A geological steering trajectory automatic correction and optimization system

Abstract

This invention discloses an automatic correction and optimization system for geological steering trajectories, relating to the field of geological steering control technology. It includes: a multi-scale boundary perception and prior calibration module, which acquires drilling-while-drilling azimuth gamma, azimuth resistivity, and seismic prior data to construct a posterior boundary distribution, compares the boundary position in the drilled section, and calibrates the seismic prior deviation to correct the forward distribution; a probabilistic boundary rolling pre-simulation module, which pre-simulates the probabilistic boundary position at each depth and calculates the dynamic collision probability to generate an evolutionary map; a risk-coupled correction decision module, which solves the correction sequence for tool facet angle and drilling speed under mechanical constraints based on the evolutionary map; and a pre-simulation deviation-driven model evolution module, which compares the actual drilling trajectory with the probabilistic boundary position to calculate the pre-simulation deviation, and transmits back the correction calibration coefficients and layered prior parameters to achieve synchronous closed-loop evolution of the geological model and correction decisions. This invention solves the systemic bottleneck of traditional deterministic boundary interpretation and static control in geological steering.

Description

Technical Field

[0001] This invention relates to the field of geological guidance control technology, specifically to an automatic correction and optimization system for geological guidance trajectories. Background Technology

[0002] Geological steering of horizontal wells uses multi-scale geophysical data to control the drill bit trajectory in real time. The accuracy of formation boundary identification, risk quantification capability and real-time correction directly determine the reservoir encounter rate and the success or failure of the operation.

[0003] The existing system adopts the "multi-source data fusion - deterministic boundary interpretation - threshold-triggered correction" approach: weighted fusion of near-scale gamma, mesoscale resistivity and far-scale seismic data, determination of single boundary location based on deterministic inversion, setting a fixed threshold based on geometric distance to trigger toolface adjustment, and correcting the model through offline comparison with actual drilling data;

[0004] This approach has proven successful in homogeneous reservoirs, but it suffers from systemic limitations in heterogeneous formations: multi-scale fusion fails to utilize near-scale data from drilled well sections to online calibrate far-scale seismic priors, leading to an underestimation of forward uncertainties; boundary identification outputs deterministic single-point locations without probabilistic representations or confidence intervals, resulting in a lack of quantitative uncertainty for subsequent control; trajectory correction employs fixed thresholds and on / off strategies, which cannot adapt to non-stationary formation evolution, and model updates and correction execution are time-separated; there is a lack of quantitative assessment of historical decision-making omission and mis-adjustment rates, making it impossible to adaptively adjust risk classification thresholds; the core defects are: deterministic boundary interpretation fails to form a probabilistic uncertainty transmission chain, static threshold control fails to establish a rolling evolution collision probability map, and geological-engineering decision-making is fragmented in the spatiotemporal dimensions;

[0005] There is an urgent need for an automatic correction and optimization scheme for geological guidance trajectories that integrates multi-scale uncertainty quantification, probabilistic boundary rolling pre-simulation, and adaptive evolution of pre-simulation deviation. Summary of the Invention

[0006] To address the aforementioned technical problems, an automatic correction and optimization system for geological guidance trajectories is provided. This technical solution solves the problems of the lack of a probabilistic uncertainty transmission chain in the interpretation of deterministic boundaries, the lack of a rolling evolution collision probability map in static threshold control, and the separation of geological-engineering decision-making in the spatiotemporal dimensions.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] An automatic correction and optimization system for geological guidance trajectories includes:

[0009] The multi-scale boundary sensing and prior calibration module acquires drilling azimuth gamma, azimuth resistivity and seismic prior data according to the detection depth, and constructs the posterior distribution of formation boundaries and the stratification prior parameters. The near-scale azimuth gamma and far-scale seismic prior data are compared with the boundary positions in the drilled well section. If the deviation exceeds the preset verification threshold, the systematic deviation of the seismic prior data is calibrated with azimuth gamma and calibration coefficients are generated. The calibration coefficients are used to correct the posterior distribution of boundaries and the stratification prior parameters in the un-drilled well section ahead.

[0010] The probabilistic boundary rolling pre-simulation module, based on the boundary posterior distribution, performs probability propagation and deduction of the trajectory of the drill bit for multiple preset steps in the future, pre-simulates the probabilistic boundary position at each depth point by point, calculates the dynamic collision probability between the probabilistic boundary position and the trajectory, and generates a collision probability evolution map distributed along the drilling depth.

[0011] The risk coupling correction decision module, based on the collision probability evolution map, takes avoiding high collision probability as the optimization objective, and solves the correction sequence of tool face angle and drilling speed under the constraints of well inclination angle and dogleg degree, so that the drill bit avoids the probabilistic boundary position and satisfies mechanical constraints.

