Method and apparatus for analyzing a stuck helical wellbore

Abstract

This invention discloses a method and apparatus for analyzing stuck pipe in spiral wells. The method includes: acquiring drill string vibration signals and drilling engineering parameters in real time using near-bit sensors; calculating the high-frequency torque fluctuation energy entropy based on the drill string vibration signals; calculating the suspended weight dispersion based on the drilling engineering parameters; performing inversion of well inclination angle data, azimuth angle data, and drill string vibration signals to obtain wellbore curvature inversion results; calculating the predicted cuttings bed thickness based on mechanical drilling rate data and the wellbore curvature inversion results; constructing a multimodal helicity index for quantifying the risk level of stuck pipe in spiral wells based on the suspended weight dispersion, the high-frequency torque fluctuation energy entropy, and the predicted cuttings bed thickness; and dynamically generating drilling pressure adjustment commands and drilling fluid viscosity adjustment commands using a deep reinforcement learning algorithm based on the multimodal helicity index. This invention aims to improve the efficiency of stuck pipe analysis and the accuracy of stuck pipe alarms in spiral wells.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for oil drilling engineering, and in particular to a method and device for analyzing stuck pipe in spiral wells. Background Technology

[0002] This section is intended to provide background or context for the embodiments of the invention described herein. The description herein is not an admission that it is prior art simply because it is included in this section.

[0003] Stuck borehole drilling is a significant safety hazard in oil drilling engineering. Its formation mechanism involves multiple factors, including mismatch between drill string and wellbore trajectory, excessively long sliding drilling time, and spiral-shaped cuttings accumulation. Traditional detection technologies have significant limitations: threshold alarm methods only monitor sudden changes in suspended weight, ignoring the coupling relationship between torque spectrum and vibration energy, leading to a high false alarm rate in hard formations; annular return velocity models are based on the assumption of idealized uniform cuttings distribution, failing to capture the actual non-uniform state of cuttings accumulation in spiral boreholes, resulting in significant prediction errors. Control measures are also severely lagging: manual judgment based on logging curves has a long response delay and cannot prevent the cuttings bed thickness from exceeding the critical value; single control models do not incorporate formation characteristics, resulting in a high failure rate under complex geological conditions. These problems lead to lengthy handling of stuck pipe incidents and significant economic losses per well. Summary of the Invention

[0004] This invention provides a method for analyzing stuck pipe in spiral wells, thereby improving the efficiency of drilling operation safety management, the efficiency of spiral well stuck pipe analysis, and the accuracy of spiral well stuck pipe alarms. The method includes:

[0005] Real-time acquisition of drill string vibration signals and drilling engineering parameters is achieved through near-bit sensors.

[0006] The high-frequency fluctuation energy entropy of torque is calculated based on the drill string vibration signal; the suspended weight dispersion is calculated based on the drilling engineering parameters; the suspended weight dispersion is used to reflect the degree of non-uniform fluctuation of the drill string force; the high-frequency fluctuation energy entropy of torque is used to characterize the abnormal state of drill string rotation.

[0007] The extended Kalman filter algorithm is used to invert wellbore inclination angle data, azimuth angle data, and drill string vibration signals to obtain wellbore curvature inversion results; the cuttings bed thickness prediction value is calculated based on the mechanical drilling rate data and the wellbore curvature inversion results; the cuttings bed thickness prediction value is used to quantify wellbore cleaning efficiency.

[0008] Based on the suspended weight dispersion, the high-frequency fluctuation energy entropy of torque, and the predicted value of cuttings bed thickness, a multimodal helicity index is constructed to quantify the risk level of stuck pipe in spiral wellbores.

[0009] Based on the multimodal helicity index, a deep reinforcement learning algorithm is used to dynamically generate drilling pressure adjustment commands and drilling fluid viscosity adjustment commands; the drilling pressure adjustment commands are used to control and adjust the hydraulic cylinder pressure; the drilling fluid viscosity adjustment commands are used to control the injection of nano-sealants to adjust the rheology of the drilling fluid.

[0010] This invention also provides a spiral wellbore stuck pipe analysis device to improve the efficiency of drilling operation safety management, the efficiency of spiral wellbore stuck pipe analysis, and the accuracy of spiral wellbore stuck pipe alarms. The device includes:

[0011] The data acquisition module is used to acquire drill string vibration signals and drilling engineering parameters in real time through near-bit sensors;

[0012] The torque high-frequency fluctuation energy entropy and suspended weight dispersion calculation module is used to calculate the torque high-frequency fluctuation energy entropy based on the drill string vibration signal; and to calculate the suspended weight dispersion based on the drilling engineering parameters; the suspended weight dispersion is used to reflect the degree of non-uniform fluctuation of the drill string force; and the torque high-frequency fluctuation energy entropy is used to characterize the abnormal state of drill string rotation.

[0013] The cuttings bed thickness prediction calculation module is used to invert well inclination angle data, azimuth angle data, and drill string vibration signals using an extended Kalman filter algorithm to obtain wellbore curvature inversion results; it calculates the cuttings bed thickness prediction value based on the mechanical drilling rate data and the wellbore curvature inversion results; the cuttings bed thickness prediction value is used to quantify wellbore cleaning efficiency.

[0014] The multimodal helicity index construction module is used to construct a multimodal helicity index for quantifying the risk level of stuck pipe in spiral wellbores based on the suspended weight dispersion, the high-frequency fluctuation energy entropy of torque, and the predicted value of cuttings bed thickness.

[0015] The instruction generation module is used to dynamically generate drilling pressure adjustment instructions and drilling fluid viscosity adjustment instructions based on the multimodal helicity index using a deep reinforcement learning algorithm; the drilling pressure adjustment instructions are used to control and adjust the hydraulic cylinder pressure; the drilling fluid viscosity adjustment instructions are used to control the injection of nano-sealant to adjust the rheology of the drilling fluid.

[0016] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described spiral wellbore sticking analysis method.

[0017] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described spiral wellbore sticking analysis method.

[0018] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described spiral wellbore sticking analysis method.

[0019] In this embodiment of the invention, drill string vibration signals and drilling engineering parameters are acquired in real time using near-bit sensors; high-frequency torque fluctuation energy entropy is calculated based on the drill string vibration signals; and suspended weight dispersion is calculated based on the drilling engineering parameters. The suspended weight dispersion reflects the degree of non-uniform fluctuation in drill string stress; the high-frequency torque fluctuation energy entropy characterizes the abnormal state of drill string rotation; and wellbore curvature inversion results are obtained by inverting well inclination angle data, azimuth angle data, and drill string vibration signals using an extended Kalman filter algorithm; and the wellbore curvature inversion is performed based on the mechanical drilling rate data and the wellbore curvature inversion. The results show the predicted value of the cuttings bed thickness; this predicted value is used to quantify wellbore cleaning efficiency; based on the suspended weight dispersion, the high-frequency torque fluctuation energy entropy, and the predicted cuttings bed thickness, a multimodal helicity index is constructed to quantify the risk level of stuck pipe in a spiral wellbore; based on the multimodal helicity index, a deep reinforcement learning algorithm is used to dynamically generate drill pressure adjustment commands and drilling fluid viscosity adjustment commands; the drill pressure adjustment command is used to control and adjust the hydraulic cylinder pressure; the drilling fluid viscosity adjustment is used to control the injection of nano-plugging agents to adjust the rheology of the drilling fluid. This invention calculates vibration energy entropy based on axial vibration signals to reflect the wellbore contact state, extracts high-frequency torque fluctuation energy entropy to characterize drill string rotation anomalies, and combines suspended weight discreteness to quantify non-uniform fluctuations in drill string stress, forming a multi-dimensional physical field feature base. An extended Kalman filter algorithm is used to fuse well inclination angle, azimuth angle, and vibration signals to invert wellbore curvature. The annulus cross-sectional area parameter is dynamically corrected through wellbore geometry, and a cuttings bed thickness prediction value is constructed based on mechanical drilling rate data to accurately quantify wellbore cleaning efficiency. A multimodal helicity index is constructed, which integrates suspended weight discreteness, high-frequency torque fluctuation energy entropy, and cuttings bed thickness prediction value to achieve quantitative classification of helical wellbore stuck hole risk levels. This solves the false alarm defects of the single-parameter threshold method in existing technologies. Multimodal feature fusion significantly improves early warning accuracy compared to traditional methods. Deep reinforcement learning closed-loop control improves response speed. The wellbore curvature inversion correction model exhibits excellent adaptability in complex conditions such as hard formations, thereby reducing the incidence of stuck hole accidents, shortening the accident handling cycle, and significantly saving single-well operating costs, providing reliable technical support for high-risk conditions such as shale oil and gas horizontal wells. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0021] Figure 1 This is a flowchart illustrating a spiral wellbore sticking analysis method according to an embodiment of the present invention;

[0022] Figure 2 This is a specific example diagram of a spiral wellbore sticking analysis method in an embodiment of the present invention;

[0023] Figure 3 This is a specific example diagram of a spiral wellbore sticking analysis method in an embodiment of the present invention;

[0024] Figure 4 This is a specific example diagram of a spiral wellbore sticking analysis method in an embodiment of the present invention;

[0025] Figure 5 This is a schematic diagram of the structure of a spiral wellbore sticking analysis device according to an embodiment of the present invention;

[0026] Figure 6 This is a schematic diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0028] In this document, the term "and / or" merely describes a relationship, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" in this document means any combination of at least two of any one or more elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0029] In the description of this specification, the terms "comprising," "including," "having," and "containing" are open-ended terms, meaning that they include but are not limited to. The terms "an embodiment," "a specific embodiment," "some embodiments," and "for example," etc., refer to specific features, structures, or characteristics described in connection with that embodiment or example that are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, or characteristics described can be combined in any suitable manner in one or more embodiments or examples. The order of steps involved in the various embodiments is used to illustrate the implementation of this application, and the order of steps is not limited and can be adjusted appropriately as needed.

[0030] The acquisition, storage, use, and processing of data in this application comply with relevant regulations. The information collected in this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant standards, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation interfaces are provided for users to choose to authorize or refuse.

[0031] It should be noted that in the embodiments of this application, certain existing solutions in the industry, such as software, components, and models, may be mentioned. For example, some existing software tools, components, algorithm models, or solutions well-known in other technical fields may be cited. These should be considered exemplary, and their purpose is only to illustrate the feasibility of implementing the technical solution of this application. These mentions should be understood as typical examples, and their core purpose is to illustrate and verify the rationality and feasibility of implementing the technical solution proposed in this application. However, this does not mean that the applicant has already used or necessarily used the solution. Such citations do not imply that the applicant has actually adopted these existing solutions, or that it will necessarily adopt these methods in its technical implementation process in the future. In other words, these mentions are only illustrative in nature, helping to understand the connection and transcendence of the innovation points of this application with the prior art, and do not constitute an endorsement or reliance statement on a specific prior art product.

[0032] Stuck borehole drilling is a significant safety hazard in oil drilling engineering. Its formation mechanism involves multiple factors, including mismatch between drill string and wellbore trajectory, excessively long sliding drilling time, and spiral-shaped cuttings accumulation. Traditional detection technologies have significant limitations: threshold alarm methods only monitor sudden changes in suspended weight, ignoring the coupling relationship between torque spectrum and vibration energy, leading to a high false alarm rate in hard formations; annular return velocity models are based on the assumption of idealized uniform cuttings distribution, failing to capture the actual non-uniform state of cuttings accumulation in spiral boreholes, resulting in significant prediction errors. Control measures are also severely lagging: manual judgment based on logging curves has a long response delay and cannot prevent the cuttings bed thickness from exceeding the critical value; single control models do not incorporate formation characteristics, resulting in a high failure rate under complex geological conditions. These problems lead to lengthy handling of stuck pipe incidents and significant economic losses per well.

