A trajectory correction method based on drilling survey data
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-05
- Publication Date
- 2026-08-14
Smart Images
Figure CN122565373A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of oil drilling automation, and in particular to a trajectory correction method based on measurement-while-drilling (MWD) data. Background Technology
[0002] In oil, gas, and geological mining drilling, trajectory correction technology is crucial to ensuring the drill bit accurately follows the predetermined design path. Current technologies typically utilize measurement-while-drilling (MWD) tools to collect geometric position parameters such as inclination angle, azimuth angle, and tool face angle in real time, transmitting this data to the surface via mud pulses or electromagnetic waves. Surface technicians or automated surface systems calculate correction parameters based on the geometric deviations between the measured values and the designed trajectory, then transmit this data downlink to downhole actuators such as rotary steering systems. These actuators correct the drilling direction by adjusting flange pressure or changing the deflection vector. To improve the scientific rigor of correction, some solutions have begun to incorporate simple dynamic models, attempting to aid decision-making by analyzing the forces acting on the drill bit.
[0003] Existing technologies, when processing complex drilling dynamic data, cannot effectively distinguish between different dynamic sources causing trajectory deviations. This results in the system's inability to accurately determine whether the trajectory deviation is caused by formation environment disturbances or by actuator performance fluctuations. This lack of clarity regarding the cause of deviations and the resulting blind control strategies directly limit the accuracy and effectiveness of trajectory correction. Summary of the Invention
[0004] To overcome the above shortcomings, this invention provides a trajectory correction method based on drilling survey data, which aims to improve the problem that the system cannot accurately determine whether the trajectory deviation is caused by formation environment disturbance or by performance fluctuation of the actuator.
[0005] This invention provides the following technical solution: a trajectory correction method based on drilling survey data, comprising:
[0006] S1. Real-time acquisition of drilling survey data, surface drilling parameters, and near-bit high-frequency vibration signals, which are then used as input parameters into the drilling system energy transfer model to construct a lateral power flow model;
[0007] S2. Use the blind source separation algorithm to perform feature decomposition on the lateral power flow model to extract the environmental disturbance term and the controlled deflection term;
[0008] S3. Map the environmental disturbance term, the controlled deflection term, and the drilling survey data to a high-dimensional feature space, extract the feature manifold through a manifold learning algorithm, and determine the timing of the correction trigger based on the feature manifold.
[0009] S4. When the correction triggering time is reached, the trajectory evolution model is used to solve the problem with the energy loss in the lateral power flow model and the geometric distance corresponding to the drilling survey data as the objective function, and the correction evolution operator is obtained.
[0010] S5. The shadow model is run synchronously between the ground terminal and the downhole node. The ground terminal converts the deviation evolution operator into an incremental correction offset and transmits the correction offset to the downhole node through the communication link.
[0011] S6. The downhole node receives the correction offset and performs self-calibration processing on the correction offset in combination with the real-time monitored near-drill bit pose data to generate an execution command, driving the downhole actuator to perform trajectory correction actions.
[0012] Preferably, in S1, the energy transfer model of the drilling system includes:
[0013] The total input mechanical work of the drilling system is calculated based on the surface drilling parameters, and the spatial force configuration of the drilling system is constructed by combining the drilling survey data.
[0014] By utilizing the spatial force configuration, the total input mechanical work loss is analyzed, and the axial propulsion loss, tubing rotation torsion loss, and friction loss generated by the contact between the drill string and the well wall are extracted during the drilling process.
[0015] The real-time total effective energy flow transmitted to the drill bit is determined by subtracting the axial propulsion loss, the torsional loss of the tubing rotation, and the friction loss from the total input mechanical work.
[0016] Preferably, in S1, constructing the lateral power flow model includes:
[0017] The lateral displacement field of the drill bit at the bottom plane is identified by using the high-frequency vibration signal near the drill bit, and the lateral virtual work component of the drill bit at the bottom plane is calculated by combining the total effective energy flow.
[0018] Obtain the bending strain characteristics of the drilling system near the drill bit, and calculate the lateral bending strain energy components of the drilling system on the correction trajectory;
[0019] A lateral power flow model is constructed by vector superposition of the lateral virtual work component and the lateral bending strain energy component.
[0020] Preferably, in S2, the eigenvalue decomposition of the lateral power flow model using the blind source separation algorithm includes:
[0021] The lateral power flow model is resampled in the time dimension to construct the observation signal matrix;
[0022] Using the independent component analysis algorithm as the benchmark algorithm for blind source separation, the observed signal matrix is de-aliased to determine a set of statistically independent source signal components.
[0023] By performing spectral analysis on each source signal component, the frequency characteristics and energy density corresponding to each source signal component are identified, and the feature decomposition of the lateral power flow model is completed.
[0024] Preferably, in S2, the extraction of the environmental disturbance term and the controlled deflection term includes:
[0025] A preset feature library is established, which includes the operating frequency modes of the actuator and the disturbance feature modes of the formation anisotropy;
[0026] The decomposed source signal components are matched with a preset feature library, and the source signal components that conform to the action frequency mode are reconstructed into controlled deflection terms.
[0027] The source signal components that conform to the disturbance characteristic mode and whose frequency distribution is within a preset random bandwidth are reconstructed into environmental disturbance terms.
