Image processing method for adaptive trajectory control of rotary steerable drilling

Abstract

The present application relates to the field of logging while drilling and intelligent steering control technology in oil and gas drilling engineering, in particular to a rotary steering drilling adaptive trajectory control method for image processing. It comprises: receiving original logging while drilling image data frames of downhole transmission channels; generating a high-resolution predicted geological texture map; synchronously generating a feature uncertainty distribution map representing the pixel probability distribution characteristics in the high-resolution predicted geological texture map; calculating the matching degree of the high-resolution predicted geological texture map and a preset stress-induced artifact feature library; generating a sensor contamination marker signal; calling a safety baseline geological model constructed before drilling; calculating the geological texture structure difference value relative to the safety baseline geological model; generating a geological illusion entropy; receiving the geological illusion entropy and the sensor contamination marker signal; and driving a downhole steering tool according to the activated mode. The present application ensures that the steering instruction is always based on the real existing geological information, and avoids the deviation of the drilling trajectory caused by the fictitious texture.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of logging while drilling and intelligent steering control in oil and gas drilling engineering, in particular to a self-adaptive trajectory control method for rotary steering drilling based on image processing. BACKGROUND

[0002] The rotary steering drilling system is the core equipment for oil and gas resource exploration and development, and its trajectory control precision directly determines the reservoir drilling rate and engineering safety. The system mainly relies on real-time analysis of formation texture features through logging while drilling imaging data, downhole steering tools and decision algorithms to guide the drill bit to track the target reservoir.

[0003] The existing technology is limited by the downhole mud pulse or electromagnetic wave transmission bandwidth. The original logging while drilling image data generally has the problems of low sampling rate, large noise and blurred edges. Although the deep learning super-resolution technology can improve the image quality, the generative model is easy to produce algorithm hallucinations that violate physical common sense in the lack of data support area. Moreover, the stress-induced artifacts caused by sensor crystal distortion in high temperature and high pressure environment are easy to be misjudged as real geological layering. The traditional closed-loop control system defaults that the front-end sensing data is absolutely reliable, lacks dynamic checking and fusing mechanism for AI model uncertainty and hardware physical failure, and is difficult to distinguish between real geological mutations and false signal interference, which leads to the system executing incorrect steering instructions when the model cognitive bias or sensor failure occurs, resulting in trajectory deviation or even engineering accidents. Therefore, there is an urgent need for a self-adaptive trajectory control scheme that can effectively suppress the sensing noise and model hallucination risk.

[0004] The above information disclosed in the above background section is only used to strengthen the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] To solve the above technical problems, the present application discloses a self-adaptive trajectory control method for rotary steering drilling based on image processing. Specifically, the technical scheme of the present application includes:

[0006] The low-quality perception reconstruction unit receives the original logging while drilling image data frame transmitted by the downhole transmission channel;

[0007] The deep learning image super-resolution model is run to process the original logging while drilling image data frame, and a high-resolution predicted geological texture map is generated;

[0008] A feature uncertainty distribution map representing the pixel probability distribution features in the high-resolution predicted geological texture map is synchronously generated;

[0009] The pseudo-feature discrimination unit calculates the matching degree between the high-resolution predicted geological texture map and the preset stress-induced artifact feature library;

[0010] generating a sensor contamination flag signal when the matching degree exceeds a preset reference;

[0011] calling a safety baseline geologic model built before drilling through a geologic hallucination entropy calculation unit;

[0012] calculating a geologic texture structure difference value of the high-resolution predicted geologic texture map relative to the safety baseline geologic model;

[0013] extracting data in the feature uncertainty distribution map, and performing weighted operation on the geologic texture structure difference value to generate a geologic hallucination entropy;

[0014] receiving the geologic hallucination entropy and the sensor contamination flag signal through a dual-mode arbitration control unit;

[0015] if the geologic hallucination entropy is lower than a preset threshold value and the sensor contamination flag signal is not received, activating an intelligent navigation mode and giving a trajectory planning algorithm control authority;

[0016] if the geologic hallucination entropy exceeds a preset cognitive breakdown threshold value or the sensor contamination flag signal is received, activating a mechanical baseline mode and depriving the trajectory planning algorithm of the control authority;

[0017] driving a downhole steering tool according to the activated mode through an execution and feedback correction unit.

