work area has a strike of 60°, a dip of 150°, and a dip angle of 25°, it indicates that the stratum extends at 60° in a northeast-southwest direction and dips steadily at a slope of 25° towards the southeast at 150°.

[0044] The target work area refers to a specific geographical area designated for shale gas exploration and development.

[0045] In this embodiment, the processor can establish a spatial model by performing stratigraphic occurrence analysis on the target work area; and then obtain analysis results by analyzing the spatial model to predict the trajectory trend of the drilling horizontal section.

[0046] Preferably, such as Figure 2 As shown, step S1 further includes: Step S110: Obtain seismic interpretation stratigraphic data within the target work area and single-well stratigraphic data of the study area, and perform correction processing on the two types of data to obtain corrected data.

[0047] Seismic interpretation stratigraphic data are used in geological exploration to determine geological structures and lithological distribution. Examples include structural maps of the top surface of the target layer, isopyrographs, and fault plane distribution maps.

[0048] Seismic interpretation stratigraphic data can be obtained in various ways. For example, it can be generated by processing seismic wave signals collected in the field and then tracing and plotting them in specialized software by professionals based on the characteristics of the phase axis of the reflected waves.

[0049] The single-well stratification data in the study area refers to the results of accurately dividing and marking the geological strata of the drilling profile using well logging, core data, and other wellbore data. Examples include the depth of the top and bottom interfaces of the target layer, the thickness of sublayers, lithology transition surfaces, and the top depth of marker layers.

[0050] Stratified data from a single well in the study area can be obtained in various ways. For example, based on the abrupt changes in the morphology of the integrated logging curves and core logging data, the precise depths of the interfaces between different lithologies can be determined on the single well profile through manual comparative analysis.

[0051] Step S120: Based on the correction data, an initial three-dimensional geological model is established using a spatial modeling algorithm.

[0052] The initial three-dimensional geological model refers to a digital three-dimensional structural model that can characterize the spatial morphology and structural features of the strata interface in the target work area.

[0053] Based on the aforementioned correction data, this embodiment selects the Kriging space interpolation algorithm to establish an initial three-dimensional geological model. Under the constraint of known drilling stratification points, surface fitting is performed on the formation interfaces to ultimately form a three-dimensional geological model that includes formation attitude, faults, and structural features. This model can realize the query and three-dimensional visualization of formation attitude at any location, providing a geological basis for horizontal well trajectory design.

[0054] Step S130: Based on the initial three-dimensional geological model, the first analysis result is obtained through analysis.

[0055] The first analysis result is a comprehensive geological understanding result used to guide the design of horizontal well trajectories. In this embodiment, the first analysis result can be obtained manually based on experience.

[0056] Step S140: Based on the first analysis result, slice the initial three-dimensional geological model to generate a geological-guided two-dimensional model; The geological-guided 2D model is a simplified profile model generated by directional cutting and projection along the expected path of the preset horizontal well trajectory based on the initial 3D geological model.

[0057] In this embodiment, professionals manually predict and design a horizontal well trajectory based on the first analysis results. By loading this trajectory as a spatial cutting line into the initial three-dimensional geological model, vertical cross-sectional slices are made along the trajectory to generate a geological guidance two-dimensional model.

[0058] The horizontal well trajectory planning is the ideal geometric path for the drill bit to enter from the surface entry point, pass through the build-up section, and finally enter and travel horizontally through the target high-quality reservoir.

[0059] Step S150: Obtain the drilling-while-drilling curve data, compare the drilling-while-drilling curve data with the corresponding predicted curve in the geological steering two-dimensional model in real time to obtain the real-time comparison result, and update the three-dimensional geological model based on the real-time comparison result.

[0060] The drilling while-drilling curve data is data that is collected in real time by the drilling measurement tool and transmitted to the surface during the current drilling operation.

[0061] The predicted curve in the geological steering two-dimensional model refers to the theoretical response curve that reflects the expected formation properties at different depths, which is pre-derived on the two-dimensional profile after slicing along the designed wellbore trajectory based on the initial three-dimensional geological model.