[0028] Preferably, in S3, the extraction of feature manifolds via a manifold learning algorithm includes:
[0029] By fusing environmental disturbance terms, controlled deflection terms, and drilling survey data with time-series features, a high-dimensional phase space matrix is constructed.
[0030] The geodesic distance between each sampling point in the high-dimensional phase space matrix is calculated using the equidistant mapping algorithm to generate a global distance matrix.
[0031] Nonlinear dimensionality reduction is performed on the global distance matrix to find low-dimensional structures embedded in the high-dimensional phase space and extract feature manifolds that reflect the trajectory evolution law.
[0032] Preferably, in S3, determining the correction triggering timing based on the characteristic manifold includes:
[0033] Real-time monitoring of the motion trajectory of the characteristic manifold in low-dimensional space, and calculation of the instantaneous curvature of the motion trajectory and the diffusion velocity of the region deviating from the design manifold;
[0034] Establish a correction evaluation function, wherein the instantaneous curvature is a variable;
[0035] When the instantaneous curvature exceeds the preset curvature threshold, or the diffusion speed exceeds the preset trend threshold, the current moment is determined as the timing for correction triggering.
[0036] Preferably, in S4, the solution process, using the energy loss in the lateral power flow model and the geometric distance corresponding to the drilling survey data as the objective function, includes:
[0037] A trajectory evolution model is constructed by a bidirectional gated recurrent neural network, and the historical pose sequence corresponding to the drilling survey data and the real-time feature parameters corresponding to the lateral power flow model are input into the trajectory evolution model to generate multiple sets of candidate evolution trajectories.
[0038] A collaborative optimization objective function is established based on the energy loss in the lateral power flow model and the geometric distance corresponding to the drilling measurement data, and the collaborative optimization objective function is used to evaluate and score each candidate evolution trajectory.
[0039] The candidate evolutionary trajectory with the highest evaluation score is selected as the target regression path, and a correction evolution operator is generated based on the control parameters corresponding to the target regression path.
[0040] Preferably, in S5, the synchronous operation of the shadow model between the ground terminal and the downhole node includes:
[0041] Neural network models with consistent initial parameters are pre-set as shadow models in both the ground terminal and the downhole node, and a unified causal evolution state space is established for the shadow models.
[0042] The ground terminal acquires the correction evolution operator in real time, calculates the parameter increment of the correction evolution operator relative to the current state of the shadow model, and encapsulates the parameter increment as a correction offset for transmission.
[0043] After receiving the corrected offset, the downhole node uses the corrected offset to perform local parameter reconstruction on its locally running shadow model.
[0044] Preferably, in S6, the step of performing self-calibration processing on the correction offset based on real-time monitored near-drill bit pose data to generate execution instructions includes:
[0045] The measured pose vector is constructed using the near-drill bit pose data, and the theoretical pose vector predicted by the shadow model at the current moment is retrieved.
[0046] Calculate the spatial residual between the measured pose vector and the theoretical pose vector, and input the spatial residual into a preset Kalman filter calibrator to calculate the real-time gain compensation coefficient for the prediction bias of the shadow model.
[0047] The received correction offset is weighted and corrected using a real-time gain compensation coefficient to generate an execution instruction.
[0048] The present invention has the following beneficial effects:
[0049] 1. In this invention, the precise decoupling of environmental disturbance terms and controlled deflection terms is achieved through the drill string energy transfer model and blind source separation algorithm, which solves the problem that the traditional deviation correction method cannot accurately separate the natural formation deflection force, resulting in unclear deviation causes and blind control strategies.
[0050] 2. In this invention, the feature manifold of the high-dimensional phase space is extracted by manifold learning algorithm, and the dynamic capture of the correction triggering time is realized by using instantaneous curvature and diffusion velocity. This solves the problem of decision lag and drastic fluctuation of wellbore trajectory caused by the traditional method relying on a fixed geometric deviation threshold for triggering.
[0051] 3. In this invention, through the incremental synchronization mechanism of the ground and downhole shadow models and the edge self-calibration processing based on spatial residuals, the efficient transmission of decision parameters and the accurate compensation of execution instructions under complex working conditions are realized, which solves the problems of ground decision transmission delay under limited communication bandwidth and deviation between theoretical model prediction and actual working conditions. Attached Figure Description
[0052] Figure 1 This is a flowchart of a trajectory correction method based on drilling survey data proposed in this invention;
[0053] Figure 2 This is a flowchart of the energy transfer and lateral power flow construction of a drilling tool system based on a trajectory correction method using drilling measurement-while-drilling data, as proposed in this invention.
[0054] Figure 3 This is a feature extraction logic diagram based on blind source separation for a trajectory correction method based on drilling survey data proposed in this invention;
[0055] Figure 4 This invention presents a feature manifold extraction and correction triggering decision graph for a trajectory correction method based on drilling survey data.
[0056] Figure 5 This is a flowchart illustrating the downhole edge self-calibration execution process of a trajectory correction method based on drilling measurement data proposed in this invention. Detailed Implementation
[0057] The technical solutions in 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 some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0058] In embodiments of the present invention, a trajectory correction method based on drilling survey data is provided, such as... Figures 1-5 As shown, it includes the following steps:
[0059] S1. Real-time acquisition of drilling survey data, surface drilling parameters, and near-bit high-frequency vibration signals, which are then used as input parameters into the drilling system energy transfer model to construct a lateral power flow model;
[0060] Furthermore, in S1, the energy transfer model of the drilling system includes:
[0061] The total mechanical work input of the drilling tool system is calculated based on the ground drilling parameters, and the spatial force configuration of the drilling tool system is constructed by combining the drilling survey data.