[0018] Preferably, the data in the feature uncertainty distribution map is extracted, and the geologic texture structure difference value is subjected to weighted operation to generate a geologic hallucination entropy, including:

[0019] performing a mode alignment operation to project and rasterize the safety baseline geologic model into a baseline trend field;

[0020] the baseline trend field has the same spatial resolution as the high-resolution predicted geologic texture map;

[0021] calculating a gray scale deviation or a gradient deviation of the high-resolution predicted geologic texture map and the baseline trend field pixel by pixel;

[0022] for a region where the gray scale deviation or the gradient deviation exists, extracting a feature uncertainty value corresponding to the region in the feature uncertainty distribution map;

[0023] if the feature uncertainty value is higher than a preset high-confidence threshold value, exponentially amplifying a weight of the region deviation value in total entropy calculation;

[0024] if the feature uncertainty value is lower than a preset low-confidence threshold value, suppressing the weight of the region deviation value in total entropy calculation;

[0025] aggregating the weighted deviation values of the full image to obtain the geohallucination entropy.

[0026] Preferably, the safety baseline geologic model is constructed based on low-frequency seismic data or regional geologic rules;

[0027] The safety baseline geologic model only contains the basic stratigraphic trend of the average stratigraphic dip angle;

[0028] The stress-induced artifact feature library contains regular stripe signal features generated by the pressure deformation of the sensor;

[0029] The regular stripe signal features have specific spatial frequency and directionality.

[0030] Preferably, the mechanical baseline mode is activated to deprive the trajectory planning algorithm of control authority, including:

[0031] Stopping executing the image feature-based build-up instruction or the twist azimuth instruction;

[0032] Locking the current wellbore trajectory tangent direction;

[0033] Executing a pure-geometry build-up drilling operation to guide the drill bit to drill through the interference area.

[0034] Preferably, the method further comprises:

[0035] Monitoring the change trend of the geohallucination entropy in real time during the operation of the mechanical baseline mode;

[0036] When the geohallucination entropy is monitored to fall back to the safety interval and remain for a preset time window, determining that the data quality is restored;

[0037] Automatically releasing the locking state of the intelligent navigation mode;

[0038] Re-enabling the artificial intelligence algorithm to intervene and fine-tune the trajectory.

[0039] Preferably, the downhole steering tool is driven according to the activated mode by executing the feedback correction unit, including:

[0040] When the instruction of the intelligent navigation mode is received, the push plate or the pointing motor is driven to track the reservoir;

[0041] When the instruction of the mechanical baseline mode is received, the control parameter is locked;

[0042] The response sensitivity gain of the downhole steering tool execution mechanism is reduced;

[0043] High-frequency oscillation instructions caused by residual image noise are filtered.

[0044] Compared with the prior art, the present application has the following beneficial effects:

[0045] 1、The present application introduces a geological hallucination entropy calculation unit, which innovatively uses low-frequency seismic data to construct a safe baseline geological model, and compares it with the topological structure difference of the AI-generated super-resolution image. Combined with the weighted operation of the feature uncertainty distribution map, the system can quantitatively evaluate the credibility of the AI reconstructed image. This mechanism effectively prevents the deep learning model from generating false layering or fracture features when processing low-quality logging data, ensuring that the guidance instructions are always based on real existing geological information, and avoiding the deviation of the drilling trajectory caused by the algorithm's fabricated texture.

[0046] 2、The present application is aimed at the problem that sensors in deep well drilling often produce deformation and output interference signals due to high pressure. The present application sets up a artifact feature discrimination unit. By matching the real-time image with the stress-induced artifact feature library containing regular stripe signals, the system can accurately identify and mark non-geologically caused sensor pollution signals. Once such hardware noise is detected, the sensor pollution marker is triggered, preventing the control system from misjudging mechanical deformation as a formation boundary or a fracture, thereby eliminating the engineering risks of false build-up or twist azimuth caused by hardware failure data.