[0062] In this embodiment, the drilling-while-drilling curve data is aligned with the predicted curve at the same location in the geological steering 2D model according to the depth coordinates. The morphological difference between the measured and predicted curves is calculated using a dynamic time warping algorithm or point-by-point interpolation analysis. When a systematic deviation is detected, the deviation value is used as a feedback signal, and the spatial coordinates of the stratigraphic interface in the corresponding area of ​​the 3D geological model are corrected in reverse using a Kriging interpolation algorithm, thereby achieving dynamic calibration of the geological model. Systematic deviations include overall curve morphological shifts or differences in the depth of characteristic strata.

[0063] In this embodiment, a three-dimensional geological model integrating well and seismic data is constructed to provide a scientific basis for predicting the trajectory trend of horizontal wells. A precise geological navigation benchmark is established by generating a two-dimensional geological guidance model. By combining real-time data with the dynamic comparison of model predictions and the model's self-updating mechanism, a closed-loop guidance system with continuous optimization capabilities is formed.

[0064] Step S2: Obtain geological feature data and historical drilling curve data within the target work area. Analyze the geological attributes of the target strata based on the geological feature data and historical drilling curve data to obtain a second analysis result. The second analysis result includes geological layer labels and multi-source input features.

[0065] Geological characteristic data are multi-source information collections used to characterize the lithology, physical properties, and structural attributes of underground strata. Examples include structural morphology data obtained through exploration, reservoir physical parameters analyzed based on core and well logging data, and lithofacies distribution characteristics obtained through geological modeling.

[0066] Historical drilling curve data refers to physical parameter data that reflects the characteristics of the drilled formation, collected and recorded by drilling measurement tools during the drilling operations completed in the target work area.

[0067] In this embodiment, the processor can obtain geological feature data within the target work area by integrating seismic interpretation results, core experimental analysis results, and well logging interpretation results of drilled wells; and obtain historical drilling curve data by extracting the database of adjacent drilled wells within the target work area.

[0068] The target formation refers to the thin, high-quality reservoir that needs to be precisely encountered and traversed for a long period during the horizontal well directional drilling process.

[0069] The second analysis result is a standardized dataset used to build a deep learning algorithm learning library, formed by comprehensively analyzing geological feature data and historical drilling curve data. In this embodiment, the second analysis result includes geological layer labels and multi-source input features.

[0070] In this embodiment, the geological layer label includes middle, superior, Down, middle, The five tags are: drilling time, total hydrocarbons, methane, far-end gamma, near-bit average gamma, upward gamma value, downward gamma value, and lithological evolution index.

[0071] Step S3: Based on the second analysis results, construct a deep learning algorithm learning library; this deep learning algorithm learning library consists of multi-source input features and geological layer labels.

[0072] Preferably, such as Figure 3 As shown, step S3 further includes: Step S310: Clean the data of the second analysis results and extract multi-source input features.

[0073] The purpose of data cleaning is to remove invalid data, correct outliers, and standardize data benchmarks. In this embodiment, data cleaning may include outlier removal and missing value handling.

[0074] Outlier removal involves identifying and removing values ​​that are outside the range of physical meaning. For example, removing negative gamma values ​​or invalid records with zero values ​​in the drilling time curve.

[0075] Missing value handling addresses transiently missing segments in data records. In this embodiment, the processor uses linear interpolation to fill in these missing segments, and for continuous missing segments that cannot be effectively repaired, the entire segment is deleted.

[0076] In this embodiment, after the processor cleans the second analysis result, it directly extracts the required eight multi-source input features from the cleaned data.

[0077] Step S320: Based on the second analysis results after cleaning, manually label the geological sub-layers; Step S330: Based on multi-source input features and geological layer labels, a deep learning algorithm learning library is constructed through feature label matching.