[0062] By utilizing the spatial force configuration, the total input mechanical work loss is analyzed, and the axial propulsion loss, tubing rotation torsion loss, and friction loss generated by the contact between the drill string and the well wall are extracted during the drilling process.
[0063] The real-time total effective energy flow transmitted to the drill bit is determined by subtracting the axial propulsion loss, the torsional loss of the tubing rotation, and the friction loss from the total input mechanical work.
[0064] Furthermore, in S1, constructing the lateral power flow model includes:
[0065] The lateral displacement field of the drill bit at the bottom plane is identified by using the high-frequency vibration signal near the drill bit, and the lateral virtual work component of the drill bit at the bottom plane is calculated by combining the total effective energy flow.
[0066] Obtain the bending strain characteristics of the drilling system near the drill bit, and calculate the lateral bending strain energy components of the drilling system on the correction trajectory;
[0067] A lateral power flow model is constructed by vector superposition of the lateral virtual work component and the lateral bending strain energy component.
[0068] Specifically, the system acquires real-time measurement-while-drilling (MWD) data, surface drilling parameters, and near-bit high-frequency vibration signals through various sensors integrated on the drill string assembly. The MWD data primarily includes the inclination angle and azimuth angle, used to determine the three-dimensional pose of the drill string in the underground space; the surface drilling parameters cover dynamic information such as rotary table torque, hook load, and drilling pressure; and the near-bit high-frequency vibration signals reflect the microscopic motion state of the drill bit during rock breaking. These raw data are uniformly used as input parameters and fed into a pre-defined drill string system energy transfer model to construct a lateral power flow model reflecting the lateral offset trend of the drill bit.
[0069] In practice, the drilling system energy transfer model first calculates the total input mechanical work of the drilling system using surface drilling parameters. This total input mechanical work represents the total energy coupled from the surface power system to the drill string. Subsequently, the system combines real-time acquired drilling survey data to construct the spatial force configuration of the drilling system. Under the constraints of this spatial force configuration, the model performs in-depth loss analysis on the total input mechanical work. Through the mechanical torque balance relationship and friction model, it extracts the axial propulsion loss, torsional loss of the drill string rotation, and friction loss generated by the contact between the drill string and the wellbore during the actual drilling process. The axial propulsion loss mainly originates from the work done to overcome the formation resistance, the torsional loss originates from the internal energy consumption during the rotation of the drill string, and the friction loss is obtained by calculating the contact normal pressure and friction coefficient between each node of the drill string and the wellbore.
[0070] After obtaining the aforementioned loss terms, the model performs an energy balance calculation. By subtracting axial propulsion loss, tubing rotation torsion loss, and friction loss from the total input mechanical work, the residual energy is calculated. This residual energy is logically defined as the real-time total effective energy flow transferred to the drill bit, i.e., the net work actually used for rock breaking and potentially causing trajectory deviation.
[0071] Subsequently, the system enters the stage of constructing the lateral power flow model. Using near-bit high-frequency vibration signals, and through acceleration integration and signal filtering techniques, the lateral displacement field of the drill bit at the bottom plane is identified. This displacement field characterizes the instantaneous oscillation amplitude of the drill bit in a plane perpendicular to the design axis. The system, combined with the previously obtained real-time total effective energy flow, calculates the lateral virtual work component of the drill bit at the bottom plane. The lateral virtual work component is calculated using the formula... The calculation shows that, Indicates the lateral virtual work component. This represents the equivalent lateral force. This represents the displacement vector in the lateral displacement field.
[0072] Simultaneously, the system acquires the bending strain characteristics of the drilling tool system near the drill bit using strain gauges mounted on the drill string. These characteristics reflect the degree of deformation of the drill string due to wellbore bending or uneven stress. Based on the principles of mechanics of materials, the lateral bending strain energy components of the drilling tool system on the current correction trajectory are calculated. Lateral bending strain energy components. Calculation reference formula In the specific calculations of this scheme, it is simplified to In this formula, Represents the lateral bending strain energy component. This represents the equivalent bending stiffness coefficient of the drilling system. This indicates the change in the bending angle of the drill bit relative to the designed trajectory.
[0073] Finally, the system vector-superimposes the calculated lateral virtual work component with the lateral bending strain energy component. This superposition comprehensively considers the kinetic energy consumption trend generated by motion and the potential energy accumulation trend generated by deformation, thus constructing a complete lateral power flow model. In this model, the magnitude and direction of the power flow reflect the dynamic tendency of the drill bit to deviate from the predetermined trajectory, providing accurate data support for subsequent deviation decoupling and correction command generation. In this scheme, the equivalent bending stiffness coefficient... Preferably, the values are pre-calibrated based on the material and cross-sectional characteristics of the drill bit to ensure the accuracy of energy calculation.
[0074] S2. Use the blind source separation algorithm to perform feature decomposition on the lateral power flow model to extract the environmental disturbance term and the controlled deflection term;
[0075] Furthermore, in S2, the eigenvalue decomposition of the lateral power flow model using the blind source separation algorithm includes:
[0076] The lateral power flow model is resampled in the time dimension to construct the observation signal matrix;
[0077] Using the independent component analysis algorithm as the benchmark algorithm for blind source separation, the observed signal matrix is de-aliased to determine a set of statistically independent source signal components.
