A machine learning-based torsional oscillation directional drilling drill string friction state recognition method
By using machine learning models to predict the trend of mechanical drilling rate changes, the problem of difficulty in identifying drill string friction state in torsional directional drilling was solved, enabling accurate identification of drill string friction state and optimization of torsional directional strategy, thereby improving drilling efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SINOPEC OILFIELD SERVICE CORPORATION
- Filing Date
- 2022-08-26
- Publication Date
- 2026-08-04
AI Technical Summary
In existing torsional directional drilling technology, it is difficult and the accuracy of identifying drill string friction is low, making it impossible to accurately predict the trend of mechanical drilling speed changes.
By employing machine learning methods, a machine learning model for identifying drill string friction status is constructed through the collection and processing of wellhead torque data. A BP neural network is then used to predict the changing trend of mechanical drilling speed, thereby achieving accurate identification of drill string friction status.
It enables accurate identification of drill string friction and mechanical drilling speed, assists drillers in optimizing torsion strategies, and improves the efficiency and accuracy of torsion directional drilling.
Smart Images
Figure CN117684875B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of oil and gas drilling technology, specifically to a machine learning-based method for identifying the frictional state of drill string in torsional directional drilling. Background Technology
[0002] The working principle of the drill string torsion system is to control the top drive movement to achieve clockwise and counterclockwise torsional movement of the drill string at the wellhead, converting the static friction of the upper drill string into dynamic friction, thereby reducing the loss of drilling pressure transmitted by the axial friction of the drill string. Based on the friction distribution, the drill string is divided into three zones: the "surface ripple zone" due to the torsional effect of the surface, the "static friction zone" that keeps the drill string stationary circumferentially, and the "anti-torque zone" affected by the reverse torque of the power drilling tool.
[0003] The drill string torsion system mainly consists of two parts: software and hardware. The hardware part is the ground control system, which is connected to the driller's console. The driller's console sends control commands to the top drive, controlling it to rotate alternately clockwise and counterclockwise. Simultaneously, this system needs to input information such as riser pressure, wellhead torque, tool face angle, inclination angle, and azimuth angle.
[0004] The control system software includes human-machine interface software and PLC control software. The human-machine interface software is used to set control parameters such as the peak torque of left and right swing, the impact torque of left and right impact, the wellhead torsion angle and speed of left and right swing. The system also displays drilling information such as riser pressure and downhole tool face angle.
[0005] During directional drilling, the drill string torsion system operates similarly to that used in conventional sliding drilling. First, before drilling begins, the drill string torsion system is connected to the top drive module adapter, and then connected to the MWD via a network cable to enable the control system to monitor the riser pressure and downhole tool face. Next, when the drill string has reached the bottom but the drill bit has not yet touched the bottom, the top drive is activated to rotate the drill string, and the drill string's no-load torque and riser pressure are measured. The control system's human-machine interface software is then opened, and the left and right swing limit torques are set to no more than 50% of the drill string's no-load torque. Considering the counter-torque of the balancing guide motor, the right swing limit torque must be greater than the left swing limit torque. Then, the tool face is positioned, the pump is started, and the drill string is lowered to the bottom, recording the riser pressure changes. Finally, by judging and adjusting the peak torque at the surface, the driller can determine the depth at which the drill string is oscillating. As drilling progresses, the optimal peak swing torque of the drill string needs to be dynamically detected and judged based on the engineer's experience and through trial and error.
[0006] The toolface adjustment process of the drill string torsion system differs from conventional drilling. During drilling, the drill string torsion system directly adjusts the toolface angle by "impacting" the working face to the left or right with a torque exceeding the limit value, without stopping drilling. The torque value for impacting the toolface is still determined based on engineer experience and trial and error. Because the toolface is adjusted progressively, the wellbore trajectory achieved by the drill string torsion system is smoother than that of traditional directional drilling. Standby pressure is a crucial parameter for understanding the bit pressure in the drill string torsion system. The difference in standby pressure signals reflects the changes in bit pressure at the bottom of the well. After conversion based on engineer experience, the bit pressure is obtained. The spectral data of standby pressure can identify factors unrelated to standby pressure and counter-torque, such as cuttings bed accumulation and nozzle blockage.