[0078] In this embodiment, the training samples of the deep learning algorithm learning library consist of 8 multi-source input features, and the corresponding output labels are 5 geological sub-layer labels corresponding to the training samples. Specifically, the processor can construct a sample set based on 2500 sets of drilling-while-drilling curve data from drilled horizontal wells, wherein the first 2000 sets are the training dataset and the last 500 sets are the validation dataset.

[0079] The training dataset includes several training samples and their corresponding output labels. Multiple iterations are performed, with at least one iteration comprising: selecting one or more training samples from the training dataset; inputting these training samples into the deep learning algorithm learning library to obtain predicted outputs for the samples; substituting the predicted outputs and their corresponding output labels into a predefined loss function formula to obtain the loss function value; and updating the hyperparameters in the deep learning algorithm learning library in reverse order based on the loss function value. This step can be performed using various methods, such as gradient descent. The iteration ends when the termination condition is met, resulting in the trained deep learning algorithm learning library.

[0080] In this embodiment, a high-quality deep learning algorithm learning library was constructed through systematic data cleaning, feature extraction and manual annotation. This library accurately associates multi-source input features with geological layer labels, providing a structured standard dataset for model training and fundamentally ensuring the accuracy of the automatic layer identification model.

[0081] Step S4: Based on the deep learning algorithm learning library, a small-layer automatic recognition model is constructed through feature optimization and algorithm training; the small-layer automatic recognition model is a BP neural network model trained to a preset threshold for prediction accuracy.

[0082] The automatic sub-layer identification model is used to identify the geological sub-layer category at the drill bit's location in real time. In this embodiment, the automatic sub-layer identification model is a machine learning model. For example, the automatic sub-layer identification model may include any one or a combination of a BP neural network model or other custom model structures.

[0083] Preferably, such as Figure 4 As shown, step S4 further includes: Step S410: Analyze the correlation between the multi-source input features and the geological layer labels based on the Spearman correlation coefficient method to form a correlation heatmap.

[0084] In this embodiment, the deep learning algorithm learning library consists of 8 multi-source input features and 5 geological layer labels. The processor can analyze the correlation between the multi-source input features and the geological layer labels based on the Spearman correlation coefficient method. By calculating the rank correlation coefficient between each feature and the label and quantifying its monotonic correlation strength, a learning library is formed. Figure 7 The correlation heatmap shown enables objective screening of key features for predicting geological layer classification.

[0085] In the correlation heatmap, the color intensity of each cell represents the magnitude of the Spearman correlation coefficient between the two corresponding variables; warm colors indicate a strong positive correlation, and cool colors indicate a strong negative correlation. The color intensity of the color block is directly proportional to the absolute value of the correlation coefficient. Based on this, the core feature combination most relevant to the geological layer discrimination can be identified.

[0086] Step S420: Based on the correlation heatmap, determine the multi-source input features with correlation greater than a preset threshold to obtain the preferred input features.

[0087] The preferred input features refer to the subset of core features that are significantly related to the geological layer labels and are selected through correlation analysis.

[0088] The preset threshold can be manually preset based on experience, and can be represented by a specific value of the Spearman correlation coefficient (such as 0.3, 0.5, etc.). When the feature correlation coefficient is greater than the preset threshold, it is considered to be significantly related to the geological layer label and is retained.

[0089] In this embodiment, a preset threshold is set to 0.07, and multi-source input features with a feature correlation coefficient greater than this preset threshold are selected as preferred input features, such as... Figure 7 As shown, this represents the near-bit average gamma, far-bit gamma, total hydrocarbons, methane, and the difference between the upper and lower gamma values.

[0090] Step S430: Perform range standardization on the selected input features, encode the geological layer labels, and construct standardized input and output matrices to obtain the model input layer and model output layer of the automatic layer recognition model.

[0091] The input layer is the first layer of the BP neural network model, responsible for receiving raw data provided externally. This raw data can be images, text, audio signals, or other machine-readable data formats.