[0078] By performing spectral analysis on each source signal component, the frequency characteristics and energy density corresponding to each source signal component are identified, and the feature decomposition of the lateral power flow model is completed.
[0079] Furthermore, in S2, the extraction of environmental disturbance terms and controlled deflection terms includes:
[0080] Establish a pre-set feature library, which includes the action frequency modes of the actuators and the disturbance feature modes of the formation anisotropy;
[0081] The decomposed source signal components are matched with a preset feature library, and the source signal components that conform to the action frequency mode are reconstructed into controlled deflection terms.
[0082] The source signal components that conform to the disturbance characteristic mode and whose frequency distribution is within a preset random bandwidth are reconstructed into environmental disturbance terms.
[0083] Specifically, after acquiring the lateral power flow model, the system enters the feature decoupling stage, aiming to extract the different source dynamic features affecting the trajectory from the complex dynamic response. The system first uses a blind source separation algorithm to perform feature decomposition on the lateral power flow model. Since the lateral power flow signal exhibits non-stationary characteristics in the time domain and is mixed with various mechanical vibrations and ground feedback, the system first resamples the lateral power flow model in the time dimension. The resampling process involves applying anti-aliasing filtering to the original signal and then discretizing it according to a uniform time step to construct an observation signal matrix. This observation signal matrix serves as a multi-dimensional input source, containing the power fluctuation situation at different sampling points.
[0084] After constructing the observed signal matrix, the system uses Independent Component Analysis (ICA) as the baseline algorithm for blind source separation. ICA is a computational method that finds linearly combined components by maximizing the statistical independence of signals. It dealiases the mixed signals in the observed signal matrix by optimizing the contrast function. Specifically, it searches for a demixing matrix that minimizes the mutual information between the components of the demixed output vector, thereby identifying a set of statistically independent source signal components. The demixing process is illustrated by the following formula:
[0085] ;
[0086] in, This represents the demixed source signal component vector. The calculated unmixing matrix, This represents the input observed signal matrix. In this scheme, the iteration step size of the unmixing matrix is preferably 0.01.
[0087] After obtaining the source signal components, the system performs spectral analysis on each component. Spectral analysis uses a Fast Fourier Transform (FFT) to convert the time-domain source signal into a frequency-domain representation, thereby identifying the frequency characteristics and energy density of each component. The frequency characteristics reflect the periodic variation of the signal, while the energy density characterizes the intensity distribution of that component in the power contribution. Thus, the system completes the eigenvalue decomposition of the lateral power flow model, transforming the continuous power flow signal into independent components with distinct frequency attributes.
[0088] Subsequently, the system executes the extraction process for environmental disturbance terms and controlled deflection terms. The system first establishes a pre-built feature library, which is stored as prior knowledge in the computational unit. This library contains the operating frequency modes of the actuators and the disturbance characteristic modes of formation anisotropy. The operating frequency modes of the actuators typically correspond to the extension / retraction frequency of the deflector blades of the rotary steering system or the rotation frequency of the tool face, exhibiting clear regularity and narrow-band characteristics. The disturbance characteristic modes of formation anisotropy, on the other hand, are related to random excitations associated with abrupt changes in formation hardness or formation dip angle.
[0089] During the extraction process, the system performs correlation matching between the decomposed source signal components and modes in a pre-set feature library. By calculating the energy concentration of the source signal components within a specific frequency band, the source signal components conforming to the action frequency mode are reconstructed into a controlled deflection term. This term represents the effective work performed by the actuator driven by manual or automatic correction commands. Simultaneously, the system reconstructs source signal components conforming to the disturbance characteristic mode and whose frequency distribution falls within a preset random bandwidth into an environmental disturbance term. The environmental disturbance term reflects the interference effect of the natural formation build-up force on the drill bit. In this scheme, the preset random bandwidth is preferably between 5 Hz and 50 Hz. Through the above logic, the system achieves the quantitative separation of correction execution force and environmental disturbance force, providing clean feature input for subsequent trajectory steady-state evaluation.
[0090] S3. Map the environmental disturbance term, the controlled deflection term, and the drilling survey data to a high-dimensional feature space, extract the feature manifold through the manifold learning algorithm, and determine the timing of the correction trigger based on the feature manifold.
[0091] Furthermore, in S3, the extraction of feature manifolds through manifold learning algorithms includes:
[0092] By fusing environmental disturbance terms, controlled deflection terms, and drilling survey data with time-series features, a high-dimensional phase space matrix is constructed.
[0093] The geodesic distance between each sampling point in the high-dimensional phase space matrix is calculated using the equidistant mapping algorithm to generate a global distance matrix.
[0094] Nonlinear dimensionality reduction is performed on the global distance matrix to find low-dimensional structures embedded in the high-dimensional phase space and extract feature manifolds that reflect the trajectory evolution law.
[0095] Furthermore, in S3, determining the correction trigger timing based on the characteristic manifold includes:
[0096] Real-time monitoring of the motion trajectory of the characteristic manifold in low-dimensional space, and calculation of the instantaneous curvature of the motion trajectory and the diffusion velocity of the region deviating from the design manifold;
[0097] Establish a correction evaluation function, with instantaneous curvature as the variable;
[0098] When the instantaneous curvature exceeds the preset curvature threshold, or the diffusion speed exceeds the preset trend threshold, the current moment is determined as the timing for correction triggering.