[0012] The model evolution module driven by the pre-drilling deviation compares the actual drilling trajectory with the probabilistic boundary position at the corresponding depth, calculates the pre-drilling deviation, and sends it back to the multi-scale boundary perception and prior calibration module to correct the calibration coefficients and hierarchical prior parameters with the weighted correction of the pre-drilling deviation.

[0013] Preferably, in the multi-scale boundary sensing and prior calibration module, the detection depth of the drilling azimuth gamma data is 0.5m to 3m, the detection depth of the azimuth resistivity data is 3m to 10m, and the detection depth of the far-scale seismic prior data is greater than 10m; before constructing the posterior distribution of the formation boundary, the multi-scale boundary sensing and prior calibration module performs depth registration and consistency verification on the drilling azimuth gamma, azimuth resistivity, and seismic prior data.

[0014] Preferably, in the multi-scale boundary perception and prior calibration module, the boundary position comparison uses a cross-correlation algorithm to solve the position deviation; the preset verification threshold is dynamically set according to the statistical variance of the historical comparison deviation of the drilled well section, and the larger the statistical variance, the wider the preset verification threshold.

[0015] Preferably, in the multi-scale boundary sensing and prior calibration module, the systematic bias of the seismic prior data is calibrated using azimuth gamma and calibration coefficients are generated, including:

[0016] The calibration coefficients are generated by solving the linear scaling factor and depth offset through least squares fitting of the systematic bias.

[0017] The boundary location information in the seismic prior data of the un-drilled well section is corrected with calibration coefficients. The corrected seismic prior data, along with the drilling azimuth gamma and azimuth resistivity, are then re-input into the variational Bayesian inference framework to update the hyperparameters of the layered prior parameters and re-infer the boundary posterior distribution.

[0018] Preferably, in the probability boundary rolling pre-simulation module, the preset step size and probability propagation deduction method include:

[0019] The preset step size is determined by the combined constraints of the current dogleg degree of the drill bit and the tool face angle change rate. The closer either parameter of the dogleg degree or the tool face angle change rate is to the allowable boundary of the well inclination angle and dogleg degree constraints, the shorter the preset step size will be.

[0020] The time domain of the probability propagation deduction is measured by the drilling depth, covering a well section from 15m to 30m in front of the drill bit. The arrangement density of each preset step size node in the time domain is non-uniformly distributed according to the variance of the boundary posterior distribution. The higher the variance, the higher the node density in the depth interval.

[0021] Preferably, in the probabilistic boundary rolling pre-simulation module, the probabilistic boundary position at each depth is pre-simulated point by point, including:

[0022] Using unscented transformation propagation, the mean and covariance of the boundary posterior distribution are extracted. Based on the mean and covariance, a set of sigma sampling points is selected. Each sigma sampling point is propagated along the drilling direction to a depth node of a preset step size. At each depth node, the mean and covariance of the probabilistic boundary position are reconstructed.

[0023] Preferably, in the probabilistic boundary rolling pre-simulation module, calculating the dynamic collision probability between the probabilistic boundary position and the trajectory includes:

[0024] Calculate the shortest distance from the trajectory to the probabilistic boundary position. Based on the covariance matrix between the shortest distance and the probabilistic boundary position, calculate the failure probability value of the trajectory crossing the probabilistic boundary position, which is used as the dynamic collision probability.

[0025] Preferably, in the probabilistic boundary rolling pre-simulation module, generating a collision probability evolution map distributed along the drilling depth includes:

[0026] The collision probability evolution map generates a one-dimensional probability curve with drilling depth as the x-axis and dynamic collision probability as the y-axis;

[0027] The collision probability evolution map also includes a confidence envelope of the dynamic collision probability, which is superimposed with an uncertainty bandwidth on the dynamic collision probability based on the quantiles of the boundary posterior distribution.

[0028] The uncertainty bandwidth widens at depths where the variance of the boundary posterior distribution increases.

[0029] Preferably, in the risk coupling correction decision module, the correction sequence solution method includes:

[0030] Model predictive control is adopted, and the correction sequence of tool face angle and drilling speed is solved by rolling under the constraints of well inclination angle and dogleg degree. The prediction time domain of model predictive control is synchronized with the preset step size of the probability boundary rolling pre-simulation module.

[0031] The solution to the correction sequence also incorporates the tool face angle change rate constraint as an additional mechanical constraint.

[0032] Preferably, the method for calculating and weighting the pre-exercise deviation in the pre-exercise deviation-driven model evolution module includes:

[0033] The actual drilling trajectory is compared with the probabilistic boundary position at the corresponding depth, and the shortest distance between the actual drilling trajectory and the probabilistic boundary position is calculated as the pre-drilling deviation.