[0092] The output layer is the last layer of the BP neural network model, responsible for converting the information passed from the intermediate layers into the final output format, such as classification labels or continuous numerical values. In this embodiment, the output layer outputs the probability distribution of different actual geological sub-layers.

[0093] Step S440: Based on the model input layer and the model output layer, set the model intermediate layer; the model intermediate layer includes the number of intermediate layers, the number of hidden neurons, and the activation function.

[0094] The intermediate layers of a model are one or more hidden layers located between the model input layer and the model output layer. Each layer contains several neurons that are connected to the output of the previous layer in a weighted manner.

[0095] In this embodiment, the model has two intermediate layers and six hidden neurons, with the sigmoid function as the activation function.

[0096] Step S450: Construct an initial small-layer automatic recognition model based on the model input layer, model output layer, and model intermediate layer.

[0097] Step S460: Train and iterate the initial small-layer automatic recognition model to obtain the trained small-layer automatic recognition model.

[0098] Preferably, such as Figure 5 As shown, step S460 further includes: Step S461: Initialize the initial small-layer automatic recognition model, randomly assign initial weights, and obtain the prediction result of the first model.

[0099] Step S462: Based on the prediction results of the first model and the geological layer labels, calculate the training error of the initial layer automatic identification model using a preset loss function.

[0100] Step S463: Based on the training error, the initial small-layer automatic recognition model is adjusted in reverse weights, and the trained small-layer automatic recognition model is obtained through iterative optimization.

[0101] The model training is complete when the loss function of the initial small-layer automatic recognition model meets preset conditions, resulting in a trained small-layer automatic recognition model. These preset conditions can include loss function convergence, the number of iterations reaching a threshold, etc.

[0102] In this embodiment, the processor constructs a BP neural network model by selecting features that are strongly correlated with geological sub-layers. After training, the model can identify geological sub-layers in real time with high accuracy based on drilling data, effectively improving the penetration rate of high-quality reservoirs in horizontal wells.

[0103] Step S5: Preprocess the drilling curve data and input the preprocessed drilling curve data into the small layer automatic identification model to obtain the small layer probability distribution.

[0104] The preprocessing refers to processing the drilling curve data according to the same standards and methods as the training stage, so that it conforms to the format and specifications of the pre-trained small-layer automatic recognition model for the model input layer.

[0105] The sublayer probability distribution refers to a multi-dimensional vector output by the sublayer automatic identification model for the current drill bit position. Each dimension's value represents the predicted probability that the drill bit position belongs to a specific geological sublayer category; the sum of all probability values ​​is 1.

[0106] Step S6: Based on the probability distribution of the sublayer, the actual geological sublayer classification at the drill bit location is obtained through data inverse encoding.

[0107] Step S7: Based on the actual geological layer classification at the drill bit location, generate trajectory control instructions to improve the geological guidance penetration rate of high-quality reservoirs in horizontal wells.

[0108] The trajectory control command is a specific control command generated based on the identification results of the small-layer automatic identification model, used to adjust the drilling trajectory. By adjusting the actuator of the downhole rotary steerable tool, precise, meter-level real-time control of the drilling trajectory is achieved.

[0109] Furthermore, such as Figure 6 As shown, step S7 further includes: Step S710: Based on the actual geological layer classification of the drill bit location, determine the relative position of the drill bit in the target window and generate trajectory control instructions; the target window is a three-dimensional spatial target body preset based on the location of high-quality reservoirs.

[0110] In this embodiment, as Figure 8 As shown, the processor can use the recognition results of the small layer automatic recognition model to adjust the drilling trajectory in real time to ensure a high penetration rate in high-quality reservoirs.

[0111] For example, if the actual geological sub-layer classification shows that the drill bit is located at the lower part of the target window, a deflection command is generated; if the actual geological sub-layer classification shows that the drill bit is located at the middle of the target window, a stabilization command is generated; if the actual geological sub-layer classification shows that the drill bit is located at the upper part of the target window, a deflection command is generated. Furthermore, the slope adjustment angle for the horizontal section does not exceed 1.5° / 10m.