[0033] Specifically, the formation mechanism and hazards of stuck pipe in spiral wells are as follows:

[0034] Spiral wellbore formations are non-uniform wellbore shapes caused by a mismatch between drill string stiffness and wellbore trajectory (e.g., excessive dogleg length), excessively long sliding drilling time (>60%), or spiral accumulation of cuttings. Their formation process can be divided into three stages:

[0035] 1. Initial micro-helical stage: When the well inclination angle is <30°, slight friction between the drill string and the well wall causes periodic fluctuations in the suspended weight (standard deviation σ<5kN), but no obvious keyway has yet been formed.

[0036] 2. Mid-term spiral expansion stage: Cuttings accumulate at the bottom edge of the wellbore to form a spiral cuttings bed. The torque dominance frequency offset (Δf_peak) increases from the normal <5Hz to >10Hz, and the weight dispersion (CV=σ / μ) exceeds 0.2.

[0037] 3. Late-stage stuck state: The spiral borehole causes distortion in the logging-while-drilling (LWD) signal (e.g., periodic fluctuations in the density curve of ±0.3 g / cm³). 3 The friction coefficient (μ) between the drill string and the wellbore increases by more than 30%, eventually leading to accidents such as pressure buildup and keyway stuck drill bits.

[0038] Spiral wellbore sticking has the following drawbacks:

[0039] 1. Safety risks: The average time to handle a stuck drill bit incident in a spiral well is 72 hours, and the direct cost loss per well exceeds 5 million yuan.

[0040] 2. Efficiency loss: Substandard wellbore quality leads to a 25% increase in the failure rate of well completion tool deployment and a 15%-20% extension of the block development cycle.

[0041] 3. Data distortion: The spiral wellbore trajectory causes an error of 20%-35% in the interpretation of logging-while-drilling data, affecting the accuracy of reservoir evaluation.

[0042] The existing spiral wellbore stuck risk analysis scheme has the following bottlenecks:

[0043] 1. Limitations of testing technology:

[0044] (1) Threshold alarm method: It only monitors sudden changes in suspended weight (±5 tons), but does not combine torque spectrum (energy distribution of 50-150Hz) and vibration energy entropy (H_vib), resulting in a false alarm rate of up to 40% in hard strata.

[0045] (2) Annular return velocity model: The cuttings bed thickness prediction model based on the Herschel-Bulkley equation assumes uniform distribution of cuttings. In reality, the cuttings in the spiral well are spirally stacked (thickness difference > 50%), and the prediction error exceeds 30%.

[0046] 2. Lagging regulatory measures:

[0047] (1) Manual experience adjustment: Engineers rely on visual judgment of logging curves, with an average response delay of 2 hours, which cannot prevent the cuttings bed thickness from exceeding the critical value (≥3cm).

[0048] (2) Single model control: A single LSTM model was used, without the formation drillability index (DCI) and drill string stiffness coefficient (SRF), and the failure rate was controlled to 38% in hard formations (quartz content > 60%).

[0049] To address the aforementioned problems, this invention provides a method for analyzing stuck pipe in spiral wellbores, thereby improving the efficiency of drilling operation safety management, the efficiency of spiral wellbore stuck pipe analysis, and the accuracy of spiral wellbore stuck pipe alarms. Figure 1 This is a flowchart illustrating a spiral wellbore sticking analysis method according to an embodiment of the present invention. See [link / reference]. Figure 1 The method may include:

[0050] Step 101: Acquire drill string vibration signals and drilling parameters in real time using near-bit sensors;

[0051] Step 102: Calculate the high-frequency torque fluctuation energy entropy based on the drill string vibration signal; calculate the suspended weight dispersion based on the drilling engineering parameters; the suspended weight dispersion is used to reflect the degree of non-uniform fluctuation of the drill string force; the high-frequency torque fluctuation energy entropy is used to characterize the abnormal state of drill string rotation;

[0052] Step 103: Using the extended Kalman filter algorithm, invert the well inclination angle data, azimuth angle data, and drill string vibration signal to obtain the wellbore curvature inversion result; calculate the cuttings bed thickness prediction value based on the mechanical drilling rate data and the wellbore curvature inversion result; the cuttings bed thickness prediction value is used to quantify the wellbore cleaning efficiency;

[0053] Step 104: Based on the suspended weight dispersion, the high-frequency fluctuation energy entropy of torque, and the predicted value of cuttings bed thickness, construct a multimodal helicity index for quantifying the risk level of stuck pipe in spiral wellbores;

[0054] Step 105: Based on the multimodal helicity index, a deep reinforcement learning algorithm is used to dynamically generate drilling pressure adjustment commands and drilling fluid viscosity adjustment commands; the drilling pressure adjustment commands are used to control and adjust the hydraulic cylinder pressure; the drilling fluid viscosity adjustment commands are used to control the injection of nano-sealant to adjust the rheology of the drilling fluid.

[0055] In this embodiment of the invention, drill string vibration signals and drilling engineering parameters are acquired in real time using near-bit sensors; high-frequency torque fluctuation energy entropy is calculated based on the drill string vibration signals; and suspended weight dispersion is calculated based on the drilling engineering parameters. The suspended weight dispersion reflects the degree of non-uniform fluctuation in drill string stress; the high-frequency torque fluctuation energy entropy characterizes the abnormal state of drill string rotation; and wellbore curvature inversion results are obtained by inverting well inclination angle data, azimuth angle data, and drill string vibration signals using an extended Kalman filter algorithm; and the wellbore curvature inversion is performed based on the mechanical drilling rate data and the wellbore curvature inversion. The results show the predicted value of the cuttings bed thickness; this predicted value is used to quantify wellbore cleaning efficiency; based on the suspended weight dispersion, the high-frequency torque fluctuation energy entropy, and the predicted cuttings bed thickness, a multimodal helicity index is constructed to quantify the risk level of stuck pipe in a spiral wellbore; based on the multimodal helicity index, a deep reinforcement learning algorithm is used to dynamically generate drill pressure adjustment commands and drilling fluid viscosity adjustment commands; the drill pressure adjustment command is used to control and adjust the hydraulic cylinder pressure; the drilling fluid viscosity adjustment is used to control the injection of nano-plugging agents to adjust the rheology of the drilling fluid. This invention calculates vibration energy entropy based on axial vibration signals to reflect the wellbore contact state, extracts high-frequency torque fluctuation energy entropy to characterize drill string rotation anomalies, and combines suspended weight discreteness to quantify non-uniform fluctuations in drill string stress, forming a multi-dimensional physical field feature base. An extended Kalman filter algorithm is used to fuse well inclination angle, azimuth angle, and vibration signals to invert wellbore curvature. The annulus cross-sectional area parameter is dynamically corrected through wellbore geometry, and a cuttings bed thickness prediction value is constructed based on mechanical drilling rate data to accurately quantify wellbore cleaning efficiency. A multimodal helicity index is constructed, which integrates suspended weight discreteness, high-frequency torque fluctuation energy entropy, and cuttings bed thickness prediction value to achieve quantitative classification of helical wellbore stuck hole risk levels. This solves the false alarm defects of the single-parameter threshold method in existing technologies. Multimodal feature fusion significantly improves early warning accuracy compared to traditional methods. Deep reinforcement learning closed-loop control improves response speed. The wellbore curvature inversion correction model exhibits excellent adaptability in complex conditions such as hard formations, thereby reducing the incidence of stuck hole accidents, shortening the accident handling cycle, and significantly saving single-well operating costs, providing reliable technical support for high-risk conditions such as shale oil and gas horizontal wells.

[0056] In practice, the first step is to obtain drill string vibration signals and drilling parameters in real time through near-bit sensors.

[0057] In one embodiment, the near-bit sensor includes a triaxial accelerometer mounted on the rear of the drill bit; the drill string vibration signal includes: axial vibration acceleration signal, lateral impact acceleration signal, and dynamic torque fluctuation signal; the axial vibration acceleration signal is used to quantify the longitudinal vibration energy distribution of the drill bit; the lateral impact acceleration signal is used to monitor the irregular contact state of the wellbore; the dynamic torque fluctuation signal is the dynamic torque fluctuation signal during the drill string rotation process obtained by a deployed torque meter, used to characterize the energy distribution features extracted by rapid spectrum analysis; the drilling engineering parameters include: suspended weight parameters, pump pressure parameters, and rotational speed parameters.

[0058] In the above embodiments, during specific implementation, the near-bit sensor system includes a triaxial accelerometer deployed in the rear section of the drill bit. This device simultaneously captures axial vibration acceleration signals and lateral impact acceleration signals. The axial vibration signal is used to quantify the longitudinal energy distribution characteristics of the drill string, while the lateral impact signal monitors changes in the irregular contact state of the wellbore. Dynamic torque fluctuation signals are acquired in real time by an independently installed high-frequency torque meter, and abnormal energy distribution characteristics during drill string rotation are extracted using rapid spectrum analysis technology. The drilling engineering parameter acquisition channel integrates a suspended weight sensor, a pump pressure sensor, and a rotation speed meter. The suspended weight parameter records the temporal changes in the hook load, the pump pressure parameter reflects the pressure fluctuation state of the drilling fluid circulation system, and the rotation speed parameter is correlated with the trend of drill string rotation speed changes.

[0059] All sensor data undergoes synchronous multi-source signal processing via edge computing nodes. Axial and lateral vibration signals are used to generate vibration energy entropy based on frequency domain energy distribution characteristics. Torque fluctuation signals are analyzed for energy dispersion within a preset high-frequency band. Suspension weight parameters are calculated for dispersion using statistical fluctuation quantification methods. The processed vibration signals, engineering parameters, and timestamp information are standardized and encapsulated to form a unified data stream output for the multimodal sensing layer, providing a temporally consistent basic input for the downstream feature fusion engine.

[0060] After data encapsulation, the edge computing node performs time alignment on axial vibration energy entropy, lateral impact characteristics, high-frequency torque fluctuation energy entropy, suspended weight dispersion, pump pressure change trend, and rotational speed fluctuation, and pushes this data to the intelligent decision-making layer in real time via a parallel transmission channel. This data transmission mechanism ensures strict synchronization between well depth coordinates and sensor acquisition actions, avoiding the feature analysis distortion problem caused by transmission delays in traditional systems, and establishing a highly reliable data foundation for dynamic risk quantification in spiral wells.

[0061] In specific implementation, after step 101: acquiring drill string vibration signals and drilling engineering parameters in real time through near-bit sensors, step 102: calculating the high-frequency fluctuation energy entropy of torque based on the drill string vibration signals; calculating the suspended weight dispersion based on the drilling engineering parameters; the suspended weight dispersion is used to reflect the degree of non-uniform fluctuation of the drill string force; the high-frequency fluctuation energy entropy of torque is used to characterize the abnormal state of drill string rotation.