[0034] The calibration coefficients and stratified prior parameters of the un-drilled well section ahead are corrected by weighted adjustment of the pre-drilling deviation. An exponentially weighted moving average rule is adopted, in which the weight of the pre-drilling deviation decreases exponentially with the increase of the depth of the un-drilled well section ahead.

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0036] This invention proposes a multi-scale boundary perception and prior calibration module, which calibrates the systematic bias of far-scale seismic priors using near-scale data from drilled well sections and corrects the posterior distribution of the forward boundary, thereby reducing the uncertainty of forward boundary prediction online. It also proposes a probabilistic boundary rolling pre-simulation module, which pre-simulates the probabilistic boundary positions at each depth point by point through probability propagation and generates a collision probability evolution map, realizing the transformation of trajectory risk from a static threshold to a probabilistic quantitative understanding of rolling evolution along depth. Furthermore, it proposes a risk coupling correction decision module, which solves the correction sequence under mechanical constraints based on the collision probability evolution map, achieving direct coupling and continuous optimization of geological risk evolution and engineering control decisions in the drilling time domain. Finally, it proposes a pre-simulation bias-driven model evolution module, which corrects calibration coefficients and layered prior parameters by comparing the actual drilling trajectory with the probabilistic boundary positions, achieving synchronous closed-loop evolution of the geological model and correction decisions. Attached Figure Description

[0037] Figure 1 This is a system framework diagram of the present invention. Detailed Implementation

[0038] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0039] Reference Figure 1As shown, an automatic correction and optimization system for geological guidance trajectories includes:

[0040] The multi-scale boundary sensing and prior calibration module acquires drilling azimuth gamma, azimuth resistivity and seismic prior data according to the detection depth, and constructs the posterior distribution of formation boundaries and the stratification prior parameters. The near-scale azimuth gamma and far-scale seismic prior data are compared with the boundary positions in the drilled well section. If the deviation exceeds the preset verification threshold, the systematic deviation of the seismic prior data is calibrated with azimuth gamma and calibration coefficients are generated. The calibration coefficients are used to correct the posterior distribution of boundaries and the stratification prior parameters in the un-drilled well section ahead.

[0041] In the multi-scale boundary sensing and prior calibration module, the detection depth of the drilling azimuth gamma data is 0.5m to 3m, the detection depth of the azimuth resistivity data is 3m to 10m, and the detection depth of the far-scale seismic prior data is greater than 10m. Before constructing the posterior distribution of the formation boundary, the multi-scale boundary sensing and prior calibration module performs depth registration and consistency verification on the drilling azimuth gamma, azimuth resistivity, and seismic prior data.

[0042] The physical basis for probing depth using three-scale data is as follows:

[0043] The half-amplitude point of the spatial response function of the azimuth gamma probe during drilling corresponds to a radial depth of approximately 0.5 m, and the 90% signal contribution radius does not exceed 3 m. Therefore, 0.5 m to 3 m is considered the effective detection depth at the near-scale. The azimuth resistivity instrument operates at a frequency of 400 kHz–2 MHz. The propagation depth of electromagnetic waves under conventional reservoir resistivity conditions is 3 m to 10 m, which is considered the effective detection depth at the mesoscale. The wavelength resolution of well-side seismic or VSP data corresponds to a formation interface depth scale greater than 10 m, which is considered the effective detection depth at the far-scale.

[0044] The coordinate system for depth registration is as follows:

[0045] The drilling azimuth gamma, azimuth resistivity, and seismic prior data are unified into the same vertical depth coordinate system. Specifically, based on the depth, inclination, and azimuth sequences recorded in the well logging, the wellbore trajectory is calculated using the minimum curvature method, establishing the conversion relationship between depth measurements and vertical depth. After time-depth conversion, the seismic prior data is transformed from the elevation datum to the same vertical depth datum as the well logging data, completing the depth coordinate alignment of the three scales of data.

[0046] The consistency check is as follows:

[0047] Within the overlapping depth range after depth registration, the Pearson correlation coefficient and variance ratio of the three-scale data are calculated. The judgment threshold is set as a correlation coefficient greater than 0.8 and a variance ratio between 0.67 and 1.5. If the data in a certain depth segment does not meet the above threshold, it is determined that there is an inconsistency between scales.