[0112] In this embodiment, by intelligently comparing the real-time identified geological layer classification results with the preset target window position, corresponding commands for deflection reduction, stabilization, or deflection increase are automatically generated, realizing adaptive and precise control of the drilling trajectory in high-quality reservoirs, thereby significantly improving the geological steering penetration rate of horizontal wells.

[0113] The basic concepts have been described above. Obviously, for those skilled in the art, the detailed disclosure above is merely illustrative and does not constitute a limitation of this specification. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and corrections to this specification. Such modifications, improvements, and corrections are suggested in this specification and therefore remain within the spirit and scope of the exemplary embodiments described herein.

[0114] Furthermore, unless expressly stated in the claims, the order of processing elements and sequences, the use of numbers and letters, or other names described in this specification are not intended to limit the order of the processes and methods described herein. Although various examples have been discussed in the foregoing disclosure of some embodiments of the invention that are currently considered useful, it should be understood that such details are for illustrative purposes only, and the appended claims are not limited to the disclosed embodiments; rather, the claims are intended to cover all modifications and equivalent combinations that conform to the spirit and scope of the embodiments described herein. For example, while the system components described above can be implemented using hardware devices, they can also be implemented solely using software solutions, such as installing the described system on existing servers or mobile devices.

[0115] Similarly, it should be noted that, in order to simplify the description disclosed herein and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of embodiments in this specification may sometimes combine multiple features into a single embodiment, drawing, or description thereof. However, this method of disclosure does not imply that the subject matter of this specification requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of a single embodiment disclosed above.

[0116] In some embodiments, numbers describing the quantity of components and attributes are used. It should be understood that such numbers used in the description of embodiments are modified in some examples with the terms "approximately," "approximately," or "generally." Unless otherwise stated, "approximately," "approximately," or "generally" indicates that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may be changed depending on the characteristics required by individual embodiments. In some embodiments, numerical parameters should take into account specified significant digits and employ a general method of digit reservation. Although the numerical ranges and parameters used to confirm their breadth of range in some embodiments of this specification are approximate values, in specific embodiments, such values ​​are set as precisely as feasible.

[0117] For each patent, patent application, patent application publication, and other material such as articles, books, specifications, publications, and documents referenced in this specification, the entire contents of which are incorporated herein by reference. This excludes historical application documents that are inconsistent with or conflict with the content of this specification, as well as documents that limit the broadest scope of the claims in this specification (currently or subsequently appended to this specification). It should be noted that in the event of any inconsistency or conflict between the descriptions, definitions, and / or terminology used in the supplementary materials to this specification and the content of this specification, the descriptions, definitions, and / or terminology used in this specification shall prevail.

[0118] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be consistent with the teachings of this specification, rather than as examples or limitations. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A method for improving geological guidance crossing rate based on deep learning, characterized in that, The method is executed by a processor and includes: S1. Conduct stratigraphic attitude analysis on the target work area to obtain the first analysis result, and manually predict the trajectory trend of the horizontal section of the well based on the first analysis result; S2. Obtain geological feature data and historical drilling curve data within the target work area. Analyze the geological attributes of the target strata based on the geological feature data and historical drilling curve data to obtain a second analysis result. The second analysis result includes geological sub-layer labels and multi-source input features. S3. Based on the second analysis results, construct a deep learning algorithm learning library; the deep learning algorithm learning library consists of multi-source input features and geological layer labels; S4. Based on the deep learning algorithm learning library, a small-layer automatic recognition model is constructed through feature optimization and algorithm training; the small-layer automatic recognition model is a BP neural network model trained to a prediction accuracy that meets a preset threshold. S5. Preprocess the drilling curve data and input the preprocessed drilling curve data into the small layer automatic identification model to obtain the small layer probability distribution. S6. Based on the probability distribution of the sub-layers, the actual geological sub-layer classification at the location of the drill bit is obtained through data inverse encoding; S7. Based on the actual geological layer classification of the drill bit location, generate trajectory control instructions to improve the geological guidance penetration rate of high-quality reservoirs in horizontal wells.