[0007] The existing torsional directional drilling technology faces difficulties in determining the torsional strategy and parameters. During directional drilling, the maximum torque of the drill string swinging left and right at the wellhead needs to be determined by experienced field engineers through repeated adjustments. The depth of the surface swept zone and the reverse torque parameters of the downhole drill string need to be estimated by engineers based on experience, making it impossible to optimize drilling parameters. During tool face adjustment, the control strategy of "bumping" the left and right limit torques is too conservative and does not fully tap the potential of the technology. Summary of the Invention
[0008] The purpose of this invention is to provide a machine learning-based method for identifying the frictional state of drill string in torsion directional drilling, in order to solve the problem that existing technologies cannot accurately predict the changing trend of mechanical drilling speed, resulting in difficulties and low accuracy in identifying the frictional state of drill string in torsion directional drilling.
[0009] To solve the above-mentioned technical problems, the technical solution provided by this invention and the corresponding beneficial effects of the technical solution are as follows: The present invention provides a machine learning-based method for identifying the frictional state of drill string in torsional directional drilling, comprising the following steps: During torsional directional drilling operations, the wellhead torque at the current moment and the previous N-1 moments is obtained, and the wellhead torque at the current moment and the previous N-1 moments is converted into the corresponding wellhead torque sequence and input into the drill string friction state recognition machine learning model for recognition, so as to predict the change trend of mechanical drilling speed at the next moment and determine the change of drill string friction state, where N is greater than 1. The drill string friction state recognition machine learning model is obtained through training in the following manner: S1. Collect wellhead torque data during torsional directional drilling operations, including torque and acquisition time; S2. Extract several wellhead torque sequences from the wellhead torque data. Each wellhead torque sequence contains the wellhead torque at N consecutive acquisition times. S3. Mark the status of each wellhead torque sequence, including marking the trend of mechanical drilling rate change, wherein the trend of mechanical drilling rate change includes an upward trend and a downward trend; S4. The labeled state of the wellhead torque sequence and the corresponding mechanical drilling rate change trend is used as training data to train the constructed machine learning model to obtain the drill string friction state recognition machine learning model.
[0010] The beneficial effects of the above technical solution are as follows: Multiple wellhead torque sequences are input into a pre-constructed machine learning model for training, resulting in a drill string friction state recognition machine learning model. Then, the wellhead torque sequences corresponding to the current moment and the previous N-1 moments are input into the drill string friction state recognition machine learning model for identification, predicting the trend of mechanical drilling speed. Based on the trend of mechanical drilling speed, the state of drill string friction is determined: an increase in mechanical drilling speed leads to an increase in drill string friction, and a decrease in mechanical drilling speed leads to a decrease in drill string friction, thus determining the drill string friction state. This achieves accurate identification of the drill string friction state and the trend of mechanical drilling speed, enabling real-time monitoring and optimization of wellhead torque. It solves the problems of difficult and low-accuracy identification of drill string friction state in torsion directional drilling, further assisting drillers in adjusting torsion strategies for torsion directional drilling.
[0011] Furthermore, the wellhead torque is obtained by the following method: obtaining the inverter torque of the top drive, and correcting the inverter torque to obtain the wellhead torque.
[0012] Furthermore, before step S4, each wellhead torque sequence needs to be normalized and the labeling status of the corresponding wellhead torque sequence needs to be encoded.
[0013] Furthermore, the method for determining the trend of mechanical drilling speed change in step S3 is as follows: A linear expression is fitted to the mechanical drilling rate curve, and the trend of mechanical drilling rate is obtained based on the slope of the linear expression; if the slope of the linear expression is greater than 0, it is determined to be an upward trend; if the slope of the linear expression is less than 0, it is determined to be a downward trend.
[0014] Furthermore, an upward trend in mechanical drilling speed is marked as 1, and a downward trend in mechanical drilling speed is marked as 0.
[0015] Furthermore, the status labeling of each wellhead torque sequence in step S3 also includes labeling erroneous data; the labeling method for erroneous data is as follows: for each wellhead torque sequence, erroneous data is judged, and if there is no erroneous data, it is marked as 0; if there is erroneous data, it is marked as the sequence number of the erroneous data in the wellhead torque sequence.
[0016] Furthermore, the machine learning model described in step S4 is a BP neural network model.
[0017] Furthermore, the encoding process is a Onehot encoding process.