[0062] In the above embodiment, the drill string vibration signal processing module receives the raw data stream collected by the near-bit sensor; for the dynamic torque fluctuation signal, it focuses on a preset high-frequency fluctuation range using fast spectrum conversion technology, extracts the discrete characteristics of energy distribution within this frequency band, and generates a torque high-frequency fluctuation energy entropy index characterizing the abnormal state of drill string rotation. This calculation process uses a spectrum energy focusing analysis method to capture the torque dominant frequency offset characteristics and quantify its abnormality, and outputs the entropy value characterization result in real time through edge computing nodes.

[0063] The suspended weight parameter in drilling engineering parameters is processed using statistical analysis methods. Based on the time-series variation data of the hook load, the degree of fluctuation dispersion within a continuous time window is calculated to generate a suspended weight dispersion index. This index directly reflects the non-uniform fluctuation characteristics of the drill string stress and dynamically tracks the load change trend through a sliding window update mechanism. The calculation process integrates the data stream continuously acquired by the suspended weight sensor and outputs the dispersion characterization value through a fluctuation characteristic quantification model.

[0064] The calculation results of high-frequency torque fluctuation energy entropy and suspended weight dispersion are kept synchronized in the time domain through a timestamp alignment mechanism. Edge computing nodes standardize and encapsulate the two types of indicators to form a feature vector with unified dimensions. The torque entropy value is used to identify the risk of drill string rotation instability, while the suspended weight dispersion is associated with abnormal fluctuations in drill pressure transmission. Together, they constitute the core criteria for downhole dynamic behavior. The processed feature vector is pushed to the intelligent decision-making layer in real time, providing an input basis for the construction of a multimodal risk index.

[0065] The signal processing flow strictly follows a collaborative mechanism of physical sensing and algorithm calculation: vibration signals are analyzed for energy distribution in a preset high-frequency band to capture rotational anomalies, while engineering parameters rely on statistical models to quantify the non-uniformity of force. All calculations are completed at the edge, ensuring that the timeliness from data acquisition to feature generation meets the requirements of closed-loop control.

[0066] In one embodiment, calculating the high-frequency fluctuation energy entropy of torque based on the drill string vibration signal includes:

[0067] By analyzing the axial vibration acceleration signal, lateral impact acceleration signal, and dynamic torque fluctuation signal in the drill bit vibration signal, the vibration energy entropy and torque main frequency offset are extracted.

[0068] In one embodiment, calculating the weight dispersion based on the drilling engineering parameters includes:

[0069] By using statistical analysis methods, the weight dispersion is generated based on the time-series variation values ​​of the weight parameters in the drilling engineering parameters.

[0070] In the above embodiments, the drill string vibration signal analysis module receives axial vibration acceleration signals and lateral impact acceleration signals collected by a triaxial accelerometer, and simultaneously receives dynamic torque fluctuation signals captured by a high-frequency torque meter. The axial vibration signal undergoes frequency domain energy distribution characteristic analysis to extract a quantized value of the dispersion of vibration energy within a preset frequency band. This quantized value characterizes the instability of the contact state between the drill bit and the wellbore. The lateral impact signal is used to identify the frequency and intensity level of irregular contact events with the wellbore. The dynamic torque fluctuation signal is processed using fast spectrum conversion technology, focusing on a specific high-frequency fluctuation band, analyzing the discrete characteristics of energy distribution within this band, and generating a high-frequency torque fluctuation energy entropy index. This index directly reflects the abnormal fluctuation state during drill string rotation, and the entropy value characterization result is output in real time through edge computing nodes.

[0071] The high-frequency torque fluctuation energy entropy calculation process continuously tracks well depth changes and recalculates the spectral energy distribution characteristics within each data update window. The analysis process employs a sliding time window mechanism to ensure that the entropy index reflects the drill string rotation state of the current well section in real time. The calculation results are standardized and encapsulated to form a feature vector stream, providing torque-dimensional input parameters for subsequent multimodal risk index construction.

[0072] The suspended weight parameter processing module receives time-series data on the hook load variation collected by the suspended weight sensor. It calculates the load fluctuation characteristics within a continuous time window using statistical analysis methods, quantifying the degree of non-uniform fluctuation in drill string stress. Specifically, it extracts a suspended weight parameter sequence of a fixed time length, calculates the ratio of the standard deviation to the mean of this sequence, and generates a quantified value for the suspended weight dispersion. This value is directly related to downhole pressure transmission efficiency; increased dispersion indicates increased friction between the drill string and the wellbore.

[0073] The calculation process employs a dynamic window adjustment mechanism, automatically optimizing the statistical window length based on changes in the mechanical drilling rate. During high-speed drilling, the statistical window is shortened to improve response speed, while in complex well sections, the window is extended to ensure statistical reliability. The generated suspended weight dispersion index is synchronized with the high-frequency torque fluctuation energy entropy through a timestamp alignment mechanism, together forming a core feature pair reflecting downhole dynamic behavior.

[0074] The suspended weight dispersion index is pushed to the intelligent decision-making layer in real time through a feature fusion interface to participate in the risk assessment of stuck pipe in spiral wells. After the calculation and encapsulation are completed at the edge computing node, the index data is strictly bound to the well depth coordinates, avoiding the misjudgment of status caused by data delay in traditional methods.

[0075] The two types of feature calculation processes are executed in parallel on the edge computing nodes: the vibration signal processing channel focuses on torque fluctuation energy analysis, while the engineering parameter channel processes the statistical characteristics of the suspended weight. Calculation results are exchanged via a shared memory area, with the feature alignment module ensuring precise time axis matching. The final output feature vector contains two core indicators: high-frequency torque fluctuation energy entropy and suspended weight dispersion, and is transmitted to the feature fusion engine via a standardized data bus.

[0076] At the intelligent decision-making layer, the high-frequency fluctuation energy entropy of torque is used to identify the risk of rotational instability, while the suspended weight dispersion is correlated with the early warning of cuttings bed accumulation. When both indicators rise abnormally simultaneously, a high-risk level judgment of stuck pipe in the spiral well is triggered, providing key input for deep reinforcement learning decision-making. The entire calculation process maintains stable operation under the high temperature and high pressure environment downhole, meeting the real-time analysis needs of complex working conditions.

[0077] In specific implementation, after performing step 102: calculating the high-frequency fluctuation energy entropy of torque based on the drill string vibration signal; calculating the suspended weight dispersion based on the drilling engineering parameters; the suspended weight dispersion is used to reflect the degree of non-uniform fluctuation of the drill string force; and the high-frequency fluctuation energy entropy of torque is used to characterize the abnormal state of drill string rotation, step 103 is performed: using the extended Kalman filter algorithm to invert the well inclination angle data, azimuth angle data, and drill string vibration signal to obtain the wellbore curvature inversion result; calculating the cuttings bed thickness prediction value based on the mechanical drilling rate data and the wellbore curvature inversion result; and the cuttings bed thickness prediction value is used to quantify the wellbore cleaning efficiency.

[0078] In the above embodiments, the dynamic data streams of well inclination and azimuth are acquired in real time through the measurement while drilling (MWD) system, and the time series of drill string vibration signals monitored by near-bit sensors are acquired simultaneously. An extended Kalman filter algorithm framework is deployed on the edge computing nodes, using the gradient of well inclination change, azimuth offset trend, and vibration energy characteristics as core input measurements. Wellbore trajectory parameters are dynamically corrected through state prediction iteration and measurement update channels. The algorithm continuously eliminates MWD noise interference and drill string vibration signal distortion, outputting high-precision wellbore curvature inversion results. These results map the geometric evolution characteristics of the wellbore trajectory in real time and quantify the development trend of wellbore helicalization.

[0079] In the embodiments, Figure 2 This is a specific example diagram of a spiral wellbore sticking analysis method according to an embodiment of the present invention, such as... Figure 2 As shown, the extended Kalman filter algorithm is used to invert wellbore inclination angle data, azimuth angle data, and drill string vibration signals to obtain wellbore curvature inversion results, including:

[0080] Step 201: Acquire well inclination angle and azimuth angle data in real time using the measurement while drilling system;

[0081] Step 202: At the preset edge computing node, the extended Kalman filter algorithm is used to invert the well inclination angle data, azimuth angle data and drill string vibration signal to obtain the wellbore curvature inversion result.

[0082] In the above embodiments, the inclination angle and azimuth angle data streams are acquired in real time through the measurement while drilling (MWD) system, and the time series of vibration signals from the short section behind the drill bit is acquired simultaneously. An extended Kalman filter algorithm processing framework is deployed on the edge computing nodes, using the inclination angle change rate, azimuth angle change rate, and vibration energy characteristics as input measurements. The wellbore trajectory parameters are iteratively corrected through a dual-channel approach of state prediction and measurement update. The algorithm continuously eliminates MWD noise and drill string vibration interference, outputting high-precision wellbore curvature inversion results. These results reflect the geometric morphological changes of the wellbore trajectory in real time and quantify the evolution trend of wellbore helicalization.

[0083] The inversion process deeply integrates physical sensing and dynamic models: well inclination data provides the basis for trajectory dip angle, azimuth data describes horizontal offset, and vibration signals capture the microscopic contact state between the drill string and the wellbore. Through nonlinear transformation, these three types of data are mapped into a continuous variable of wellbore curvature, which directly characterizes the trajectory bending intensity per unit well depth. The calculation results are strictly bound to the current well depth coordinates, forming a dynamic digital mapping of the wellbore geometry.

[0084] The wellbore curvature inversion results provide crucial input to the downstream calculation module: by modifying the annular cross-sectional area parameter to optimize the cuttings bed thickness prediction model, the spatial constraint effect of curved well sections on the cuttings transport channel is eliminated; simultaneously, geometric boundary conditions are provided for drill string dynamics analysis to assess the stress distribution state of the drill string in the spiral wellbore. After completing the inversion calculation, the edge computing nodes synchronize the curvature data with features such as vibration energy entropy and suspended weight dispersion in time and space to construct the spatial geometric dimension input of the multimodal risk index.

[0085] In one embodiment, Figure 3 This is a specific example diagram of a spiral wellbore sticking analysis method according to an embodiment of the present invention, such as... Figure 3 As shown, the predicted value of the cuttings bed thickness is calculated based on the mechanical drilling rate data and the wellbore curvature inversion results, including:

[0086] Step 301: Calculate the volume of cuttings generated per unit time based on the mechanical drilling rate data;

[0087] Step 302: Determine the cuttings transport efficiency based on the cuttings generation volume and the annular return velocity parameters monitored by the drilling fluid circulation system; dynamically correct the annular geometric cross-sectional area parameters using the wellbore curvature inversion results;

[0088] Step 303: Calculate the cuttings transport lag time parameter based on the corrected annular geometric cross-sectional area parameter and cuttings transport efficiency;

[0089] Step 304: Based on the cuttings transport lag time parameter, calculate the thickness of the cuttings accumulation that is not carried out of the wellbore using the cuttings dynamic balance model, and generate a predicted value for the cuttings bed thickness.

[0090] In the above embodiments, the volume of cuttings generated per unit time is calculated using real-time acquired mechanical drilling rate data. This volume reflects the changing trend of drill bit rock-breaking efficiency. Combined with annular return velocity parameters monitored by the drilling fluid circulation system, the cuttings transport efficiency status within the annulus is determined. The annular geometric cross-sectional area parameters are dynamically corrected using wellbore curvature inversion results to accurately reflect the spatial constraint effect of curved well sections on the cuttings transport channel. Based on the corrected annular geometric cross-sectional area parameters and the cuttings transport efficiency status, the lag time parameter for cuttings transport within the annulus is calculated.