[0047] 3、The present application designs an automatic switching logic between intelligent navigation mode and mechanical baseline mode. When the geological hallucination entropy exceeds the standard or the sensor is polluted, the dual-mode arbitration control unit will immediately deprive the trajectory planning algorithm of its authority and forcibly switch to the mechanical baseline mode. In this mode, by locking the current wellbore tangent direction, reducing the sensitivity of the actuator and performing pure geometric build-up operation, the system can drill through the interference zone when the data environment deteriorates. This fault-oriented safety design ensures that the drilling tool can maintain the smoothness and stability of the wellbore trajectory when the AI is blind or hallucinatory.

[0048] 4、The present application significantly improves the stability of control through unique weighted entropy calculation and lag recovery strategy. On the one hand, by exponentially amplifying the deviation weight of high-confidence areas, the system is extremely sensitive to potential risks; on the other hand, when the mechanical baseline mode is lifted, the geological hallucination entropy needs to fall back to the safe interval and remain for a preset time window. This sharp triggering and cautious recovery strategy effectively filters high-frequency oscillation commands caused by residual image noise, avoids frequent switching between intelligent and mechanical control modes, and ensures the smoothness of the wellbore trajectory in long well section operations. BRIEF DESCRIPTION OF DRAWINGS

[0049] The present application will be further explained in conjunction with the accompanying drawings and examples:

[0050] Figure 1 is a method flowchart of the present application. DETAILED DESCRIPTION

[0051] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with specific examples.

[0052] Embodiment 1

[0053] Please refer to Figure 1 A self-adaptive trajectory control method for rotary steerable drilling image processing, comprising:

[0054] Receiving, by a low-quality perception reconstruction unit, an original logging-while-drilling image data frame of a downhole transmission channel;

[0055] Processing, by a deep learning image super-resolution model, the original logging-while-drilling image data frame to generate a high-resolution predicted geological texture map;

[0056] Synchronously generating a feature uncertainty distribution map representing the pixel probability distribution features in the high-resolution predicted geological texture map;

[0057] Calculating, by an artifact feature discrimination unit, the matching degree between the high-resolution predicted geological texture map and a preset stress-induced artifact feature library;

[0058] When the matching degree exceeds a preset reference, generating a sensor contamination marker signal; and calling, by a geological hallucination entropy calculation unit, a safety baseline geological model constructed before drilling;

[0059] Calculating the geological texture structure difference value of the high-resolution predicted geological texture map relative to the safety baseline geological model;

[0060] Extracting data in the feature uncertainty distribution map, performing weighted operation on the geological texture structure difference value, and generating a geological hallucination entropy;

[0061] Receiving, by a dual-modal arbitration control unit, the geological hallucination entropy and the sensor contamination marker signal;

[0062] If the geological hallucination entropy is lower than a preset threshold and no sensor contamination marker signal is received, activating an intelligent navigation mode and giving control authority to a trajectory planning algorithm;

[0063] If the geological hallucination entropy exceeds a preset cognitive breakdown threshold or a sensor contamination marker signal is received, activating a mechanical baseline mode and depriving the trajectory planning algorithm of control authority;

[0064] Driving, by an execution and feedback correction unit, a downhole steering tool according to the activated mode;

[0065] The low-quality perception reconstruction unit, as the primary analysis hub of the data stream, is responsible for solving the problem of raw data sparsity caused by the limited transmission of downhole mud pulses or electromagnetic waves. The unit is configured to continuously ingest raw LWD image data frames through a physical telemetry interface. Such data is typically limited by low sampling rates and exhibits high noise and edge blur characteristics. To reconstruct the geological details, the unit loads and executes a pre-trained deep learning image super-resolution model. The model uses a generative adversarial network architecture to learn the mapping relationship between a large number of core scan data and logging responses, and reconstructs the original low-quality data into high-resolution predicted geological texture maps with sub-millimeter texture characteristics.