[0018] Furthermore, the normalization process involves dividing the wellhead torque by the stable wellhead torque; the stable wellhead torque is obtained using the following method: Raise the drill bit from the bottom of the well and rotate the drill string until the inverter torque is stable. Record the stable inverter torque. Correct the stable inverter torque to obtain the stable wellhead torque. Attached Figure Description
[0019] Figure 1 This is a flowchart illustrating the implementation of a machine learning-based well drill string friction state identification method in an embodiment of the present invention. Detailed Implementation
[0020] Method Implementation Examples: The overall approach of this embodiment is as follows: This invention discloses a machine learning-based method for identifying drill string friction states in torsional directional drilling. It collects the torque and rotation angle of the top drive motor inverter during torsional directional drilling operations, along with the time of rotation, and constructs a discrete state model of the inverter torque. Then, it establishes an inverter moment of inertia model and corrects the inverter torque to obtain the actual wellhead torque. Based on the wellhead unloaded torque when the drill bit is lifted from the bottom of the well, it performs wellhead torque normalization to obtain the normalized wellhead torque. The wellhead torque sequence is then labeled with its state, and the labeled states of the wellhead torque sequence are combined using Onehot encoding to obtain the state code corresponding to each wellhead torque. The wellhead torque sequence and the corresponding state code are used as training data; a pre-set BP neural network is trained to obtain a drill string friction state identification model. Finally, during the torsion directional drilling operation, the normalized wellhead torque sequence at the current moment is obtained. This normalized wellhead torque sequence is then input into a drill string friction state recognition machine learning model for identification, resulting in a state code. This state code is then compared with the corresponding state codes mentioned above to obtain the trend of mechanical drilling rate changes, thereby determining the current drill string friction state. Therefore, this invention can accurately identify the wellhead torque state, drill string friction state, and reasonable operating parameters in torsion directional drilling, achieving real-time monitoring of the drill string friction state in torsion directional drilling.
[0021] To achieve the above-mentioned objectives, the drill string friction state identification method based on a BP neural network of this invention is divided into two stages: (1) establishing a drill string friction state identification machine learning model; (2) applying the trained drill string friction state identification machine learning model to perform online state identification, such as... Figure 1 As shown, the specific steps include the following: Phase 1: Establishing a machine learning model for identifying drill string friction conditions.
[0022] S1: Keep the top drive unloaded, control the top drive to complete reciprocating motion, record the inverter torque, identify the rotational inertia of the top drive, i.e., the top drive is unloaded, add a speed disturbance signal, record the speed and inverter torque, and calculate the motor rotational inertia; lift the drill bit off the bottom of the well, rotate the drill string until the inverter torque is in a stable state, and record the stable inverter torque.
[0023] S2: Construct a rotational inertia model of the inverter torque to correct the inverter torque to obtain the wellhead torque; the wellhead torque is equal to the inverter torque minus the no-load torque, where the inverter torque can be read by the industrial control computer, and the no-load torque refers to the inverter torque when the top drive is unloaded.
[0024] The stable inverter torque is corrected using a moment of inertia model to obtain a stable wellhead torque, which is then used as a parameter for subsequent normalization. Specifically, the moment of inertia model is: Torque / Acceleration = Moment of Inertia. After measuring the rotational speed and torque in the time domain, the moment of inertia is obtained through linear fitting.
[0025] S3: Collect wellhead torque data for torsional directional drilling, including time series data of the left and right swing processes; mechanical drilling rate data also needs to be collected.
[0026] S4: Extract several wellhead torque sequences from the wellhead torque data. Each wellhead torque sequence contains wellhead torques that are consecutive or spaced N times apart; N is greater than 1.
[0027] S5: Label the status of each wellhead torque sequence, including erroneous data and the trend of mechanical drilling rate changes. The specific method is as follows: Error data is identified in the wellhead torque sequence to improve the accuracy of the drill string friction state identification machine learning model. If no error data exists, it is marked as 0. If error data exists, it is marked as the sequence number of the error data in the wellhead torque sequence. The above error data is abrupt data, such as individual data that differ greatly from the data before and after it.
[0028] To determine the trend of mechanical drilling speed, during the above wellhead torque sequence, if the mechanical drilling speed is on an upward trend, it is marked as 1; if the mechanical drilling speed is on a downward trend, it is marked as 0.
[0029] Furthermore, the method for determining the trend of mechanical drilling speed is as follows: A linear expression is fitted to the mechanical drilling speed curve, and the trend of the mechanical drilling speed is obtained based on the slope of the linear expression. If the slope of the linear expression is greater than 0, the mechanical drilling speed is considered to be on an upward trend, and the label status is 1; if the slope of the linear expression is less than 0, the mechanical drilling speed is considered to be on a downward trend, and the label status is 0.
[0030] The start and end times of the mechanical drilling rate change trend are fixed periods, and the wellhead torque sequence is sampled once every period.
[0031] S6: Normalize each wellhead torque sequence to the range [0, 1] to obtain a normalized wellhead torque sequence. The wellhead torque sequence normalization is performed using Min-Max normalization. Specifically, normalization is to divide each wellhead torque in each wellhead torque sequence by the stable wellhead torque in step S2 above.