[0091] By employing a dynamic equilibrium model to address the coupling relationship between cuttings generation volume, transport efficiency, and lag time parameters, the thickness of cuttings accumulation that is not carried out of the wellbore by drilling fluid is calculated, generating a real-time updated predicted value for the cuttings bed thickness. This calculation process is completed at edge computing nodes, and the model continuously tracks changes in well depth to dynamically adjust parameter weights, ensuring consistency in prediction accuracy across different well sections.

[0092] When the predicted value exceeds a preset threshold, a drilling fluid control command is triggered. Simultaneously, it serves as an input to the formation coupling dimension constructed using the helicity index, participating in the quantitative assessment of wellbore sticking risk. The calculation process is spatiotemporally synchronized with the wellbore curvature inversion results, and error verification is performed by comparing measured data through a digital twin platform to ensure that the predicted results meet engineering reliability standards. The volume of cuttings generation is correlated with changes in mechanical drilling rate; transport efficiency is affected by both annular return velocity and cross-sectional area; and the lag time parameter reflects the time delay characteristics of cuttings returning from the bottom of the well to the surface. The model output results are pushed to the intelligent decision-making layer through a feature fusion interface, providing a quantitative basis for wellbore cleanliness for closed-loop control.

[0093] The automatic optimization model parameters increase with well depth, maintaining predictive stability during horizontal section extension. The predicted cuttings bed thickness is used not only for real-time early warning but also participates in the calculation of the deep reinforcement learning reward function, optimizing the balance between mechanical drilling rate and wellbore cleanliness. The entire computing system operates stably in a high-temperature, high-pressure downhole environment, meeting the real-time analysis needs of complex operating conditions.

[0094] In specific implementation, in step 103: using the extended Kalman filter algorithm, the well inclination angle data, azimuth angle data, and drill string vibration signal are inverted to obtain the wellbore curvature inversion result; the cuttings bed thickness prediction value is calculated based on the mechanical drilling rate data and the wellbore curvature inversion result; after the cuttings bed thickness prediction value is used to quantify the wellbore cleaning efficiency, step 104: based on the suspended weight dispersion, the high-frequency torque fluctuation energy entropy, and the cuttings bed thickness prediction value, a multimodal helicity index is constructed to quantify the risk level of stuck pipe in a spiral wellbore.

[0095] In one embodiment, a multimodal helicity index for quantifying the sticking risk level of a spiral wellbore is constructed based on the suspended weight dispersion, the high-frequency fluctuation energy entropy of torque, and the predicted value of the cuttings bed thickness, including:

[0096] The suspended weight dispersion, the high-frequency fluctuation energy entropy of torque, and the predicted value of cuttings bed thickness are input into a multiphysics coupling model for weighted fusion processing to generate a continuously changing multimodal helicity index. The multiphysics coupling model is used to quantify the dynamic interaction between the wellbore, drill string, and formation, and fuse discrete parameters into a multimodal helicity index for quantifying the risk level of stuck pipe in a spiral wellbore.

[0097] In the above embodiments, the feature fusion engine deployed on the edge computing node synchronously receives four types of core input parameters: the discrete quantification of the suspended weight quantifies the degree of non-uniform fluctuation of the drill string force; the vibration energy entropy characterizes the instability of the contact state between the drill string and the wellbore; the torque dominant frequency offset reflects the abnormal characteristics of drill string rotation; and the predicted value of the cuttings bed thickness quantifies the wellbore cleaning efficiency. These parameters are input into a multiphysics coupling model for dynamic weighted fusion processing. This model, by quantifying the real-time interaction of wellbore geometric constraints, drill string dynamic response, and formation characteristics, fuses the discrete parameters into a continuously changing multimodal helicity index.

[0098] A multiphysics coupling model establishes a linkage mechanism between wellbore trajectory constraints, drill string vibration transmission, and formation lithology feedback: wellbore curvature inversion results provide geometric boundary conditions, suspended weight dispersion and vibration energy entropy jointly describe the drill string dynamic state, and the predicted cuttings bed thickness reflects the formation cuttings transport characteristics. The model generates a continuous variable of helicity index through a nonlinear weighted algorithm, which maps the helical wellbore sticking risk level in real time. The risk level classification strictly corresponds to the characteristics of different development stages of helical wellbore formation.

[0099] The weighting coefficients of different characteristics are dynamically adjusted with increasing well depth. In vertical well sections, the focus is on the coupled analysis of suspended weight dispersion and vibration energy entropy, while in horizontal sections and sections with large doglegs, the weighting of torque dominance frequency offset and cuttings bed thickness is strengthened. The generated helicity index is bound to the current well depth coordinates through a timestamp to form a spatiotemporal evolution map of downhole risk status.

[0100] When the index is in the low-risk range, the current drilling parameters are maintained; when it enters the medium-risk range, a short trip-out frequency optimization command is triggered; when the high-risk threshold is reached, drilling pressure adjustment and drilling fluid rheology optimization programs are simultaneously initiated. The index calculation process is verified for reliability in real time through a digital twin platform, and correlation analysis is performed between the predicted risk level and actual downhole operating data to ensure that the risk quantification results meet engineering reliability standards.

[0101] In specific implementation, after step 104: constructing a multimodal helicity index for quantifying the risk level of stuck pipe in spiral wellbore based on the suspended weight dispersion, the high-frequency fluctuation energy entropy of torque, and the predicted value of cuttings bed thickness, step 105: dynamically generating drilling pressure adjustment commands and drilling fluid viscosity adjustment commands based on the multimodal helicity index using a deep reinforcement learning algorithm; the drilling pressure adjustment command is used to control and adjust the hydraulic cylinder pressure; the drilling fluid viscosity adjustment is used to control the injection of nano-plugging agents to adjust the rheology of the drilling fluid.

[0102] In one embodiment, based on the multimodal helicity index, a deep reinforcement learning algorithm is used to dynamically generate drilling pressure adjustment commands and drilling fluid viscosity adjustment commands, including:

[0103] The multimodal helicity index and mechanical drilling rate are used as state inputs to the spiral wellbore control depth reinforcement learning model. The spiral wellbore control depth reinforcement learning model is used to train a pre-set depth reinforcement learning model based on the spiral wellbore control dataset, by analyzing the coupling relationship between downhole risk level and drilling efficiency through a proximal strategy optimization algorithm. The spiral wellbore control dataset includes historical data of drilling pressure adjustment commands and drilling fluid viscosity adjustment commands corresponding to different multimodal helicity indices and mechanical drilling rates.

[0104] Receive drilling pressure adjustment commands and drilling fluid viscosity adjustment commands corresponding to the state input quantities output by the spiral wellbore control depth reinforcement learning model.

[0105] In the above embodiments, the multimodal helicity index and real-time mechanical drilling rate are jointly analyzed as state inputs in the spiral well control depth reinforcement learning model deployed in the intelligent decision-making layer. The model uses a proximal strategy optimization algorithm to analyze the coupling relationship between downhole risk level and drilling efficiency, dynamically generating drilling pressure adjustment commands and drilling fluid rheological optimization commands. During the model training phase, a historical spiral well control dataset is used. This dataset contains the mapping relationship between the multimodal helicity index, mechanical drilling rate, and their corresponding control commands under different operating conditions. An association mechanism between risk state and optimal control action is established through iterative learning.

[0106] The model inference process receives the current state input in real time and calculates the optimal control strategy through pre-trained network weights. Drill pressure adjustment commands are transmitted to the hydraulic cylinder pressure regulating unit, which uses a closed-loop control algorithm to change the applied drill pressure load intensity, maintaining the dynamic balance of the drill string. Drill fluid rheological optimization commands drive the nano-plugging agent injection system, adjusting the drilling fluid rheological properties according to preset particle size characteristics and injection rate, thereby influencing cuttings bed deposition conditions by increasing the dynamic-plastic ratio.

[0107] When the helicity index is in the low-risk range, a fine-tuning command for drilling pressure is output; when entering the medium-risk stage, drilling pressure adjustment and short tripping frequency optimization are linked; when the high-risk threshold is reached, drilling pressure control and drilling fluid performance enhancement regulation are triggered simultaneously. All commands are virtually pre-simulated through a digital twin platform, and the stress distribution of the drill string and the trend of wellbore cleanliness changes are verified in a three-dimensional simulation environment to ensure that the commands meet the engineering safety boundaries.

[0108] In specific implementation, the spiral wellbore resistance analysis method provided in this embodiment of the invention may further include: aligning the feature spatial distribution of the feature data collected from the new well with the feature data of neighboring wells in the knowledge base using the CORRELATIONALIGNMENT algorithm, wherein the neighboring well data includes historical drilling data of a certain bay or basin block, eliminating the feature distribution differences between geological blocks; simultaneously fixing the long short-term memory hidden layer parameters of the pre-trained deep reinforcement learning model, and only fine-tuning and optimizing the connection weights of the output layer, so as to quickly complete the model transfer adaptation under the condition that the number of new well data samples is less than 500 sets.

[0109] In specific implementation, the spiral wellbore sticking analysis method provided in this embodiment of the invention may further include: constructing a dynamic model of the drilling process based on a three-dimensional simulation engine, with input parameters including wellbore trajectory geometry, drill string assembly structural characteristics, and formation lithology parameters, and outputting wellbore cleanliness status and drill string stress distribution in real time; verifying the prediction accuracy and working condition mapping reliability of the digital twin model by comparing the error rate between the predicted value of the cuttings bed thickness and the downhole measured data, as well as the correlation coefficient between the drill string vibration simulation signal and the sensor measured signal, and ensuring that the control commands generated by the deep reinforcement learning algorithm conform to the engineering safety boundary.

[0110] The verification platform overcomes the limitations of traditional cuttings transport models in predicting errors through a dual mechanism of controlling the error rate of cuttings bed thickness prediction and verifying the correlation of vibration signals. The control commands that pass the verification are synchronously sent to the physical execution unit, forming a seamless connection from virtual pre-simulation to on-site implementation, providing engineering-level reliability assurance for blockage control under complex geological conditions.

[0111] The following is a specific embodiment to illustrate the specific application of the device of the present invention.

[0112] This embodiment proposes a real-time early warning and dynamic control system and method for stuck pipe in spiral wells based on multimodal data fusion and deep reinforcement learning. The technical architecture is as follows:

[0113] This system adopts a three-layer architecture of "perception-decision-execution". Figure 4 This is a specific example diagram of a spiral wellbore sticking analysis method according to an embodiment of the present invention, such as... Figure 4 As shown, it specifically includes the following:

[0114] 1. Start: The system starts up and enters the running state.

[0115] 2. Multimodal perception layer data acquisition: Collect multimodal data (such as images, sound, distance, etc.) from the environment through multiple sensors.

[0116] 3. Data transmission to edge computing nodes: The collected data is sent to the nearest edge computing node for preliminary processing.

[0117] 4. Determine: Does the data require cloud-based DRL training?

[0118] Yes: If the data is highly complex and requires a deep reinforcement learning (DRL) model to participate in the decision-making process, then the data will be uploaded to the cloud.

[0119] No: If the data can be processed directly at the edge node, then cloud processing is skipped.

[0120] 5. Edge computing nodes process data directly: For data that does not require cloud intervention, edge nodes use local models or rule bases for real-time processing.

[0121] 6. Upload data to the cloud: Transfer the data that needs to be trained to the cloud server.