[0066] The model generates uncertainty using the Monte Carlo dropout method while outputting the predicted texture. That is, during the inference phase, the Dropout layer in the network is kept active, and the same frame of raw LWD image data is continuously executed second random forward propagation to obtain predicted maps that are slightly different from each other. The gray variance of each pixel point in the predicted maps is calculated , and the variance matrix is normalized to generate a feature uncertainty distribution map.

[0067] This distribution map is essentially a two-dimensional scalar field aligned with the spatial resolution of the texture map. The numerical value of any coordinate point in the field quantifies the model's confidence in reconstructing the geological features at that point: high numerical values indicate that the texture details at that point are mainly derived from the model's statistical association and completion, and lack direct support from the original sensor signals; low numerical values indicate that the features at that point have a strong basis in the original data response.

[0068] The artifact feature discrimination unit performs signal screening for non-geological interference. The unit uses a pattern matching algorithm to correlate the frequency domain features of the high-resolution predicted geological texture map with a pre-set stress-induced artifact feature library. When the matching degree index obtained from the operation exceeds the pre-set benchmark, it indicates that the stripe features in the current image highly match the lattice distortion signals produced by the sensor crystal under the action of high temperature and high pressure stress in terms of spatial frequency and directionality, rather than natural sedimentary structures. At this time, the unit immediately triggers the sensor contamination marker signal.

[0069] The geologic hallucination entropy calculation unit initiates quantitative evaluation of the AI cognitive risk; the unit calls a safety baseline geologic model representing the physical bottom limit of geologic structure, calculates the geologic texture structure difference value of the current predicted texture map relative to the baseline model in the aspect of stratigraphic trend and continuity; in order to accurately distinguish real geologic mutations from false algorithm hallucinations, the unit introduces a feature uncertainty distribution map as a weighting operator to perform nonlinear weighted aggregation on the difference value, thereby outputting the geologic hallucination entropy; this parameter is a dynamic changing dimensionless index, its mathematical nature is similar to the weighted relative entropy or Kullback-Leibler divergence between the predicted distribution and the physical baseline distribution, and its physical meaning is to represent the risk degree of the generative model deviating from the physical baseline and over-associating in the absence of data support;

[0070] The decision terminal is composed of a dual-mode arbitration control unit, which establishes the reverse coupling logic of perception confidence and control authority based on the design principle of safety bottoming; in real-time monitoring operation, if the geologic hallucination entropy remains below the preset threshold and there is no pollution marker, the system determines that the inference logic of the AI model is convergent and credible, and then activates the intelligent navigation mode, authorizing the trajectory planning algorithm to call the steering tool to maximize the reservoir drilling rate; once the geologic hallucination entropy breaks through the preset cognitive collapse threshold or the sensor pollution marker signal is detected, the system determines that the current environment has exceeded the effective reasoning boundary of the AI model, and there is a high risk of induced collision, and immediately switches to the mechanical baseline mode; this operation will cut off the instruction output path of the optimization algorithm, forcing the system to execute the deterministic bottom protection drilling strategy;

[0071] This embodiment elaborates how the system establishes a dynamic balance between data-driven flexibility and engineering safety; the core innovation is to break through the limitation of the traditional closed-loop control system that assumes the front-end perception data to be absolutely reliable, to build a fuse protection mechanism for generative AI hallucination defects by introducing entropy evaluation and physical feature library double-checking, effectively preventing major engineering accidents caused by the drilling platform deviating from the safety window due to false geological images.