[0048] The strategy for handling inconsistent depth segments is as follows: If the segment is located in an already drilled section, the near-scale azimuth gamma data is used as the benchmark, and the depth domain is fine-tuned and the amplitude is corrected for the azimuth resistivity data. If the segment is located in an un-drilled section ahead, the fusion weight of the far-scale seismic prior data in the variational Bayesian inference framework is reduced. The weight reduction coefficient is linearly set to 0.5 to 0.8 according to the degree of inconsistency, so as to avoid low-quality prior data from contaminating the posterior distribution of the ahead boundary.

[0049] In the multi-scale boundary sensing and prior calibration module, the boundary position comparison uses a cross-correlation algorithm to solve for position deviations. The preset verification threshold is dynamically set based on the statistical variance of historical comparison deviations in the drilled well section. The larger the statistical variance, the wider the preset verification threshold. The statistical variance is calculated based on a sliding window of historical comparison deviations within the drilled well section, with a window length of 50 meters and a step size of 1 meter. The preset verification threshold is set according to the principle that the threshold equals the safety factor multiplied by the square root of the statistical variance. The safety factor is selected between 1.5 and 2.5 based on formation stability.

[0050] The multi-scale boundary sensing and prior calibration module uses azimuth gamma calibration to determine the systematic bias of seismic prior data and generates calibration coefficients, including:

[0051] The calibration coefficients are generated by solving the linear scaling factor and depth offset through least squares fitting of the systematic bias.

[0052] Systematic bias sequence sample construction: Within the drilled well section, a paired sample of the near-scale azimuth gamma measured boundary depth and the far-scale seismic prior predicted boundary depth is extracted at 0.5m depth intervals. The difference between the two is calculated as the systematic bias sample value for that depth point. The systematic bias sequence is constructed by traversing all depth nodes in the drilled well section.

[0053] The boundary location information in the seismic prior data of the un-drilled well section is corrected using calibration coefficients. The corrected seismic prior data, along with the drilling azimuth gamma and azimuth resistivity data, are re-input into the variational Bayesian inference framework to update the hyperparameters of the layered prior parameters and re-infer the boundary posterior distribution. The variational Bayesian inference framework adopts a three-layer layered prior structure: the first layer is the near-scale azimuth gamma data layer, set as a Gaussian prior, with initial hyperparameters including the boundary location mean and accuracy; the second layer is the mesoscale azimuth resistivity data layer, set as a Gaussian prior, with initial hyperparameters including the boundary location mean and accuracy; the third layer is the far-scale seismic prior data layer, set as a range-constrained prior, with initial hyperparameters including the boundary trend mean and trend range width. Each layer of prior is coupled through a shared parent node. The hyperparameters of the parent node are updated online by the hard constraint data of the drilled well section, and the updated parent node hyperparameters re-constrain the boundary posterior distribution of the un-drilled well section.

[0054] Posterior update of hyperparameters: After correcting the boundary location information in the seismic prior data of the un-drilled well section with the calibration coefficients, the corrected seismic prior data, along with the drilling azimuth gamma and azimuth resistivity, are re-input into the variational Bayesian inference framework. In the variational Bayesian inference framework, using the calibration coefficients as observation variables, the hyperparameters of the hierarchical prior parameters are updated by maximizing the lower bound of evidence. Iterative optimization is performed using gradient descent with a step size set to 0.01 to 0.05. Updates stop when the change in the lower bound of evidence between two adjacent iterations is less than a preset convergence threshold, which is set to 0.0001. The updated hyperparameters then re-constrain the posterior boundary distribution of the un-drilled well section.

[0055] The probabilistic boundary rolling pre-simulation module, based on the boundary posterior distribution, performs probability propagation and deduction of the trajectory of the drill bit for multiple preset steps in the future, pre-simulates the probabilistic boundary position at each depth point by point, calculates the dynamic collision probability between the probabilistic boundary position and the trajectory, and generates a collision probability evolution map distributed along the drilling depth.

[0056] The probability boundary rolling pre-simulation module includes a preset step size and probability propagation deduction method, comprising:

[0057] The preset step size is determined by the combined constraints of the current dogleg degree of the drill bit and the tool face angle change rate. The closer either the dogleg degree or the tool face angle change rate is to the allowable boundary of the well inclination angle and dogleg degree constraints, the shorter the preset step size. The preset step size is determined by looking up a table of dogleg degree and tool face angle change rate. The table stores the step size values ​​corresponding to different dogleg degree intervals and tool face angle change rate intervals in advance. The closer the dogleg degree or tool face angle change rate is to the allowable boundary, the smaller the corresponding step size value.

[0058] The step size lookup method is represented as follows: a dogleg degree of 0–3° / 30m and a tool face angle change rate of 0–5° / 30m correspond to a step size of 5m; a dogleg degree of 3–5° / 30m or a tool face angle change rate of 5–10° / 30m correspond to a step size of 3m; a dogleg degree >5° / 30m or a tool face angle change rate >10° / 30m correspond to a step size of 1m.