[0122] 7. DRL training in the cloud: The cloud uses deep reinforcement learning algorithms to train or infer data to generate better decision models.

[0123] 8. Generate execution instructions: Whether the processing is done directly at the edge or the results are returned after training in the cloud, specific execution instructions will be generated in the end.

[0124] 9. End: The instruction has been executed and the process has ended.

[0125] I. Multimodal Sensing Layer:

[0126] The near-bit sensor system is deployed at the rear sub of the drill bit and includes a triaxial accelerometer and a dynamic torque capture device. The triaxial accelerometer simultaneously acquires axial vibration and lateral impact signals. The axial vibration signal is used to analyze the longitudinal energy distribution characteristics of the drill bit, while the lateral impact signal monitors irregular contact events with the wellbore. The dynamic torque capture device acquires the torque fluctuation characteristics during drill string rotation in real time, using fast spectrum conversion technology to focus on specific high-frequency fluctuation ranges and extract abnormal energy distribution characteristics within that frequency band. The output signals from both devices are processed in parallel through edge computing nodes to ensure the synchronous output of the disordered characteristics of vibration energy and the abnormal characteristics of torque fluctuations.

[0127] The wellbore morphology inversion module employs a dynamic filtering algorithm framework, integrating wellbore inclination trend, azimuth offset direction, and drill string vibration signals provided by the measurement-while-drilling system. Through dual-channel iterative processing of state prediction and measurement updates, measurement noise and downhole vibration interference are eliminated, generating a quantitative representation of the wellbore trajectory geometry in real time. This result reflects the wellbore's spiral evolution process, providing a spatial morphological benchmark for dynamic correction of the annular cross-sectional area parameter. The inversion process is linked to the well depth coordinate increment, forming a real-time three-dimensional wellbore model that varies with drilling depth.

[0128] The vibration signal input inversion algorithm optimizes trajectory calculation accuracy, and the geometric parameters output from the inversion guide the weight allocation of vibration features. An edge computing architecture achieves end-to-end integration from raw signals to high-order features, avoiding data latency caused by traditional layered processing. The processed vibration energy features, torque fluctuation features, and wellbore geometric parameters are pushed to the decision-making layer through a standardized interface, constituting the core input for risk quantification.

[0129] The following are examples:

[0130] 1. Near-bit sensor group:

[0131] (1) Triaxial MEMS accelerometer (model ADI ADXL355): Installed on the rear section of the drill bit, with a range of ±50g and a sampling rate of 2kHz, it monitors axial vibration (Z-axis) and lateral impact (X / Y-axis) in real time. The vibration energy entropy (H_vib) is calculated using the following formula:

[0132]

[0133] Where P(f) is the normalized power spectral density of the acceleration signal at frequency f.

[0134] (2) High-frequency torque meter (model HBM T40): dynamic response time <5ms, measurement range 0-40kN·m, high-frequency torque fluctuation characteristics of 50-150Hz are extracted by fast Fourier transform (FFT).

[0135] 2. Wellbore morphology inversion module:

[0136] (3) Based on the extended Kalman filter (EKF) algorithm, the well inclination angle (α) and azimuth angle from the measurement while drilling (MWD) are fused. Data and drill string vibration signals are used to invert the wellbore curvature (K) in real time. The calculation formula is as follows:

[0137]

[0138] Where s is the well depth, and the inversion accuracy is ±0.1° / 30m.

[0139] II. Intelligent Decision-Making Layer:

[0140] The feature fusion engine extracts dynamic feature sets from multi-source data streams, covering three dimensions: time domain, frequency domain, and formation coupling. Time domain features include suspended weight dispersion and torque fluctuation variation coefficient. Suspended weight dispersion measures the degree of fluctuation dispersion of drill string load, while the torque variation coefficient characterizes the relative rate of change of torque fluctuation amplitude. Simultaneously, it calculates the proportion of sliding drilling mode time to total drilling time, reflecting the proportion of sliding drilling operations. Frequency domain features consist of vibration energy entropy and torque dominant frequency offset. Vibration energy entropy is generated by analyzing the frequency domain energy distribution characteristics of acceleration signals, characterizing the disordered state of vibration energy. Torque dominant frequency offset is extracted based on fast spectrum conversion of dynamic torque fluctuation signals, capturing anomalous features of the dominant frequency component deviating from the baseline state. The formation coupling feature integrates the formation drillability index and the cuttings bed thickness prediction. The formation drillability index obtains the rock drill resistance characteristics through the geological parameter interface, while the cuttings bed thickness prediction is dynamically calculated based on the mechanical drilling rate, annular return rate, annular cross-sectional area and cuttings transport lag time. The annular cross-sectional area parameter is corrected by the wellbore curvature inversion results to reflect the geometric constraints of the curved well section.

[0141] All features are standardized to form feature vectors with uniform dimensions, which are then input into the state space of the deep reinforcement learning model. State space parameters include key variables such as well depth, applied drill pressure (DP), drill string rotation speed, weight dispersion, vibration energy entropy, and formation drillability index. The deep reinforcement learning model employs a proximal policy optimization algorithm, defining three core control actions in the action space: DP adjustment amplitude, short tripping frequency, and drilling fluid rheological optimization intensity. The model balances the relationship between weight stability, mechanical drilling efficiency, and DP adjustment amplitude through a reward function: prioritizing weight fluctuations within a safe threshold, optimizing the mechanical drilling rate to the block benchmark level, and minimizing DP adjustment amplitude to avoid frequent disturbances.

[0142] The feature fusion engine verifies the validity of input features in real time, and triggers a cross-block transfer learning framework to align feature distributions when anomalies are detected. The deep reinforcement learning model dynamically adjusts the action space exploration rate according to the risk level, focusing on weight stability optimization in high-risk states and increasing drilling speed weight in low-risk states. After the control commands are generated, they are pre-verified through a digital twin platform to ensure that the drill string stress distribution and wellbore cleanliness changes meet the safety boundaries.

[0143] The core innovation of this technical solution lies in solving the misjudgment problem caused by isolated feature analysis in traditional methods through multi-physics feature coupling and adaptive decision-making mechanisms. The feature fusion engine transforms temporal fluctuations, frequency anomalies, and geological characteristics into unified risk quantification inputs, while the deep reinforcement learning model outputs the optimal control action through a multi-objective balancing strategy, forming a closed-loop optimization link from data to decision.

[0144] The following are examples:

[0145] 1. Feature Fusion Engine:

[0146] Extracting 32-dimensional dynamic features, including:

[0147] Time-domain characteristics: suspended weight dispersion (σ), torque variation coefficient (CV_TQ), sliding drilling ratio (T_slide / T_total);

[0148] Frequency domain characteristics: vibration energy entropy (H_vib), torque dominant frequency offset (Δf_peak);

[0149] Formation coupling characteristics: Drillability Index (DCI), predicted cuttings bed thickness (h_cuttings):

[0150]

[0151] Where A_bit is the drill bit cross-sectional area, Q_eff is the amount of cuttings returned, v_ann is the annular return velocity, A_ann is the annular cross-sectional area, and t_lag is the cuttings transport lag time.

[0152] 2. Deep Reinforcement Learning Model (DRL):

[0153] State space: 12-dimensional vector, including well depth, drilling weight (WOB), rotation speed (RPM), σ, H_vib, DCI, etc.;

[0154] Action range: 5 control actions, including drilling pressure adjustment (±5%-15%), short start-up frequency (once every 50-150m), and drilling fluid viscosity adjustment (±5%-10%).

[0155] The reward function R is shown below:

[0156]

[0157] Where sigma_max = 10kN is the maximum allowable weight fluctuation, ROP_base is the block baseline mechanical drilling rate; ROP is the mechanical drilling rate; ΔWOB represents the change in drilling pressure.

[0158] III. Dynamic Execution Layer:

[0159] The adaptive pressure-to-drill (PPD) system receives the target load command from a deep reinforcement learning model and adjusts the applied load intensity of the hydraulic cylinder in real time through a closed-loop pressure control algorithm. The control process comprehensively considers the differences in formation lithology, automatically adjusting the weights of control parameters for different rock formations such as sandstone and mudstone to achieve formation-adaptive PPD optimization. The system employs a continuous feedback mechanism, dynamically correcting the output pressure after comparing the actual load measurement value with the target command, ensuring that the stress state of the drill string quickly converges to the target range. The adjustment process is synchronized with changes in well depth, maintaining PPD application accuracy during horizontal extension.

[0160] The drilling fluid intelligent control unit integrates a rheological performance monitoring device and a chemical agent injection system. When the predicted cuttings bed thickness exceeds a preset critical threshold, a nanoscale plugging agent injection procedure is automatically triggered. During the injection process, the injection rate is dynamically adjusted based on real-time drilling fluid rheological monitoring data, optimizing the drilling fluid's cuttings carrying capacity by increasing the dynamic-plastic ratio parameter. The system establishes a rheological change tracking mechanism; when insufficient yield value improvement is detected, the injection intensity is automatically increased to ensure that annular cuttings transport efficiency returns to a safe level.

[0161] The two execution systems establish a collaborative working mechanism: when the drill string pressure adjustment unit optimizes the dynamic balance of the drill string, it simultaneously notifies the drilling fluid control unit to predict the trend of cuttings disturbance; after the drilling fluid rheology is improved, the feedback signal assists the drill string pressure adjustment unit in correcting the formation friction coefficient parameters. The execution results are double-verified with actual working condition data through a digital twin platform. When the hydraulic cylinder pressure response or dynamic plastic ratio changes deviate from expectations, the online compensation mechanism for command parameters is triggered.

[0162] Drill pressure control and fluid regulation form a physical closed loop: hydraulic cylinder pressure sensors provide real-time feedback of actual load values, while online rheometers continuously monitor yield value changes. These two signals together form the basis for verifying the execution effect. In hard formation drilling, emphasis is placed on ensuring drill pressure control accuracy, while in high dogleg sections, the frequency of drilling fluid rheological adjustment is strengthened. An adaptive operating condition strategy balances the system load.

[0163] The entire execution layer architecture meets the reliability requirements under extreme operating conditions: the pressure regulation system maintains control stability under high temperature and vibration environments, and the fluid control unit maintains injection accuracy under drilling fluid contamination conditions. Through coupled analysis of drill string dynamics and annular cleanliness, the system initiates collaborative prevention and control in the early stages of cuttings bed accumulation, overcoming the response lag defects of traditional execution units operating independently.

[0164] The following are examples:

[0165] 1. Drilling pressure adaptive regulator:

[0166] The hydraulic cylinder pressure is adjusted in real time according to the target drill pressure (WOB_target) output by the DRL, with an adjustment accuracy of ±2kN and a response time of <30 seconds.

[0167] Drilling pressure adjustment logic:

[0168]

[0169] Where k is the formation coefficient (k = 0.8 for sandstone, k = 1.2 for mudstone); the current MSE represents the current mechanical specific energy; the theoretical MSE represents the theoretical mechanical specific energy; WOB 初始 This indicates the initial drilling pressure.

[0170] 2. Drilling fluid intelligent control unit:

[0171] Integrating an online rheometer (measuring viscosity and yield value) with an automatic dosing system, when h_cuttings>2cm, a nano-plugging agent (particle size 50-100nm) is injected at a rate of 0.1-5L / min, increasing the drilling fluid dynamic-plastic ratio (YP / PV) by 15%-25%.