[0072] Embodiment 2

[0073] Extracting data in the feature uncertainty distribution map, performing weighted operation on the geologic texture structure difference value, and generating geologic hallucination entropy, including:

[0074] Performing a modality alignment operation to project and rasterize the safety baseline geologic model into a baseline trend field;

[0075] The baseline trend field and the high-resolution predicted geologic texture map have the same spatial resolution; the gray scale deviation or gradient deviation of the high-resolution predicted geologic texture map and the baseline trend field is calculated pixel by pixel;

[0076] For the area with gray scale deviation or gradient deviation, the feature uncertainty value corresponding to the area in the feature uncertainty distribution map is extracted;

[0077] If the feature uncertainty value is higher than the preset high confidence threshold, the weight of the area deviation value in the total entropy calculation is exponentially amplified;

[0078] If the feature uncertainty value is lower than the preset low confidence threshold, the weight of the area deviation value in the total entropy calculation is suppressed;

[0079] The weighted deviation values of the entire image are aggregated to obtain the geological hallucination entropy;

[0080] The geological hallucination entropy calculation unit performs a strict quantitative operation process, aiming to separate the real geological signal from the algorithmic noise by mathematical means; The unit performs a modal alignment operation, projects the low-dimensional safety baseline geological model along the wellbore trajectory section using a spatial interpolation algorithm, and rasterizes it into a baseline trend field with pixel-level correspondence with the high-resolution predicted geological texture map; Perform Gaussian pyramid downsampling or low-pass filtering processing on the high-resolution predicted geological texture map to make its spatial resolution consistent with the baseline trend field, thereby constructing a unified deviation calculation coordinate system; This step ensures that the subsequent difference calculation is based on a unified spatial coordinate system and resolution;

[0081] The unit performs pixel-by-pixel deviation calculation to obtain the numerical difference of the predicted texture map relative to the baseline trend field in gray intensity or local gradient direction; The key processing logic lies in the subsequent weighted operation: for each pixel point with significant deviation, the system indexes its corresponding value in the feature uncertainty distribution map;

[0082] When the uncertainty value is higher than the preset high confidence threshold, the system determines that the deviation at this point is most likely to belong to hallucination features without cause; Therefore, the system uses an exponential function to weight and amplify the deviation value of this area, making it dominate the weight in the final entropy value composition, thereby significantly raising the overall risk score;

[0083] When the uncertainty value is lower than the preset low confidence threshold, the system determines that the deviation at this point deviates from the baseline model, but belongs to the real existing geological heterogeneity; At this time, the system uses a logarithmic decay or linear suppression function to reduce the weight of the deviation value, preventing real unexpected geological discoveries from triggering the fuse mechanism;

[0084] The system integrates or accumulates the deviation values of all pixel points in the whole image after weighted adjustment, and outputs the geological hallucination entropy; by constructing this nonlinear weighted evaluation system based on reliability, the embodiment realizes fine quality inspection of the AI generated content, ensures that the system only alarms when the model generates non-physical real characteristics, and remains silent when the model discovers real details not contained in the baseline based on data, thereby retaining the detection sensitivity for complex geology under the premise of ensuring safety.

[0085] Embodiment 3

[0086] The safety baseline geological model is constructed based on low-frequency seismic data or regional geological rules;

[0087] The safety baseline geological model only contains the basic stratigraphic trend of the average stratigraphic dip;

[0088] The stress-induced artifact feature library contains regular stripe signal features generated by the deformation of the sensor under pressure;

[0089] The regular stripe signal features have specific spatial frequency and directionality;

[0090] The physical source and attribute of the system decision benchmark are clearly defined, which is the cornerstone of ensuring the objectivity of the system; the safety baseline geological model refers to a coarse-grained model constructed by using low-frequency ground seismic wave inversion data or regional exploration historical geological rules; the construction principle of the model is physical conservation, only the basic stratigraphic trend of low frequency, large scale such as average stratigraphic dip and main structure trend is retained, and all high-frequency microscopic texture details are removed; this design aims to provide a physically absolutely reliable background plate for highlighting and testing whether the fine texture generated by AI is in line with the basic physical logic;

[0091] The stress-induced artifact feature library is established based on the physical property test of the sensor hardware and the analysis of historical failure data; the stress-induced artifacts included therein refer to specific interference patterns output by the logging-while-drilling instrument when detecting the elastic deformation of the crystal or the thermal noise of the circuit under the extreme environment of high temperature and high pressure at the bottom of the well; such patterns show periodic signals with fixed spatial frequency in the frequency domain, and regular stripes with high direction consistency in the spatial domain, which have significant statistical differences with the randomness and non-uniformity characteristics of natural sedimentary layering.