[0059] The time domain of the probability propagation simulation is measured in terms of drilling depth, covering a well section 15m to 30m in front of the drill bit. Within the time domain, the node density for each preset step size is non-uniformly distributed based on the variance of the boundary posterior distribution. The higher the variance, the higher the node density in depth intervals. Specifically, the node density is proportional to the square root of the variance; for every doubling of the variance, the node density increases by 50%. For example, in depth intervals where the variance is at the baseline level, the node density is 2 nodes per meter; in depth intervals where the variance increases to 4 times the baseline level, the node density increases accordingly to 4 nodes per meter. The upper limit of the node density is set at 10 nodes per meter, and the lower limit is set at 2 nodes per meter, ensuring that the computational load of the probability propagation simulation does not exceed the real-time processing capability of the downhole embedded system, while ensuring sufficient pre-simulation resolution in areas with large variance.

[0060] In the probabilistic boundary rolling pre-simulation module, the probabilistic boundary positions at each depth are pre-simulated point by point, including:

[0061] Using unscented transformation propagation, the mean and covariance of the boundary posterior distribution are extracted. Based on the mean and covariance, a set of sigma sampling points is selected. Each sigma sampling point is propagated along the drilling direction to a depth node of a preset step size. At each depth node, the mean and covariance of the probabilistic boundary position are reconstructed.

[0062] The sigma sampling point set for unscented transform propagation is selected as follows:

[0063] In one embodiment, the unscented transformation propagation selects a deterministic sigma sampling point set based on the dimension of the boundary posterior distribution. The dimension is determined by the number of state variables at the probabilistic boundary location, including the three-dimensional spatial coordinates of the boundary points and the corresponding uncertainty parameters, with a dimension value ranging from 3 to 6. The number of sigma sampling points is twice the state dimension plus one, covering the mean point and sampling points symmetrically distributed along the principal direction of the covariance matrix. The sampling points are determined based on the square root decomposition results of the mean and the covariance matrix. The dispersion distance of the sampling points to the mean is adjusted by a scaling factor, which is adaptively selected based on the state dimension to ensure that the sampling point set covers the main uncertainty range of the boundary posterior distribution during nonlinear propagation.

[0064] The state transfer and propagation along the drilling direction are specifically as follows:

[0065] The state transition function comprises two parts: a formation interface extension model and a drill bit trajectory extrapolation model. The formation interface extension model projects the probabilistic boundary position of the current depth node along the tangential extension direction of the formation interface to the depth node of the next preset step size, based on the dip and dip angles of the drilled sections. The drill bit trajectory extrapolation model calculates the trajectory increment of the drill bit within the preset step size according to three-dimensional kinematic relationships, based on the current tool face angle, well inclination angle, and drilling speed. Each sigma sampling point is propagated from the current depth node to the depth node of the next preset step size through the combined action of the formation interface extension model and the drill bit trajectory extrapolation model.

[0066] The reconstruction of the mean and covariance at depth nodes is as follows:

[0067] At each depth node with a preset step size, the mean is calculated by weighting according to the spatial coordinates of each sigma sampling point after propagation and the weighting distribution rule of the unscented transformation propagation, and is used as the mean estimate of the probabilistic boundary position at that depth node; the covariance matrix is ​​calculated by weighting according to the scatter matrix of each sigma sampling point relative to the mean, and is used as the uncertainty estimate of the probabilistic boundary position at that depth node, thus completing the point-by-point pre-playback.

[0068] In the probabilistic boundary rolling pre-simulation module, the dynamic collision probability of the probabilistic boundary position and trajectory is calculated, including:

[0069] Calculate the shortest distance from the trajectory to the probabilistic boundary position. Based on the covariance matrix between the shortest distance and the probabilistic boundary position, calculate the failure probability value of the trajectory crossing the probabilistic boundary position, which is used as the dynamic collision probability.