[0172] IV. Multiphysics Coupling Modeling:

[0173] In the technical solution, a multiphysics coupled modeling method is used to construct a comprehensive dynamic equation for the wellbore, drill string, and formation system, aiming to define the helicity index as a core risk assessment indicator. This method integrates drill string torsional stiffness properties, formation elastic modulus characteristics, friction coefficient influence, drill string length dimensions, bending stiffness parameters, torque dominance frequency offset characteristics, and drill string rotation frequency state to form a unified quantitative model that reflects downhole dynamic interactions in real time. The calculation process of the helicity index incorporates these variables, quantifying the mechanical response of the drill string and wellbore, as well as the feedback from the formation rock to the drilling process, directly mapping the changing trend of the helical wellbore sticking risk level.

[0174] When the helicity index exceeds a preset high threshold, the system triggers an early warning mechanism. Under hard strata conditions, this early warning threshold can be appropriately lowered to accommodate differences in rock properties. The early warning signal is directly input into the deep reinforcement learning model to dynamically optimize the control strategy, ensuring that the risk response is self-adaptive to the geological environment. The entire modeling process works in conjunction with the feature fusion engine, using the calculated helicity index as the core input of the intelligent decision-making layer, achieving a seamless connection from physical field coupling analysis to risk prevention and control.

[0175] The application of the helicity index enhances the accuracy of downhole risk quantification. By continuously tracking the dynamic state of the drill string and formation feedback, intervention can be initiated at the initial stage of cuttings bed accumulation or drill string instability, significantly improving the robustness of the system.

[0176] The following example illustrates this: Constructing the coupled dynamic equations of wellbore-drill string-formation, and defining the Spiral Hazard Index (SHI):

[0177]

[0178] Variable description:

[0179] T_w: Torsional stiffness of the drill string (N·m / rad);

[0180] E_s: Elastic modulus of formation (GPa);

[0181] μ: coefficient of friction;

[0182] L: Drill string length (m);

[0183] EI: Bending stiffness (N·m) 2 );

[0184] Delta f_peak: Torque frequency offset (Hz);

[0185] f_rot: Drill bit rotation frequency (Hz).

[0186] An early warning is triggered when SHI > 0.75 (the threshold for hard formations can be reduced to 0.6).

[0187] V. Cross-block transfer learning framework:

[0188] The cross-block transfer learning framework uses a feature distribution alignment algorithm to handle the correlation between new well data and historical data from neighboring wells, matching the feature spaces of the blocks to eliminate feature distribution shifts caused by differences in geological structures. Specifically, feature space transformation technology is used to align the statistical characteristics of new well data and knowledge base data, ensuring that core parameters such as vibration energy entropy and suspended weight dispersion in different blocks are within comparable dimensions. In the model transfer stage, the hidden layer parameters of the long-term memory network of the pre-trained deep reinforcement learning model are locked, and only the connection weights of the output layer are fine-tuned and optimized. This enables rapid model adaptation under the condition of limited new well data sample size, overcoming the model failure problem of traditional methods in drilling in new exploration areas.

[0189] The digital twin verification platform constructs a dynamic model of the drilling process based on a 3D simulation engine. It inputs the wellbore trajectory geometry, drill string assembly structural characteristics, and formation lithology parameters, and outputs real-time wellbore cleanliness status and drill string stress distribution maps. The platform verifies the accuracy of the simulation model's mapping to the physical world by comparing the error between predicted cuttings bed thickness and downhole measured data, as well as the correlation strength between drill string vibration simulation signals and actual sensor data. This verification mechanism significantly reduces cuttings bed prediction errors, demonstrating significantly better accuracy than traditional annular return velocity calculation models. Furthermore, the drill string dynamics simulation results maintain a high degree of consistency with measured conditions, providing a highly reliable verification basis for intelligent decision-making.

[0190] The control strategy generated after processing new well data using the transfer learning framework is first input into a digital twin platform for virtual simulation. The platform evaluates the feasibility of the strategy by analyzing the wellbore cleanliness change trend and drill string stress distribution. Verified commands are then sent to the physical execution unit. When there is a significant deviation between the simulation results and the transfer strategy, the feature alignment module is triggered to recalibrate the new well data distribution or adjust the parameter weights of the fine-tuning output layer. This closed-loop mechanism achieves rapid and reliable adaptation in diverse geological conditions, such as the Sichuan Basin shale formations and deep-water blocks in the South China Sea, solving the control lag problem caused by the lack of geological data in traditional methods.

[0191] The following are examples:

[0192] 1. Feature Alignment: The CORAL algorithm (COR relation AL ignment) is used to align new well data with neighboring well data (such as data from a bay or basin) in the knowledge base in the feature space, eliminating distribution differences between blocks;

[0193] 2. Parameter transfer: Fix the LSTM hidden layer parameters of the pre-trained DRL model and only fine-tune the output layer to achieve rapid adaptation under small sample data (new well data <500 sets).

[0194] Digital twin verification platform:

[0195] 3. Simulation Modeling: A drilling process simulation system is built based on the Unity3D engine. The input parameters include wellbore trajectory, drill string assembly, and formation lithology. The output parameters are wellbore cleanliness and drill string stress distribution.

[0196] 4. Accuracy verification: The error in the predicted thickness of the cuttings bed is <8% (compared to >25% for the traditional HB model), and the correlation coefficient A^2 between the drill string vibration simulation and the measured data is >0.92.

[0197] This invention aims to provide an intelligent early warning and control system for stuck pipe in spiral wells. It quantifies the risks of spiral wells through a multi-physics coupling model and achieves precise early warning and adaptive control by combining deep reinforcement learning and transfer learning. Industrial trials in 12 wells in the Sichuan Basin and the South China Sea show that the system's early warning accuracy reaches 95.3%, the stuck pipe accident rate is reduced by 82%, and the cost savings per well exceed 2 million yuan, providing technical support for safe drilling in complex well types.

[0198] In summary, the core of a spiral wellbore stuck-out early warning method lies in constructing a comprehensive risk assessment mechanism through multi-source parameter fusion. First, the fluctuation dispersion characteristics of the suspended weight parameter are collected, reflecting the non-uniform state of the drill string stress. Simultaneously, the fluctuation energy entropy of the torque signal within a specific high-frequency band is extracted to characterize abnormal drill string rotation. Combined with the predicted cuttings bed thickness calculated dynamically based on wellbore geometry and mechanical drilling rate, this value quantifies wellbore cleaning efficiency. These three types of features are input into a multimodal risk index construction module, and continuous risk variables are generated through weighted fusion.

[0199] The aforementioned early warning method employs a deep reinforcement learning algorithm to dynamically optimize control parameters, specifically utilizing a proximal strategy optimization framework. The algorithm's reward function design comprehensively considers three key factors: the stability of the suspended weight parameter, the efficiency level of the mechanical drilling rate, and the magnitude control of the drilling pressure adjustment. By balancing the weights of these three factors, the system ensures optimal decision-making between risk control and drilling efficiency.

[0200] The calculation of torque high-frequency fluctuation energy entropy is strictly limited to a preset high-frequency fluctuation range, and the original torque signal is processed using fast spectrum conversion technology. The analysis process focuses on the energy distribution characteristics within this frequency band, extracting normalized spectral density features as the basis for energy entropy calculation. This frequency band limitation significantly improves the accuracy of identifying rotational anomalies and avoids interference noise from full-band analysis.

[0201] The control system built upon the above methods integrates two core technological frameworks: a cross-block transfer learning framework that eliminates distribution differences between different geological blocks through feature space alignment technology, enabling pre-trained models to quickly adapt to new exploration well data; and a digital twin verification platform that uses a 3D simulation model to pre-simulate control strategies and verify wellbore cleanliness and drill string stress distribution in a virtual environment. These two technologies form a collaborative verification closed loop, ensuring the reliability of model transfer to field implementation.

[0202] As described above, this invention aims to provide an intelligent early warning and control system for stuck pipe in spiral wells. It quantifies the risks of spiral wells through a multi-physics coupling model and achieves precise early warning and adaptive control by combining deep reinforcement learning and transfer learning. Industrial trials in 12 wells in the Sichuan Basin and the South China Sea show that the system's early warning accuracy reaches 95.3%, the stuck pipe accident rate is reduced by 82%, and the cost savings per well exceed 2 million yuan, providing technical support for safe drilling in complex well types.

[0203] Of course, it is understood that there may be other variations of the above detailed process, and all such variations should fall within the protection scope of this invention.

[0204] In this embodiment of the invention, drill string vibration signals and drilling engineering parameters are acquired in real time using near-bit sensors; high-frequency torque fluctuation energy entropy is calculated based on the drill string vibration signals; and suspended weight dispersion is calculated based on the drilling engineering parameters. The suspended weight dispersion reflects the degree of non-uniform fluctuation in drill string stress; the high-frequency torque fluctuation energy entropy characterizes the abnormal state of drill string rotation; and wellbore curvature inversion results are obtained by inverting well inclination angle data, azimuth angle data, and drill string vibration signals using an extended Kalman filter algorithm; and the wellbore curvature inversion is performed based on the mechanical drilling rate data and the wellbore curvature inversion. The results show the predicted value of the cuttings bed thickness; this predicted value is used to quantify wellbore cleaning efficiency; based on the suspended weight dispersion, the high-frequency torque fluctuation energy entropy, and the predicted cuttings bed thickness, a multimodal helicity index is constructed to quantify the risk level of stuck pipe in a spiral wellbore; based on the multimodal helicity index, a deep reinforcement learning algorithm is used to dynamically generate drill pressure adjustment commands and drilling fluid viscosity adjustment commands; the drill pressure adjustment command is used to control and adjust the hydraulic cylinder pressure; the drilling fluid viscosity adjustment is used to control the injection of nano-plugging agents to adjust the rheology of the drilling fluid. This invention calculates vibration energy entropy based on axial vibration signals to reflect the wellbore contact state, extracts high-frequency torque fluctuation energy entropy to characterize drill string rotation anomalies, and combines suspended weight discreteness to quantify non-uniform fluctuations in drill string stress, forming a multi-dimensional physical field feature base. An extended Kalman filter algorithm is used to fuse well inclination angle, azimuth angle, and vibration signals to invert wellbore curvature. The annulus cross-sectional area parameter is dynamically corrected through wellbore geometry, and a cuttings bed thickness prediction value is constructed based on mechanical drilling rate data to accurately quantify wellbore cleaning efficiency. A multimodal helicity index is constructed, which integrates suspended weight discreteness, high-frequency torque fluctuation energy entropy, and cuttings bed thickness prediction value to achieve quantitative classification of helical wellbore stuck hole risk levels. This solves the false alarm defects of the single-parameter threshold method in existing technologies. Multimodal feature fusion significantly improves early warning accuracy compared to traditional methods. Deep reinforcement learning closed-loop control improves response speed. The wellbore curvature inversion correction model exhibits excellent adaptability in complex conditions such as hard formations, thereby reducing the incidence of stuck hole accidents, shortening the accident handling cycle, and significantly saving single-well operating costs, providing reliable technical support for high-risk conditions such as shale oil and gas horizontal wells.