[0099] Specifically, in the trajectory steady-state evaluation stage of this scheme, the system deeply integrates the extracted environmental disturbance term, controlled deflection term, and real-time feedback logging-while-drilling data, aiming to capture the essential laws of wellbore trajectory evolution through nonlinear dimensionality reduction. The system first executes a manifold learning algorithm to extract feature manifolds. In this process, the environmental disturbance term represents the interference characteristics of the natural formation build-up force, the controlled deflection term represents the active control characteristics of the actuator, and the logging-while-drilling data provides a spatial geometric coordinate reference. The system then fuses these three components using temporal features, aligning the multidimensional parameters according to the sampling timestamp to construct a high-dimensional phase space matrix. Each sampling point in this matrix represents the comprehensive dynamics and geometric pose state of the drill bit at a given instant.
[0100] After constructing the high-dimensional phase space matrix, the system uses the isometric mapping algorithm to calculate the geodesic distance between each sampling point. The isometric mapping algorithm is a geometrically based manifold learning method. Its core logic is that Euclidean distance in high-dimensional space cannot truly reflect the relationships between points on a nonlinear manifold. The system constructs a local nearest neighbor graph and uses Dijkstra's algorithm to calculate the shortest path between any two points in the graph, thus approximating the geodesic distance on the manifold surface, and generating a global distance matrix. (Geodesic distance) The computational logic lies in ensuring that distance measurement unfolds along the geometry of the manifold, rather than through gaps in high-dimensional space.
[0101] After obtaining the global distance matrix, the system performs nonlinear dimensionality reduction. Specifically, the system uses multidimensional scaling transformation techniques to perform eigenvalue decomposition on the global distance matrix, searching for low-dimensional coordinate embeddings that maintain the geodesic distance. By eliminating redundant high-dimensional noise, the system searches for and discovers low-dimensional structures embedded in the high-dimensional phase space, ultimately extracting feature manifolds that reflect the trajectory evolution. The geometric shape of the feature manifold in the low-dimensional space characterizes the operating mode of the drilling system under the current conditions, such as a steady drilling mode or an unstable oscillation mode.
[0102] After manifold extraction, the system proceeds to determine the timing of correction triggers. The system monitors the trajectory of the characteristic manifold in low-dimensional space in real time; the evolution trend of this trajectory directly corresponds to the dynamic changes in the actual wellbore trajectory. The system calculates the instantaneous curvature of the trajectory and the diffusion velocity of deviation from the designed manifold region in real time. The instantaneous curvature is calculated using the formula... The calculation shows that, Indicates instantaneous curvature, The position vector represents the trajectory of the manifold. and These represent the first and second derivatives of the position vector with respect to time, respectively. The diffusion rate characterizes how quickly the current state point moves away from the stable design region.
[0103] The system establishes a correction evaluation function to quantitatively assess the trajectory status. This function uses instantaneous curvature as the core variable and also considers the influence of diffusion velocity. Evaluation Function The expression form is:
[0104] ;
[0105] In this formula, This represents the result value of the evaluation function. This is the curvature weighting coefficient. For speed weighting coefficients, This represents the diffusion velocity deviating from the designed manifold region. In this scheme, the curvature weighting coefficient... The preferred value is 0.6, the velocity weighting coefficient. The preferred value is 0.4.
[0106] When the system detects that the instantaneous curvature exceeds a preset curvature threshold, or the diffusion velocity exceeds a preset trend threshold, it means that the trajectory is about to enter an uncontrollable, unsteady region or has already produced a serious geometric deviation. At this point, the system automatically determines the current moment as the timing for correction. Through this determination method based on phase space dynamics evolution, the system can capture signs of instability at the nonlinear trend level before the deviation fully manifests as a large geometric shift, thus providing a feedforward decision-making basis for subsequent path planning.
[0107] S4. When the correction triggering time is reached, the trajectory evolution model is used to solve the problem with the energy loss in the lateral power flow model and the geometric distance corresponding to the drilling survey data as the objective function, and the correction evolution operator is obtained.
[0108] Furthermore, in S4, the objective function is to solve for the energy loss in the lateral power flow model and the geometric distance corresponding to the drilling survey data, including:
[0109] A trajectory evolution model is constructed by a bidirectional gated recurrent neural network, and the historical pose sequence corresponding to the drilling survey data and the real-time feature parameters corresponding to the lateral power flow model are input into the trajectory evolution model to generate multiple sets of candidate evolution trajectories.
[0110] A collaborative optimization objective function is established based on the energy loss in the lateral power flow model and the geometric distance corresponding to the drilling measurement data, and the collaborative optimization objective function is used to evaluate and score each candidate evolution trajectory.
[0111] The candidate evolutionary trajectory with the highest evaluation score is selected as the target regression path, and a correction evolution operator is generated based on the control parameters corresponding to the target regression path.
[0112] Specifically, after entering the path decision-making stage, when the system determines the timing for triggering a correction based on the aforementioned manifold characteristics, it initiates the trajectory evolution model for in-depth path planning and parameter solving. The core logic of this process lies in using historical patterns to predict future trends and finding the optimal regression path through multi-objective optimization. The system first constructs the trajectory evolution model using a bidirectional gated recurrent neural network. A bidirectional gated recurrent neural network is a deep learning architecture with temporal memory capabilities. It processes sequential information simultaneously through two recursive layers, forward and backward, effectively capturing the correlation between wellbore trajectories over time.