[0070] The calculation process of the dynamic collision probability is as follows: First, the shortest distance from the trajectory to the probabilistic boundary position is calculated. This shortest distance is defined as the normal distance rather than the Euclidean distance. That is, the local normal vector of the formation interface is extracted at the probabilistic boundary position, and the displacement of the drill bit trajectory extrapolated curve to the probabilistic boundary position is projected along the direction of the local normal vector. The resulting projection length is the shortest distance. This normal distance eliminates the interference of parallel displacement along the formation strike and directly represents the penetration tendency of the drill bit in the direction perpendicular to the formation interface. Then, the probability distribution of this shortest distance is derived, and the covariance matrix of the probabilistic boundary position is... By projecting the error propagation law onto the local normal vector direction, a one-dimensional variance estimate of the normal distance is obtained, which linearly maps the uncertainty of the three-dimensional boundary position to the uncertainty of the one-dimensional normal distance. This determines the probability distribution of the shortest distance with its mean as the midpoint and the one-dimensional variance as the degree of diffusion. Finally, the failure probability value is calculated, that is, the cumulative probability that the shortest distance is less than zero is calculated. This cumulative probability represents the tail probability that the drill bit trajectory crosses the probabilistic boundary position in the normal direction, which is used as the dynamic collision probability. The larger the probability value, the higher the failure risk of the drill bit trajectory crossing the formation boundary at the corresponding depth node.

[0071] The probabilistic boundary rolling pre-simulation module generates a collision probability evolution map distributed along the drilling depth, including:

[0072] The collision probability evolution map generates a one-dimensional probability curve with drilling depth as the x-axis and dynamic collision probability as the y-axis;

[0073] The collision probability evolution map also includes a confidence envelope of the dynamic collision probability, which is superimposed with an uncertainty bandwidth on the dynamic collision probability based on the quantiles of the boundary posterior distribution.

[0074] The uncertainty bandwidth widens at depths where the variance of the boundary posterior distribution increases.

[0075] The generation process of the collision probability evolution map is as follows: First, a one-dimensional data sequence is established, with the drilling depth as the horizontal axis index and the depth step size set to 0.1 meters. Each depth node stores the dynamic collision probability floating-point value at the corresponding position, forming a probability curve continuously distributed along the well depth direction. Then, a confidence envelope is generated based on the probability curve. The 5% and 95% quantiles are selected as the upper and lower envelope boundaries, or the mean plus or minus two standard deviations is used as the bandwidth range. The dynamic collision probabilities are superimposed with the positive uncertainty bandwidth and the negative uncertainty bandwidth to obtain the upper envelope and the lower envelope. The uncertainty bandwidth is determined by the local variance of the boundary posterior distribution at the current depth node. The larger the local variance, the wider the bandwidth. The widening of the bandwidth is proportional to the square root of the local variance. That is, when the formation uncertainty increases, the envelope adaptively widens according to the multiple relationship of the square root of the variance, thereby forming a wider collision probability confidence interval at the formation abrupt change zone, indicating that the reliability of geological risk perception within this depth interval is reduced.

[0076] The risk coupling correction decision module, based on the collision probability evolution map, takes avoiding high collision probability as the optimization objective, and solves the correction sequence of tool face angle and drilling speed under the constraints of well inclination angle and dogleg degree, so that the drill bit avoids the probabilistic boundary position and satisfies mechanical constraints.

[0077] The risk coupling correction decision module includes a correction sequence solution method comprising:

[0078] Model predictive control is adopted, and the correction sequence of tool face angle and drilling speed is solved by rolling under the constraints of well inclination angle and dogleg degree. The prediction time domain of model predictive control is synchronized with the preset step size of the probability boundary rolling pre-simulation module.

[0079] The solution to the correction sequence also incorporates the tool face angle change rate constraint as an additional mechanical constraint.

[0080] The correction sequence is solved using model predictive control (MMCC), which iteratively solves for the toolface angle and drilling speed correction sequence under the constraints of well inclination angle and dogleg degree. The objective function of the MMC consists of three weighted terms: the first term is a trajectory safety probability maximization term, whose weight is adaptively adjusted based on the dynamic collision probability at the current depth node; the higher the dynamic collision probability, the larger the weight of this term, in order to preferentially avoid high collision probability regions; the second term is a toolface angle change rate smoothing penalty term, to suppress severe toolface vibration; and the third term is a drilling speed change rate energy consumption penalty term, to reduce mechanical wear. The prediction time domain of the MMC is rolled along with the probability boundary. The pre-simulation module synchronizes the preset step size. Specifically, based on the current mechanical drilling speed of the drill bit, the preset step size in the depth dimension is converted into a control cycle in the time dimension. For example, when the mechanical drilling speed is 30 m / h, a 1 m depth step size corresponds to a 2 min control cycle, ensuring that the time window for geological risk simulation is strictly aligned with the time window for engineering control decision-making. During the solution process, the well inclination angle, dogleg degree, and tool face angle change rate constraints are uniformly constructed into a constraint matrix and gradient vector. A quadratic programming solver is used for iterative optimization. The solution is terminated when the change in the objective function between two adjacent iterations is less than a preset convergence threshold, and the optimal correction sequence of the tool face angle and drilling speed is output.