[0092] Embodiment 4

[0093] Activating the mechanical baseline mode to deprive the trajectory planning algorithm of control authority, including:

[0094] Stopping executing the build-up instruction or the twist azimuth instruction based on the image features;

[0095] Locking the tangent direction of the current wellbore trajectory;

[0096] Performing pure geometric steady build-up drilling operation, guiding the drill bit to drill through the interference area;

[0097] The mechanical baseline mode is defined as an inertial navigation maintenance strategy under non-line-of-sight conditions; when the dual-mode arbitration control unit triggers this mode, the system performs a strict control handover procedure: immediately shield all control signals from the image recognition module, stop executing any dynamic adjustment instructions that attempt to build or twist the azimuth according to the formation boundary;

[0098] The system locks the current well trajectory tangent vector, switches to a pure geometric steady build-up drilling logic; in this state, the drilling control system degenerates into a gravity tool face and magnetic tool face-based attitude maintenance system, no longer responding to changes in geological parameters, but strictly maintaining the current inclination and azimuth, guiding the drill bit to pass through the signal interference area in a straight line or smooth circular arc geometric trajectory.

[0099] Embodiment 5

[0100] The method further comprises:

[0101] During the operation of the mechanical baseline mode, the change trend of the geological illusion entropy is monitored in real time;

[0102] When the geological illusion entropy is monitored to fall back to the safe interval and remain for a preset time window, the data quality is determined to be restored;

[0103] The intelligent navigation mode is automatically unlocked;

[0104] The artificial intelligence algorithm is re-allowed to intervene and fine-tune the trajectory;

[0105] The system is configured with a closed-loop recovery logic with hysteresis characteristics to solve the control stability problem under complex working conditions; although the system has entered the mechanical baseline mode and locked the control right, the background computing process still continuously monitors the real-time evolution of the geological illusion entropy; as the drilling depth increases, when the drill bit physically drives away from the stress anomaly zone that causes sensor deformation, the artifact features in the imaging data will gradually disappear, and the geological illusion entropy will show a downward trend;

[0106] To prevent frequent switching of modes due to critical fluctuations in signals, the system sets strict recovery criteria: only when the geological illusion entropy not only falls back to the safe interval, but also remains in this state for a preset time window, does the system confirm that the data environment has substantially improved; after confirmation, the arbitration unit automatically unlocks the intelligent navigation mode, allowing the artificial intelligence algorithm to re-enter the control loop and finely tune the trajectory based on the restored clear geological image.

[0107] Embodiment 6

[0108] By executing the downhole steering tool driven by the feedback correction unit according to the activated mode, including:

[0109] When receiving the instruction of the intelligent navigation mode, the driving push plate or the pointing motor traces the reservoir;

[0110] When receiving the instruction of the mechanical baseline mode, the control parameters are locked;

[0111] The response sensitivity gain of the downhole steering tool execution mechanism is reduced;

[0112] The high-frequency oscillation instructions caused by residual image noise are filtered;

[0113] The execution and feedback correction unit is not only responsible for the physical execution of the instruction, but also dynamically adjusts the response characteristics of the underlying servo system according to the currently activated control mode; when the system is in the intelligent navigation mode, the unit is set to a high-gain state, and the hydraulic push plate or the pointing motor is driven to execute small attitude adjustments at a millisecond-level response speed, so as to realize accurate tracking of thin reservoirs;

[0114] Once switched to the mechanical baseline mode, the unit not only locks the tool face angle and other key control parameters, but also actively reduces the Gain coefficient of the execution mechanism control loop; this hardware-level sensitivity reduction operation aims to build a low-pass filtering effect, directly filtering out high-frequency oscillation instructions that may be caused by residual image noise or sensor micro-vibration from the physical execution end.