[0205] As described above, this invention relates to the field of intelligent control technology for oil drilling engineering, and in particular to a real-time early warning and dynamic control system for stuck pipe in spiral wells based on multimodal data fusion and deep reinforcement learning. This invention uses high-precision sensors to collect real-time data on drill string vibration, drilling engineering parameters (suspension weight, torque, pump pressure, etc.), and formation characteristics. Combined with intelligent algorithms that integrate edge computing and cloud collaboration, it achieves early identification, risk classification, and automated closed-loop control of stuck pipe in spiral wells. It is particularly suitable for downhole safety control in complex conditions such as shale oil and gas horizontal wells, deepwater drilling, and extended reach wells, effectively solving the industry problems of delayed early warning and high false alarm rates in hard formations and high dogleg sections using traditional methods.

[0206] This invention also provides a spiral wellbore sticking analysis device, as described in the following embodiments. Since the principle by which this device solves the problem is similar to the spiral wellbore sticking analysis method, the implementation of this device can refer to the implementation of the spiral wellbore sticking analysis method; repeated details will not be elaborated further.

[0207] This invention also provides a spiral wellbore stuck pipe analysis device to improve the efficiency of drilling operation safety management, the efficiency of spiral wellbore stuck pipe analysis, and the accuracy of spiral wellbore stuck pipe alarms. Figure 5 This is a schematic diagram of the structure of a spiral wellbore sticking analysis device according to an embodiment of the present invention, as shown below. Figure 5 As shown, the device includes:

[0208] Data acquisition module 501 is used to acquire drill string vibration signals and drilling engineering parameters in real time through near-bit sensors;

[0209] The torque high-frequency fluctuation energy entropy and suspended weight dispersion calculation module 502 is used to calculate the torque high-frequency fluctuation energy entropy based on the drill string vibration signal; and to calculate the suspended weight dispersion based on the drilling engineering parameters; the suspended weight dispersion is used to reflect the degree of non-uniform fluctuation of the drill string force; and the torque high-frequency fluctuation energy entropy is used to characterize the abnormal state of drill string rotation.

[0210] The cuttings bed thickness prediction calculation module 503 is used to invert well inclination angle data, azimuth angle data and drill string vibration signal through extended Kalman filter algorithm to obtain wellbore curvature inversion results; calculate the cuttings bed thickness prediction value based on mechanical drilling rate data and the wellbore curvature inversion results; the cuttings bed thickness prediction value is used to quantify wellbore cleaning efficiency;

[0211] The multimodal helicity index construction module 504 is used to construct a multimodal helicity index for quantifying the risk level of stuck pipe in spiral wellbore based on the suspended weight dispersion, the high-frequency fluctuation energy entropy of torque, and the predicted value of cuttings bed thickness.

[0212] The instruction generation module 505 is used to dynamically generate drilling pressure adjustment instructions and drilling fluid viscosity adjustment instructions based on the multimodal helicity index using a deep reinforcement learning algorithm; the drilling pressure adjustment instructions are used to control and adjust the hydraulic cylinder pressure; the drilling fluid viscosity adjustment instructions are used to control the injection of nano-sealant to adjust the rheology of the drilling fluid.

[0213] In one embodiment, the near-bit sensor includes a triaxial accelerometer mounted on the rear of the drill bit; the drill string vibration signal includes: axial vibration acceleration signal, lateral impact acceleration signal, and dynamic torque fluctuation signal; the axial vibration acceleration signal is used to quantify the longitudinal vibration energy distribution of the drill bit; the lateral impact acceleration signal is used to monitor the irregular contact state of the wellbore; the dynamic torque fluctuation signal is the dynamic torque fluctuation signal during the drill string rotation process obtained by a deployed torque meter, used to characterize the energy distribution features extracted by rapid spectrum analysis; the drilling engineering parameters include: suspended weight parameters, pump pressure parameters, and rotational speed parameters.

[0214] In one embodiment, the torque high-frequency fluctuation energy entropy and suspended weight dispersion calculation module is specifically used for:

[0215] By analyzing the axial vibration acceleration signal, lateral impact acceleration signal, and dynamic torque fluctuation signal in the drill bit vibration signal, the vibration energy entropy and torque main frequency offset are extracted.

[0216] In one embodiment, the torque high-frequency fluctuation energy entropy and suspended weight dispersion calculation module is specifically used for:

[0217] By using statistical analysis methods, the weight dispersion is generated based on the time-series variation values ​​of the weight parameters in the drilling engineering parameters.

[0218] In one embodiment, the cuttings bed thickness prediction calculation module is specifically used for:

[0219] The well inclination angle and azimuth angle data are acquired in real time through the measurement while drilling system;

[0220] At the preset edge computing node, the extended Kalman filter algorithm is used to invert the well inclination angle data, azimuth angle data and drill string vibration signal to obtain the wellbore curvature inversion result.

[0221] In one embodiment, the cuttings bed thickness prediction calculation module is specifically used for:

[0222] Based on mechanical drilling rate data, calculate the volume of cuttings generated per unit time;

[0223] Based on the cuttings generation volume and the annular return velocity parameters monitored by the drilling fluid circulation system, the cuttings transport efficiency is determined; the annular geometric cross-sectional area parameters are dynamically corrected using the wellbore curvature inversion results.

[0224] Based on the corrected annular geometric cross-sectional area parameters and cuttings transport efficiency, the cuttings transport lag time parameters are calculated.

[0225] Based on the cuttings transport lag time parameter, the thickness of the cuttings accumulation that is not carried out of the wellbore is calculated using a cuttings dynamic equilibrium model, and a predicted value of the cuttings bed thickness is generated.

[0226] In one embodiment, the multimodal helicity index construction module is specifically used for:

[0227] The suspended weight dispersion, the high-frequency fluctuation energy entropy of torque, and the predicted value of cuttings bed thickness are input into a multiphysics coupling model for weighted fusion processing to generate a continuously changing multimodal helicity index. The multiphysics coupling model is used to quantify the dynamic interaction between the wellbore, drill string, and formation, and fuse discrete parameters into a multimodal helicity index for quantifying the risk level of stuck pipe in a spiral wellbore.

[0228] In one embodiment, the instruction generation module is specifically used for:

[0229] The multimodal helicity index and mechanical drilling rate are used as state inputs to the spiral wellbore control depth reinforcement learning model. The spiral wellbore control depth reinforcement learning model is used to train a pre-set depth reinforcement learning model based on the spiral wellbore control dataset, by analyzing the coupling relationship between downhole risk level and drilling efficiency through a proximal strategy optimization algorithm. The spiral wellbore control dataset includes historical data of drilling pressure adjustment commands and drilling fluid viscosity adjustment commands corresponding to different multimodal helicity indices and mechanical drilling rates.

[0230] Receive drilling pressure adjustment commands and drilling fluid viscosity adjustment commands corresponding to the state input quantities output by the spiral wellbore control depth reinforcement learning model.

[0231] This invention provides an embodiment of a computer device for implementing all or part of the above-described spiral wellbore sticking analysis method. The computer device specifically includes the following components:

[0232] The computer device comprises a processor, memory, a communications interface, and a bus; wherein the processor, memory, and communications interface communicate with each other via the bus; the communications interface is used to realize information transmission between related devices; the computer device can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the computer device can be implemented with reference to the embodiments for implementing the spiral wellbore stuck block analysis method and the embodiments for implementing the spiral wellbore stuck block analysis device, the contents of which are incorporated herein by reference, and repeated details will not be described again.

[0233] Figure 6 This is a schematic diagram of a computer device provided in an embodiment of the present invention, which discloses a schematic block diagram of the system configuration of a computer device 1000 according to an embodiment of this application. Figure 6As shown, the computer device 1000 may include a central processing unit 1001 and a memory 1002; the memory 1002 is coupled to the central processing unit 1001. It is worth noting that... Figure 6 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0234] In one embodiment, the spiral wellbore sticking analysis function can be integrated into the central processing unit 1001. The central processing unit 1001 can be configured to perform the following control:

[0235] Real-time acquisition of drill string vibration signals and drilling engineering parameters is achieved through near-bit sensors.

[0236] The high-frequency fluctuation energy entropy of torque is calculated based on the drill string vibration signal; the suspended weight dispersion is calculated based on the drilling engineering parameters; the suspended weight dispersion is used to reflect the degree of non-uniform fluctuation of the drill string force; the high-frequency fluctuation energy entropy of torque is used to characterize the abnormal state of drill string rotation.

[0237] The extended Kalman filter algorithm is used to invert wellbore inclination angle data, azimuth angle data, and drill string vibration signals to obtain wellbore curvature inversion results; the cuttings bed thickness prediction value is calculated based on the mechanical drilling rate data and the wellbore curvature inversion results; the cuttings bed thickness prediction value is used to quantify wellbore cleaning efficiency.

[0238] Based on the suspended weight dispersion, the high-frequency fluctuation energy entropy of torque, and the predicted value of cuttings bed thickness, a multimodal helicity index is constructed to quantify the risk level of stuck pipe in spiral wellbores.

[0239] Based on the multimodal helicity index, a deep reinforcement learning algorithm is used to dynamically generate drilling pressure adjustment commands and drilling fluid viscosity adjustment commands; the drilling pressure adjustment commands are used to control and adjust the hydraulic cylinder pressure; the drilling fluid viscosity adjustment commands are used to control the injection of nano-sealants to adjust the rheology of the drilling fluid.

[0240] In another embodiment, the spiral wellbore stuck device can be configured separately from the central processing unit 1001. For example, the spiral wellbore stuck device can be configured as a chip connected to the central processing unit 1001, and the spiral wellbore stuck analysis function can be realized through the control of the central processing unit.

[0241] like Figure 6 As shown, the computer device 1000 may further include: a communication module 1003, an input unit 1004, an audio processor 1005, a display 1006, and a power supply 1007. It is worth noting that the computer device 1000 does not necessarily need to include... Figure 6 All components shown; in addition, the computer device 1000 may also include Figure 6For components not shown, please refer to existing technologies.

[0242] like Figure 6 As shown, the central processing unit 1001, sometimes also referred to as a controller or operation control, may include a microprocessor or other processor device and / or logic device. The central processing unit 1001 receives input and controls the operation of various components of the computer device 1000.

[0243] The memory 1002 may be, for example, one or more of a cache, flash memory, hard drive, removable medium, volatile memory, non-volatile memory, or other suitable device. It may store the aforementioned device-related information, and also store programs for executing that information. The central processing unit 1001 may execute the program stored in the memory 1002 to perform information storage or processing, etc.

[0244] Input unit 1004 provides input to central processing unit 1001. This input unit 1004 may be, for example, a keypad or touch input device. Power supply 1007 provides power to computer device 1000. Display 1006 displays images, text, and other display objects. This display may be, for example, an LCD display, but is not limited to this.

[0245] The memory 1002 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs, etc. The memory 1002 can also be some other type of device. The memory 1002 includes a buffer memory 1021 (sometimes referred to as a buffer). The memory 1002 may include an application / function storage unit 1022 for storing application programs and function programs or processes for executing operations of the computer device 1000 via the central processing unit 1001.

[0246] The memory 1002 may also include a data storage unit 1023 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the computer device. The driver storage unit 1024 of the memory 1002 may include various drivers for the computer device for communication functions and / or for performing other functions of the computer device (such as messaging applications, address book applications, etc.).

[0247] The communication module 1003 is a transmitter / receiver that transmits and receives signals via the antenna 1008. The communication module (transmitter / receiver) 1003 is coupled to the central processing unit 1001 to provide input signals and receive output signals, which is the same as in a conventional mobile communication terminal.