[0081] The model evolution module driven by the pre-drilling deviation compares the actual drilling trajectory with the probabilistic boundary position at the corresponding depth, calculates the pre-drilling deviation, and sends it back to the multi-scale boundary perception and prior calibration module to correct the calibration coefficients and hierarchical prior parameters with the weighted correction of the pre-drilling deviation.

[0082] The method for calculating and weighting the pre-exercise deviation in the model evolution module driven by the pre-exercise deviation includes:

[0083] The actual drilling trajectory is compared with the probabilistic boundary position at the corresponding depth, and the shortest distance between the actual drilling trajectory and the probabilistic boundary position is calculated as the pre-drilling deviation.

[0084] The calibration coefficients and stratified prior parameters of the un-drilled well section ahead are corrected by weighted adjustment of the pre-drilling deviation. An exponentially weighted moving average rule is adopted, in which the weight of the pre-drilling deviation decreases exponentially with the increase of the depth of the un-drilled well section ahead.

[0085] The specific process of calculating and weighting the pre-drilling deviation is as follows: First, the actual drilling trajectory after correction is compared with the probabilistic boundary position at the corresponding depth. The shortest distance between the two in the normal direction of the boundary is calculated as the pre-drilling deviation. It is agreed that the pre-drilling deviation is positive when the actual drilling trajectory is above the probabilistic boundary position and negative when it is below, to distinguish the direction in which the trajectory crosses the boundary. Then, the pre-drilling deviation sequence is smoothed using an exponentially weighted moving average rule. The pre-drilling deviation of the current depth node is weighted and fused with the historical cumulative deviation value using a smoothing coefficient of 0.3 to 0.5. The higher the proportion of the new deviation, the more sensitive the system is to recent pre-drilling errors. In the weighted correction... When setting the calibration coefficients and stratified prior parameters for the un-drilled section directly ahead, the weight of the pre-drilling deviation decreases exponentially with the increase of the un-drilled depth ahead. The weight decreases by about 10% for every meter increase in depth ahead, and the weights of each depth node are normalized to make the sum of one, ensuring that the contribution of the pre-drilling deviation of the distant well section ahead to the current correction is moderately weakened. The calibration coefficients and stratified prior parameters are updated according to different correction step sizes, wherein the correction step size for the linear proportional coefficient and depth offset is set to 0.05, and the correction step size for the hyperparameters of the stratified prior parameters is set to 0.01, to avoid excessive oscillation of the hyperparameters due to the fluctuation of the pre-drilling deviation, and to achieve synchronous closed-loop evolution of the geological model and the correction decision.

[0086] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. An automatic correction and optimization system for geological guidance trajectories, characterized in that, include: The multi-scale boundary sensing and prior calibration module acquires drilling-while-drilling azimuth gamma, azimuth resistivity and seismic prior data according to the detection depth, and constructs the posterior distribution of the stratigraphic boundary and the stratified prior parameters. The boundary positions of near-scale azimuth gamma and far-scale seismic prior data in the drilled well section are compared. If the deviation exceeds the preset verification threshold, the systematic deviation of the seismic prior data is calibrated with azimuth gamma and calibration coefficients are generated. The calibration coefficients are then used to correct the boundary posterior distribution and stratified prior parameters of the un-drilled well section ahead. The probabilistic boundary rolling pre-simulation module, based on the boundary posterior distribution, performs probability propagation and deduction of the trajectory of the drill bit for multiple preset steps in the future, pre-simulates the probabilistic boundary position at each depth point by point, calculates the dynamic collision probability between the probabilistic boundary position and the trajectory, and generates a collision probability evolution map distributed along the drilling depth. The risk coupling correction decision module, based on the collision probability evolution map, takes avoiding high collision probability as the optimization objective, and solves the correction sequence of tool face angle and drilling speed under the constraints of well inclination angle and dogleg degree, so that the drill bit avoids the probabilistic boundary position and satisfies mechanical constraints. The model evolution module driven by the pre-drilling deviation compares the actual drilling trajectory with the probabilistic boundary position at the corresponding depth, calculates the pre-drilling deviation, and sends it back to the multi-scale boundary perception and prior calibration module to correct the calibration coefficients and hierarchical prior parameters with the weighted correction of the pre-drilling deviation.

2. The automatic correction and optimization system for geological guidance trajectory according to claim 1, characterized in that, In the multi-scale boundary sensing and prior calibration module, the detection depth of the drilling azimuth gamma data is 0.5m to 3m, the detection depth of the azimuth resistivity data is 3m to 10m, and the detection depth of the far-scale seismic prior data is greater than 10m. Before constructing the posterior distribution of the formation boundary, the multi-scale boundary sensing and prior calibration module performs depth registration and consistency verification on the drilling azimuth gamma, azimuth resistivity, and seismic prior data.