2. The method for improving geological guidance crossing rate based on deep learning according to claim 1, characterized in that, Step S1 further includes: S110. Obtain seismic interpretation stratigraphic data within the target work area and single-well stratigraphic data of the study area, and perform correction processing on the two types of data to obtain corrected data; S120. Based on the correction data, an initial three-dimensional geological model is established using a spatial modeling algorithm; S130. Based on the initial three-dimensional geological model, the first analysis result is obtained through analysis; S140. Based on the first analysis result, the initial three-dimensional geological model is sliced ​​to generate a geological-guided two-dimensional model; S150. Obtain the drilling-while-drilling curve data, compare the drilling-while-drilling curve data with the corresponding predicted curve in the geological steering two-dimensional model in real time, and obtain the real-time comparison result; and update the three-dimensional geological model based on the real-time comparison result.

3. The method for improving geological guidance crossing rate based on deep learning according to claim 1, characterized in that, The multi-source input features include drilling time, total hydrocarbons, methane, far-end gamma, near-bit average gamma, upward gamma value, downward gamma value, and lithological evolution index.

4. The method for improving geological guidance crossing rate based on deep learning according to claim 1, characterized in that, Step S3 further includes: S310. Perform data cleaning on the second analysis result and extract the multi-source input features; S320. Based on the second analysis results after cleaning, manually label the geological sub-layers; S330. Based on the multi-source input features and the geological layer labels, a deep learning algorithm learning library is constructed through feature label matching.

5. The method for improving geological guidance crossing rate based on deep learning according to claim 1, characterized in that, Step S4 further includes: S410. Analyze the correlation between the multi-source input features and the geological layer labels based on the Spearman correlation coefficient method to form a correlation heatmap; S420. Based on the correlation heatmap, determine multi-source input features with correlation greater than a preset threshold to obtain preferred input features; S430. Perform range standardization on the preferred input features, encode the geological layer labels, and construct standardized input and output matrices respectively to obtain the model input layer and model output layer of the automatic layer recognition model. S440. Based on the model input layer and the model output layer, a model intermediate layer is set; the model intermediate layer includes the number of intermediate layers, the number of hidden neurons, and the activation function; S450. Based on the model input layer, the model output layer, and the model intermediate layer, construct an initial small-layer automatic recognition model; S460. The initial small-layer automatic recognition model is trained and iterated to obtain the trained small-layer automatic recognition model.

6. The method for improving geological guidance crossing rate based on deep learning according to claim 4, characterized in that, Step S460 further includes: S461. Initialize the initial small-layer automatic recognition model, randomly assign initial weights, and obtain the prediction result of the first model. S462. Based on the prediction results of the first model and the geological layer labels, calculate the training error of the initial layer automatic identification model using a preset loss function; S463. Based on the training error, the initial small-layer automatic recognition model is subjected to reverse weight adjustment, and the trained small-layer automatic recognition model is obtained through iterative optimization.

7. The method for improving geological guidance crossing rate based on deep learning according to claim 1, characterized in that, Step S7 further includes: S710. Based on the actual geological layer classification of the drill bit's location, determine the relative position of the drill bit in the target window and generate a trajectory control command; the target window is a three-dimensional spatial target body preset based on the location of a high-quality reservoir. in: S711. If the actual geological layer classification shows that the drill bit is located below the target window, then a descent command is generated. S712. If the actual geological layer classification shows that the drill bit is located in the middle of the target window, then a stabilization command is generated. S713. If the actual geological layer classification shows that the drill bit is located above the target window, then generate an slant increase command.

8. The method for improving geological guidance crossing rate based on deep learning according to claim 9, characterized in that, The slope adjustment angle of the horizontal section shall not exceed 1.5° / 10m.

Images