[0113] In the specific operational process, the system uses the historical pose sequence corresponding to the drilling rig measurement data as the basic input. This sequence includes the changes in the drill bit's coordinates, inclination, and azimuth over a past period. Simultaneously, the system uses the real-time characteristic parameters corresponding to the lateral power flow model as dynamic constraint input. The trajectory evolution model simulates the drill string's motion response under different guiding forces by performing a nonlinear mapping on the input data, thereby generating multiple sets of candidate evolution trajectories. Each set of candidate evolution trajectories represents a possible correction path scheme, encompassing the spatial evolution process from the current position to the target point.
[0114] To select the optimal path from numerous options, the system establishes a collaborative optimization objective function based on energy loss in the lateral power flow model and the geometric distance corresponding to the drilling measurement data. The energy loss term constrains the severity of the correction action, preventing excessive dogleg deviation from causing drill string fatigue; the geometric distance term constrains the accuracy of the regression path. This collaborative optimization objective function is used to evaluate and score each candidate evolution trajectory, with the scoring process comprehensively balancing the timeliness and safety of the correction. Collaborative Optimization Objective Function The calculation logic is as follows:
[0115] ;
[0116] In this formula, This indicates the evaluation and scoring results. This represents the Euclidean distance between the candidate evolution trajectory and the design target line. This indicates the expected energy loss corresponding to this trajectory. and These represent the distance weighting factor and the energy weighting factor, respectively. In this scheme, the distance weighting factor... The preferred value is 0.7, energy weighting factor. The preferred value is 0.3.
[0117] After scoring, the system selects the candidate evolution trajectory with the highest evaluation score as the target regression path. This path is considered the optimal solution under the current formation constraints and mechanical limits, which can smoothly revert to the design trajectory while minimizing energy waste. Subsequently, the system generates a correction evolution operator based on the control parameters corresponding to the target regression path. The correction evolution operator is a set of parameters containing a sequence of future execution steps, which defines the attitude adjustment step size and deflection vector direction of the drill bit within the evolution cycle. Through this decision-making method based on multi-path evolution, the system transforms the complex geometric correction problem into a state optimization problem of a dynamic system, achieving an intelligent leap from "instantaneous remediation" to "evolution-oriented" approaches.
[0118] S5. The shadow model runs synchronously between the ground terminal and the downhole node. The ground terminal converts the deviation evolution operator into an incremental correction offset and transmits the correction offset to the downhole node through the communication link.
[0119] Furthermore, in S5, the synchronous operation of the shadow model between the ground terminal and the downhole node includes:
[0120] Neural network models with consistent initial parameters are pre-set as shadow models in both the ground terminal and the downhole node, and a unified causal evolution state space is established for the shadow models.
[0121] The ground terminal acquires the correction evolution operator in real time, calculates the parameter increment of the correction evolution operator relative to the current state of the shadow model, and encapsulates the parameter increment as a correction offset for transmission.
[0122] After receiving the corrected offset, the downhole node uses the corrected offset to perform local parameter reconstruction on its locally running shadow model.
[0123] Specifically, in the communication and coordination phase of this scheme, the system establishes an efficient information transmission mechanism between the ground terminal and the downhole node, overcoming the limited communication bandwidth in the well through shadow model synchronization technology. During the system initialization phase, both the ground terminal and the downhole node are pre-set with identical initial parameters, forming the shadow model system. To ensure synchronized evolution of the models at both ends, the system establishes a unified causal evolution state space for the shadow model. This space defines all possible trajectory evolution paths and their corresponding dynamic constraints, ensuring that the understanding of the drilling state by both the ground and downhole nodes is based on the same mathematical benchmark.
[0124] In actual operation, the ground terminal acquires the correction evolution operator generated by the decision logic in real time. Since this operator contains a complex path prediction sequence, direct transmission would consume significant communication resources. Therefore, the ground terminal performs incremental processing, calculating the parameter increment of the correction evolution operator relative to the current internal state of the shadow model. This parameter increment reflects the minimum weight adjustments or state updates required by the shadow model to achieve the predetermined regression path. The ground terminal then encapsulates this parameter increment as a correction offset and transmits it to the downhole node via mud pulse communication or electromagnetic wave communication links.
[0125] The calculation logic for correcting the offset is implemented through state bias mapping, and its mathematical expression is as follows:
[0126] ;
[0127] In this formula, This represents the parameter increment before encapsulation. This represents the target model state vector corresponding to the bias correction evolution operator. This represents the real-time state vector of the shadow model on the ground. In this scheme, the numerical precision of the parameter increments is preferably floating-point to ensure the accuracy of the reconstruction.
[0128] Upon receiving the corrected offset, the downhole node does not need to recalculate complex prediction logic. Instead, it directly uses the corrected offset to locally reconstruct the parameters of its locally running shadow model. This process is accomplished by superimposing the corrected offset onto the weights or bias terms of the corresponding neurons in the downhole shadow model, instantly updating the downhole shadow model to the optimal evolutionary state consistent with the surface terminal. The parameter-reconstructed downhole shadow model can autonomously deduce specific execution instructions within the causal evolutionary state space based on local real-time sensing data. Through this parameter offset-based synchronization method, the system achieves high-frequency alignment between surface decision-making intentions and downhole execution actions.