3. The automatic correction and optimization system for geological guidance trajectory according to claim 1, characterized in that, In the multi-scale boundary perception and prior calibration module, the boundary position comparison uses a cross-correlation algorithm to solve the position deviation; the preset verification threshold is dynamically set according to the statistical variance of the historical comparison deviation of the drilled well section, and the larger the statistical variance, the wider the preset verification threshold.

4. The automatic correction and optimization system for geological guidance trajectory according to claim 1, characterized in that, The multi-scale boundary sensing and prior calibration module uses azimuth gamma calibration to determine the systematic bias of seismic prior data and generates calibration coefficients, including: The calibration coefficients are generated by solving the linear scaling factor and depth offset through least squares fitting of the systematic bias. The boundary location information in the seismic prior data of the un-drilled well section is corrected with calibration coefficients. The corrected seismic prior data, along with the drilling azimuth gamma and azimuth resistivity, are then re-input into the variational Bayesian inference framework to update the hyperparameters of the layered prior parameters and re-infer the boundary posterior distribution.

5. The automatic correction and optimization system for geological guidance trajectory according to claim 1, characterized in that, The probability boundary rolling pre-simulation module includes a preset step size and probability propagation deduction method, comprising: The preset step size is determined by the combined constraints of the current dogleg degree of the drill bit and the tool face angle change rate. The closer either parameter of the dogleg degree or the tool face angle change rate is to the allowable boundary of the well inclination angle and dogleg degree constraints, the shorter the preset step size will be. The time domain of the probability propagation deduction is measured by the drilling depth, covering a well section from 15m to 30m in front of the drill bit. The arrangement density of each preset step size node in the time domain is non-uniformly distributed according to the variance of the boundary posterior distribution. The higher the variance, the higher the node density in the depth interval.

6. The automatic correction and optimization system for geological guidance trajectory according to claim 1, characterized in that, In the probabilistic boundary rolling pre-simulation module, the probabilistic boundary positions at each depth are pre-simulated point by point, including: Using unscented transformation propagation, the mean and covariance of the boundary posterior distribution are extracted. Based on the mean and covariance, a set of sigma sampling points is selected. Each sigma sampling point is propagated along the drilling direction to a depth node of a preset step size. At each depth node, the mean and covariance of the probabilistic boundary position are reconstructed.

7. The automatic correction and optimization system for geological guidance trajectory according to claim 1, characterized in that, In the probabilistic boundary rolling pre-simulation module, the dynamic collision probability of the probabilistic boundary position and trajectory is calculated, including: Calculate the shortest distance from the trajectory to the probabilistic boundary position. Based on the covariance matrix between the shortest distance and the probabilistic boundary position, calculate the failure probability value of the trajectory crossing the probabilistic boundary position, which is used as the dynamic collision probability.

8. The automatic correction and optimization system for geological guidance trajectory according to claim 1, characterized in that, The probabilistic boundary rolling pre-simulation module generates a collision probability evolution map distributed along the drilling depth, including: The collision probability evolution map generates a one-dimensional probability curve with drilling depth as the x-axis and dynamic collision probability as the y-axis; The collision probability evolution map also includes a confidence envelope of the dynamic collision probability, which is superimposed with an uncertainty bandwidth on the dynamic collision probability based on the quantiles of the boundary posterior distribution. The uncertainty bandwidth widens at depths where the variance of the boundary posterior distribution increases.

9. The automatic correction and optimization system for geological guidance trajectory according to claim 1, characterized in that, The risk coupling correction decision module includes a correction sequence solution method comprising: Model predictive control is adopted, and the correction sequence of tool face angle and drilling speed is solved by rolling under the constraints of well inclination angle and dogleg degree. The prediction time domain of model predictive control is synchronized with the preset step size of the probability boundary rolling pre-simulation module. The solution to the correction sequence also incorporates the tool face angle change rate constraint as an additional mechanical constraint.

10. The automatic correction and optimization system for geological guidance trajectory according to claim 1, characterized in that, The method for calculating and weighting the pre-exercise deviation in the model evolution module driven by the pre-exercise deviation includes: The actual drilling trajectory is compared with the probabilistic boundary position at the corresponding depth, and the shortest distance between the actual drilling trajectory and the probabilistic boundary position is calculated as the pre-drilling deviation. The calibration coefficients and stratified prior parameters of the un-drilled well section ahead are corrected by weighted adjustment of the pre-drilling deviation. An exponentially weighted moving average rule is adopted, in which the weight of the pre-drilling deviation decreases exponentially with the increase of the depth of the un-drilled well section ahead.

Images