[0115] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. An image processing-based adaptive trajectory control method for rotary steered drilling, characterized in that, include: The raw logging-while-drilling image data frames are received from the downhole transmission channel through the low-quality sensing reconstruction unit. The original logging-while-drilling image data frames are processed by running a deep learning image super-resolution model to generate a high-resolution predicted geological texture map; Simultaneously generate a feature uncertainty distribution map that characterizes the pixel probability distribution features in the high-resolution predicted geological texture map; The matching degree between the high-resolution predicted geological texture map and the preset stress-induced artifact feature library is calculated by the artifact feature identification unit. When the matching degree exceeds a preset benchmark, a sensor contamination marker signal is generated; The geological illusion entropy calculation unit calls the safety baseline geological model constructed before drilling. Calculate the difference in geological texture structure between the high-resolution predicted geological texture map and the safe baseline geological model; Data from the feature uncertainty distribution map is extracted, and the geological texture structure difference values ​​are weighted to generate geological illusion entropy. The geological illusion entropy and the sensor contamination marker signal are received by the dual-modal arbitration control unit; If the geological illusion entropy is lower than a preset threshold and no contamination marker signal is received from the sensor, the intelligent navigation mode is activated, and the trajectory planning algorithm is given control authority. If the geological illusion entropy exceeds the preset cognitive collapse threshold, or if the sensor contamination marker signal is received, the mechanical baseline mode is activated, and the trajectory planning algorithm is deprived of its control authority. The downhole directional tool is driven by the execution and feedback correction unit according to the activated mode; Data from the feature uncertainty distribution map is extracted, and the geological texture structure difference values ​​are weighted to generate a geological illusion entropy, including: Perform modal alignment operation to project and rasterize the safe baseline geological model into a baseline trend field; The baseline trend field has the same spatial resolution as the high-resolution predicted geological texture map; Calculate the grayscale deviation or gradient deviation between the high-resolution predicted geological texture map and the baseline trend field pixel by pixel; For regions where the grayscale deviation or the gradient deviation exists, extract the corresponding feature uncertainty value of the region in the feature uncertainty distribution map; If the value of the uncertainty of the feature is higher than the preset high confidence threshold, the weight of the deviation value of that region in the total entropy calculation is increased exponentially; If the value of the uncertainty of the feature is lower than the preset low confidence threshold, the weight of the deviation value of that region in the total entropy calculation is suppressed; The weighted deviation values ​​of the entire image are aggregated to obtain the geological illusion entropy.

2. The image processing-based adaptive trajectory control method for rotary steered drilling according to claim 1, characterized in that, The safety baseline geological model is constructed based on low-frequency seismic data or regional geological patterns. The safety baseline geological model only includes the basic stratigraphic trend of the average dip angle of the strata; The stress-induced artifact feature library contains regular stripe signal features generated by the sensor under pressure deformation; The regular stripe signal features have specific spatial frequency and directionality.

3. The image processing-based adaptive trajectory control method for rotary steered drilling according to claim 1, characterized in that, Activating the mechanical baseline mode and depriving the trajectory planning algorithm of control includes: Stop executing image feature-based tilting or azimuth-shifting commands; Lock the current wellbore trajectory tangent direction; Perform pure geometric steady-slope drilling operations to guide the drill bit through the interference area.

4. The image processing-based rotary steered drilling adaptive trajectory control method according to claim 3, characterized in that, The method further includes: During the operation of the mechanical baseline mode, the changing trend of the geological illusion entropy is monitored in real time; Once the geological illusion entropy is detected to have fallen back to a safe range and remained within a preset time window, it is determined that the data quality has been restored. Automatically unlock the intelligent navigation mode; Artificial intelligence algorithms are now allowed to intervene and fine-tune the trajectory.

5. The image processing-based adaptive trajectory control method for rotary steered drilling according to claim 1, characterized in that, The downhole directional tool is driven by an execution and feedback correction unit based on the activated mode, including: When the intelligent navigation mode command is received, the push plate or directional motor is driven to track the reservoir. When the instruction for the mechanical baseline mode is received, the control parameters are locked; Reduce the response sensitivity gain of the downhole guiding tool actuator; Filter out high-frequency oscillation commands caused by residual image noise.

Images