[0129] S6. The downhole node receives the correction offset and performs self-calibration processing on the correction offset in combination with the real-time monitored near-drill bit pose data to generate an execution command, which drives the downhole actuator to perform trajectory correction actions.
[0130] Furthermore, in S6, the correction offset is self-calibrated based on real-time monitored near-bit pose data to generate execution instructions, including:
[0131] The measured pose vector is constructed using near-drill bit pose data, and the theoretical pose vector predicted by the shadow model at the current moment is retrieved.
[0132] Calculate the spatial residual between the measured pose vector and the theoretical pose vector, and input the spatial residual into the preset Kalman filter calibrator to calculate the real-time gain compensation coefficient for the prediction deviation of the shadow model.
[0133] The received correction offset is weighted and corrected using a real-time gain compensation coefficient to generate an execution instruction.
[0134] Specifically, during the downhole correction execution phase, the downhole node, as the core of edge computing, achieves high-precision command output by integrating received remote decisions with local real-time sensing data in a closed loop. When the downhole node receives the correction offset via the communication link, the system does not directly convert it into mechanical action but instead enters a self-calibration process. The core of this process is to eliminate the deviation between the theoretical model and the complex underground operating conditions. The system first uses a high-precision inclinometer and inertial navigation components integrated on the drill string assembly to acquire real-time near-bit pose data. This pose data covers the current instantaneous well inclination, azimuth, and tool face angle. Based on this dynamic data, the downhole node constructs a measured pose vector, which represents the actual physical coordinates and orientation of the drill bit in three-dimensional space.
[0135] Simultaneously, the downhole node retrieves the theoretical pose vector predicted by the locally running shadow model at the current moment. This theoretical pose vector represents the position the drill bit should be in according to the predetermined correction scheme under the assumption of ideal formation. The system then calculates the spatial residual between the measured pose vector and the theoretical pose vector. The spatial residual reflects the predicted deviation of the trajectory caused by unknown disturbances such as formation heterogeneity and drill string mechanical deformation. To achieve scientific compensation for such random disturbances, the system inputs the calculated spatial residual into a preset Kalman filter calibrator. The Kalman filter calibrator is an optimal estimation algorithm based on the minimum mean square error criterion, which extracts the changing trend of system error from the noisy residual signal through continuous iteration of state prediction and quantization feedback.
[0136] Using the output of the Kalman filter calibrator, the system calculates the real-time gain compensation coefficients for the prediction bias of the shadow model. These real-time gain compensation coefficients are derived from the mapping relationship between the residual covariance matrix and the observation matrix, and their mathematical expression is as follows: In this formula, This represents the real-time gain compensation coefficient. This represents the prediction error covariance matrix. Represents the observation mapping matrix, This represents the sensor observation noise covariance. In this scheme, the observation noise covariance... Preferably, it is a constant matrix set according to the sensor calibration accuracy.
[0137] After obtaining the real-time gain compensation coefficient, the downhole node uses this coefficient to perform weighted correction on the received correction offset. The weighted correction process essentially involves linearly fusing the macroscopic evolution trend transmitted from the surface with the microscopic errors captured downhole; its correction logic is as follows: In this formula, This represents the control component in the final generated execution instructions. This represents the original weight component in the corrected offset. This represents the aforementioned spatial residual.
[0138] Through the calibration process described above, the system generates execution commands after eliminating model drift errors. These commands are then translated into specific action signals for the downhole actuators, such as adjusting the deflection pressure of the rotary steering system or changing the extension displacement of the thrust vanes. The actuators then execute trajectory correction actions according to the commands, changing the cutting direction of the drill bit to bring the actual drilling trajectory closer to the target return path. This self-calibration mechanism based on real-time residual feedback ensures that the correction actions can dynamically adapt to the changing downhole environmental conditions.
[0139] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A trajectory correction method based on drilling survey data, characterized in that, include: S1. Real-time acquisition of drilling survey data, surface drilling parameters, and near-bit high-frequency vibration signals, which are then used as input parameters into the drilling system energy transfer model to construct a lateral power flow model; S2. Use the blind source separation algorithm to perform feature decomposition on the lateral power flow model to extract the environmental disturbance term and the controlled deflection term; S3. Map the environmental disturbance term, the controlled deflection term, and the drilling survey data to a high-dimensional feature space, extract the feature manifold through a manifold learning algorithm, and determine the timing of the correction trigger based on the feature manifold. S4. When the correction triggering time is reached, the trajectory evolution model is used to solve the problem with the energy loss in the lateral power flow model and the geometric distance corresponding to the drilling survey data as the objective function, and the correction evolution operator is obtained. S5. The shadow model is run synchronously between the ground terminal and the downhole node. The ground terminal converts the deviation evolution operator into an incremental correction offset and transmits the correction offset to the downhole node through the communication link. S6. The downhole node receives the correction offset and performs self-calibration processing on the correction offset in combination with the real-time monitored near-drill bit pose data to generate an execution command, driving the downhole actuator to perform trajectory correction actions.