[0248] Based on different communication technologies, multiple communication modules 1003 can be configured in the same computer device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module (transmitter / receiver) 1003 is also coupled to a speaker 1009 and a microphone 1010 via an audio processor 1005 to provide audio output via the speaker 1009 and receive audio input from the microphone 1010, thereby realizing typical telecommunications functions. The audio processor 1005 may include any suitable buffer, decoder, amplifier, etc. Furthermore, the audio processor 1005 is also coupled to a central processing unit 1001, enabling on-device recording via the microphone 1010 and on-device playback of stored sound via the speaker 1009.

[0249] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described spiral wellbore sticking analysis method.

[0250] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described spiral wellbore sticking analysis method.

[0251] In this embodiment of the invention, drill string vibration signals and drilling engineering parameters are acquired in real time using near-bit sensors; high-frequency torque fluctuation energy entropy is calculated based on the drill string vibration signals; and suspended weight dispersion is calculated based on the drilling engineering parameters. The suspended weight dispersion reflects the degree of non-uniform fluctuation in drill string stress; the high-frequency torque fluctuation energy entropy characterizes the abnormal state of drill string rotation; and wellbore curvature inversion results are obtained by inverting well inclination angle data, azimuth angle data, and drill string vibration signals using an extended Kalman filter algorithm; and the wellbore curvature inversion is performed based on the mechanical drilling rate data and the wellbore curvature inversion. The results show the predicted value of the cuttings bed thickness; this predicted value is used to quantify wellbore cleaning efficiency; based on the suspended weight dispersion, the high-frequency torque fluctuation energy entropy, and the predicted cuttings bed thickness, a multimodal helicity index is constructed to quantify the risk level of stuck pipe in a spiral wellbore; based on the multimodal helicity index, a deep reinforcement learning algorithm is used to dynamically generate drill pressure adjustment commands and drilling fluid viscosity adjustment commands; the drill pressure adjustment command is used to control and adjust the hydraulic cylinder pressure; the drilling fluid viscosity adjustment is used to control the injection of nano-plugging agents to adjust the rheology of the drilling fluid. This invention calculates vibration energy entropy based on axial vibration signals to reflect the wellbore contact state, extracts high-frequency torque fluctuation energy entropy to characterize drill string rotation anomalies, and combines suspended weight discreteness to quantify non-uniform fluctuations in drill string stress, forming a multi-dimensional physical field feature base. An extended Kalman filter algorithm is used to fuse well inclination angle, azimuth angle, and vibration signals to invert wellbore curvature. The annulus cross-sectional area parameter is dynamically corrected through wellbore geometry, and a cuttings bed thickness prediction value is constructed based on mechanical drilling rate data to accurately quantify wellbore cleaning efficiency. A multimodal helicity index is constructed, which integrates suspended weight discreteness, high-frequency torque fluctuation energy entropy, and cuttings bed thickness prediction value to achieve quantitative classification of helical wellbore stuck hole risk levels. This solves the false alarm defects of the single-parameter threshold method in existing technologies. Multimodal feature fusion significantly improves early warning accuracy compared to traditional methods. Deep reinforcement learning closed-loop control improves response speed. The wellbore curvature inversion correction model exhibits excellent adaptability in complex conditions such as hard formations, thereby reducing the incidence of stuck hole accidents, shortening the accident handling cycle, and significantly saving single-well operating costs, providing reliable technical support for high-risk conditions such as shale oil and gas horizontal wells.

[0252] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0253] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0254] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0255] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0256] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for analyzing sticking points in spiral wells, characterized in that, include: Real-time acquisition of drill string vibration signals and drilling parameters is achieved through near-bit sensors. The high-frequency fluctuation energy entropy of torque is calculated based on the drill string vibration signal; the suspended weight dispersion is calculated based on the drilling engineering parameters; the suspended weight dispersion is used to reflect the degree of non-uniform fluctuation of the drill string force; the high-frequency fluctuation energy entropy of torque is used to characterize the abnormal state of drill string rotation. The extended Kalman filter algorithm is used to invert wellbore inclination angle data, azimuth angle data, and drill string vibration signals to obtain wellbore curvature inversion results; the cuttings bed thickness prediction value is calculated based on the mechanical drilling rate data and the wellbore curvature inversion results; the cuttings bed thickness prediction value is used to quantify wellbore cleaning efficiency. Based on the suspended weight dispersion, the high-frequency fluctuation energy entropy of torque, and the predicted value of cuttings bed thickness, a multimodal helicity index is constructed to quantify the risk level of stuck pipe in spiral wellbores. Based on the multimodal helicity index, a deep reinforcement learning algorithm is used to dynamically generate drilling pressure adjustment commands and drilling fluid viscosity adjustment commands; the drilling pressure adjustment commands are used to control and adjust the hydraulic cylinder pressure; the drilling fluid viscosity adjustment commands are used to control the injection of nano-sealants to adjust the rheology of the drilling fluid.

2. The method as described in claim 1, characterized in that, The near-bit sensor includes a triaxial accelerometer mounted at the rear of the drill bit; the drill string vibration signal includes: axial vibration acceleration signal, lateral impact acceleration signal, and dynamic torque fluctuation signal; the axial vibration acceleration signal is used to quantify the longitudinal vibration energy distribution of the drill bit; the lateral impact acceleration signal is used to monitor the irregular contact state of the wellbore; the dynamic torque fluctuation signal is the dynamic torque fluctuation signal during the drill string rotation process obtained by a deployed torque meter, used to characterize the energy distribution features extracted by rapid spectrum analysis; the drilling engineering parameters include: suspended weight parameters, pump pressure parameters, and rotational speed parameters.

3. The method as described in claim 1, characterized in that, Calculating the high-frequency fluctuation energy entropy of torque based on the drill string vibration signal includes: By analyzing the axial vibration acceleration signal, lateral impact acceleration signal, and dynamic torque fluctuation signal in the drill bit vibration signal, the vibration energy entropy and torque main frequency offset are extracted.

4. The method as described in claim 1, characterized in that, The calculation of the weight dispersion based on the drilling engineering parameters includes: By using statistical analysis methods, the weight dispersion is generated based on the time-series variation values ​​of the weight parameters in the drilling engineering parameters.

5. The method as described in claim 1, characterized in that, The extended Kalman filter algorithm is used to invert wellbore inclination angle data, azimuth angle data, and drill string vibration signals to obtain wellbore curvature inversion results, including: The well inclination angle and azimuth angle data are acquired in real time through the measurement while drilling system; At the preset edge computing node, the extended Kalman filter algorithm is used to invert the well inclination angle data, azimuth angle data and drill string vibration signal to obtain the wellbore curvature inversion result.

6. The method as described in claim 1, characterized in that, The predicted value of the cuttings bed thickness is calculated based on the mechanical drilling rate data and the wellbore curvature inversion results, including: Based on mechanical drilling rate data, calculate the volume of cuttings generated per unit time; Based on the cuttings generation volume and the annular return velocity parameters monitored by the drilling fluid circulation system, the cuttings transport efficiency is determined; the annular geometric cross-sectional area parameters are dynamically corrected using the wellbore curvature inversion results. Based on the corrected annular geometric cross-sectional area parameters and cuttings transport efficiency, the cuttings transport lag time parameters are calculated. Based on the cuttings transport lag time parameter, the thickness of the cuttings accumulation that is not carried out of the wellbore is calculated using a cuttings dynamic equilibrium model, and a predicted value of the cuttings bed thickness is generated.

7. The method as described in claim 1, characterized in that, Based on the suspended weight dispersion, the high-frequency torque fluctuation energy entropy, and the predicted cuttings bed thickness, a multimodal helicity index is constructed to quantify the sticking risk level of spiral wellbores, including: The suspended weight dispersion, the high-frequency fluctuation energy entropy of torque, and the predicted value of cuttings bed thickness are input into a multiphysics coupling model for weighted fusion processing to generate a continuously changing multimodal helicity index. The multiphysics coupling model is used to quantify the dynamic interaction between the wellbore, drill string, and formation, and fuse discrete parameters into a multimodal helicity index for quantifying the risk level of stuck pipe in a spiral wellbore.

8. The method as described in claim 1, characterized in that, Based on the multimodal helicity index, a deep reinforcement learning algorithm is used to dynamically generate drilling pressure adjustment commands and drilling fluid viscosity adjustment commands, including: The multimodal helicity index and mechanical drilling rate are used as state inputs to the spiral wellbore control depth reinforcement learning model. The spiral wellbore control depth reinforcement learning model is used to train a pre-set depth reinforcement learning model based on the spiral wellbore control dataset, by analyzing the coupling relationship between downhole risk level and drilling efficiency through a proximal strategy optimization algorithm. The spiral wellbore control dataset includes historical data of drilling pressure adjustment commands and drilling fluid viscosity adjustment commands corresponding to different multimodal helicity indices and mechanical drilling rates. Receive drilling pressure adjustment commands and drilling fluid viscosity adjustment commands corresponding to the state input quantities output by the spiral wellbore control depth reinforcement learning model.

9. A spiral wellbore sticking analysis device, characterized in that, include: The data acquisition module is used to acquire drill string vibration signals and drilling engineering parameters in real time through near-bit sensors; The torque high-frequency fluctuation energy entropy and suspended weight dispersion calculation module is used to calculate the torque high-frequency fluctuation energy entropy based on the drill string vibration signal; and to calculate the suspended weight dispersion based on the drilling engineering parameters; the suspended weight dispersion is used to reflect the degree of non-uniform fluctuation of the drill string force; and the torque high-frequency fluctuation energy entropy is used to characterize the abnormal state of drill string rotation. The cuttings bed thickness prediction calculation module is used to invert well inclination angle data, azimuth angle data, and drill string vibration signals using an extended Kalman filter algorithm to obtain wellbore curvature inversion results; it calculates the cuttings bed thickness prediction value based on the mechanical drilling rate data and the wellbore curvature inversion results; the cuttings bed thickness prediction value is used to quantify wellbore cleaning efficiency. The multimodal helicity index construction module is used to construct a multimodal helicity index for quantifying the risk level of stuck pipe in spiral wellbores based on the suspended weight dispersion, the high-frequency fluctuation energy entropy of torque, and the predicted value of cuttings bed thickness. The instruction generation module is used to dynamically generate drilling pressure adjustment instructions and drilling fluid viscosity adjustment instructions based on the multimodal helicity index using a deep reinforcement learning algorithm; the drilling pressure adjustment instructions are used to control and adjust the hydraulic cylinder pressure; the drilling fluid viscosity adjustment instructions are used to control the injection of nano-sealant to adjust the rheology of the drilling fluid.

10. The apparatus as claimed in claim 9, characterized in that, The instruction generation module is specifically used for: The multimodal helicity index and mechanical drilling rate are used as state inputs to the helical wellbore control depth reinforcement learning model. The helical wellbore control depth reinforcement learning model is used to analyze the coupling relationship between downhole risk level and drilling efficiency through a proximal strategy optimization algorithm based on the helical wellbore control dataset, and to train a pre-set depth reinforcement learning model. The spiral wellbore control dataset includes: historical data of drilling pressure adjustment commands and drilling fluid viscosity adjustment commands corresponding to different multimodal helicity indices and mechanical drilling rates; Receive drilling pressure adjustment commands and drilling fluid viscosity adjustment commands corresponding to the state input quantities output by the spiral wellbore control depth reinforcement learning model.

11. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 8.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1 to 8.

13. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method of any one of claims 1 to 8.

Images