2. The trajectory correction method based on drilling survey data according to claim 1, characterized in that, In S1, the energy transfer model of the drilling system includes: The total input mechanical work of the drilling system is calculated based on the surface drilling parameters, and the spatial force configuration of the drilling system is constructed by combining the drilling survey data. By utilizing the spatial force configuration, the total input mechanical work loss is analyzed, and the axial propulsion loss, tubing rotation torsion loss, and friction loss generated by the contact between the drill string and the well wall are extracted during the drilling process. The real-time total effective energy flow transmitted to the drill bit is determined by subtracting the axial propulsion loss, the torsional loss of the tubing rotation, and the friction loss from the total input mechanical work.
3. The trajectory correction method based on drilling survey data according to claim 1, characterized in that, In S1, the construction of the lateral power flow model includes: The lateral displacement field of the drill bit at the bottom plane is identified by using the high-frequency vibration signal near the drill bit, and the lateral virtual work component of the drill bit at the bottom plane is calculated by combining the total effective energy flow. Obtain the bending strain characteristics of the drilling system near the drill bit, and calculate the lateral bending strain energy components of the drilling system on the correction trajectory; A lateral power flow model is constructed by vector superposition of the lateral virtual work component and the lateral bending strain energy component.
4. The trajectory correction method based on drilling survey data according to claim 1, characterized in that, In S2, the eigenvalue decomposition of the lateral power flow model using the blind source separation algorithm includes: The lateral power flow model is resampled in the time dimension to construct the observation signal matrix; Using the independent component analysis algorithm as the benchmark algorithm for blind source separation, the observed signal matrix is de-aliased to determine a set of statistically independent source signal components. By performing spectral analysis on each source signal component, the frequency characteristics and energy density corresponding to each source signal component are identified, and the feature decomposition of the lateral power flow model is completed.
5. The trajectory correction method based on drilling survey data according to claim 1, characterized in that, In S2, the extraction of the environmental disturbance term and the controlled deflection term includes: A preset feature library is established, which includes the operating frequency modes of the actuator and the disturbance feature modes of the formation anisotropy; The decomposed source signal components are matched with a preset feature library, and the source signal components that conform to the action frequency mode are reconstructed into controlled deflection terms. The source signal components that conform to the disturbance characteristic mode and whose frequency distribution is within a preset random bandwidth are reconstructed into environmental disturbance terms.
6. The trajectory correction method based on drilling survey data according to claim 1, characterized in that, In S3, the extraction of feature manifolds via manifold learning algorithms includes: By fusing environmental disturbance terms, controlled deflection terms, and drilling survey data with time-series features, a high-dimensional phase space matrix is constructed. The geodesic distance between each sampling point in the high-dimensional phase space matrix is calculated using the equidistant mapping algorithm to generate a global distance matrix. Nonlinear dimensionality reduction is performed on the global distance matrix to find low-dimensional structures embedded in the high-dimensional phase space and extract feature manifolds that reflect the trajectory evolution law.
7. The trajectory correction method based on drilling survey data according to claim 1, characterized in that, In S3, determining the correction trigger timing based on the characteristic manifold includes: Real-time monitoring of the motion trajectory of the characteristic manifold in low-dimensional space, and calculation of the instantaneous curvature of the motion trajectory and the diffusion velocity of the region deviating from the design manifold; Establish a correction evaluation function, wherein the instantaneous curvature is a variable; When the instantaneous curvature exceeds the preset curvature threshold, or the diffusion speed exceeds the preset trend threshold, the current moment is determined as the timing for correction triggering.
8. The trajectory correction method based on drilling survey data according to claim 1, characterized in that, In S4, the objective function is to solve for the energy loss in the lateral power flow model and the geometric distance corresponding to the drilling survey data, including: A trajectory evolution model is constructed by a bidirectional gated recurrent neural network, and the historical pose sequence corresponding to the drilling survey data and the real-time feature parameters corresponding to the lateral power flow model are input into the trajectory evolution model to generate multiple sets of candidate evolution trajectories. A collaborative optimization objective function is established based on the energy loss in the lateral power flow model and the geometric distance corresponding to the drilling measurement data, and the collaborative optimization objective function is used to evaluate and score each candidate evolution trajectory. The candidate evolutionary trajectory with the highest evaluation score is selected as the target regression path, and a correction evolution operator is generated based on the control parameters corresponding to the target regression path.
9. The trajectory correction method based on drilling survey data according to claim 1, characterized in that, In S5, the synchronous operation of the shadow model between the ground terminal and the downhole node includes: Neural network models with consistent initial parameters are pre-set as shadow models in both the ground terminal and the downhole node, and a unified causal evolution state space is established for the shadow models. The ground terminal acquires the correction evolution operator in real time, calculates the parameter increment of the correction evolution operator relative to the current state of the shadow model, and encapsulates the parameter increment as a correction offset for transmission. After receiving the corrected offset, the downhole node uses the corrected offset to perform local parameter reconstruction on its locally running shadow model.
10. The trajectory correction method based on drilling survey data according to claim 1, characterized in that, In S6, the step of performing self-calibration processing on the correction offset based on real-time monitored near-drill bit pose data to generate execution instructions includes: The measured pose vector is constructed using the near-drill bit pose data, and the theoretical pose vector predicted by the shadow model at the current moment is retrieved. Calculate the spatial residual between the measured pose vector and the theoretical pose vector, and input the spatial residual into a preset Kalman filter calibrator to calculate the real-time gain compensation coefficient for the prediction bias of the shadow model. The received correction offset is weighted and corrected using a real-time gain compensation coefficient to generate an execution instruction.