[0032] S7: The labeled states of the wellhead torque sequence of torsion directional drilling are Onehot encoded according to their combination, including the labeled states of the mechanical drilling rate change trend, to obtain the state code of the mechanical drilling rate change trend corresponding to each torsion directional drilling wellhead torque sequence.
[0033] S8: Construct a BP neural network model, take the normalized wellhead torque sequence as input, and the state code of the corresponding mechanical drilling rate change trend as the expected output, train the BP neural network model to obtain the drill string friction state recognition machine learning model.
[0034] In another embodiment, the present invention may also use other neural networks in the prior art, such as RBF neural networks.
[0035] The second stage involves applying the trained machine learning model for drill string friction state recognition to perform online state identification. Based on the obtained trend of mechanical drilling speed changes, the trend of drill string friction changes is determined: as the mechanical drilling speed increases, drill string friction increases; as the mechanical drilling speed decreases, drill string friction decreases.
[0036] S9: During torsion directional drilling, the wellhead torque sequences corresponding to the current and previous moments are normalized according to the method in S6. Then, the normalized wellhead torque sequences are input into the drill string friction state recognition machine learning model trained in S8 to obtain the trend of mechanical drilling speed. Based on the trend of mechanical drilling speed, the state of drill string friction is determined. When the mechanical drilling speed increases, the drill string friction increases, and when the mechanical drilling speed decreases, the drill string friction decreases. Therefore, based on the trend of mechanical drilling speed, it is determined whether the current maximum torque of torsion drilling at the wellhead is reasonable, further assisting the driller in adjusting the torsion strategy for torsion directional drilling.
[0037] In another implementation, the wellhead torque sequence at the current moment and the previous N moments are used as input data for the drill string friction state recognition machine learning model, where N≥1. The sequence is normalized according to the method in S6, and then input into the drill string friction state recognition machine learning model trained in S8 to obtain the trend of mechanical drilling speed. Since a larger maximum torque results in smaller drill string friction and a faster mechanical drilling speed, and vice versa, the trend of mechanical drilling speed can be used to determine whether the current maximum torque for torsional drilling at the wellhead is reasonable, further assisting the driller in adjusting the torsional strategy for torsional directional drilling.
[0038] The purpose of this invention is to overcome the shortcomings of existing torsional directional drilling technology and provide a machine learning-based method for identifying drill string friction status. This involves constructing a variable frequency drive (VFD) moment of inertia model, and then using One-hot encoding to encode the mechanical drilling rate (MDR) state and building a backpropagation (BP) neural network. After training, a network for identifying the VFD torque state is obtained, which can accurately identify the drill string friction status and the changing trend of the MDR, enabling real-time monitoring and optimization of wellhead torque. This solves the problems of difficult and low-accuracy drill string friction status identification in torsional directional drilling, further assisting drillers in torsional directional drilling. This invention uses a BP neural network, which can learn the characteristics of drill string friction status based on a large amount of historical data, avoiding oversimplification of the physical model of drill string friction and the difficulty in picking key parameters. It is a convenient and effective method for identifying drill string friction status, capable of identifying drill string friction status and MDR trends, reducing the influence of human factors on drill string friction status judgment. This invention has strong versatility; the machine learning-based torsional directional drilling drill string friction status identification method is compatible with current mainstream MWD instruments.
Claims
1. A machine learning-based method for identifying the frictional state of a drill string in torsional directional drilling, characterized in that, Includes the following steps: During torsional directional drilling operations, the wellhead torque at the current moment and the previous N-1 moments is obtained, and the wellhead torque at the current moment and the previous N-1 moments is converted into the corresponding wellhead torque sequence and input into the drill string friction state recognition machine learning model for recognition, so as to predict the change trend of mechanical drilling speed at the next moment and determine the change of drill string friction state, where N is greater than 1. The drill string friction state recognition machine learning model is obtained through training in the following manner: S1. Collect wellhead torque data during torsional directional drilling operations, including torque and acquisition time; S2. Extract several wellhead torque sequences from the wellhead torque data. Each wellhead torque sequence contains the wellhead torque at N consecutive acquisition times. S3. Mark the status of each wellhead torque sequence, including marking the status of the mechanical drilling rate change trend. The mechanical drilling rate change trend includes an upward trend and a downward trend. The determination method is as follows: fit a linear expression to the mechanical drilling rate curve, and obtain the mechanical drilling rate change trend based on the slope of the linear expression; if the slope of the linear expression is greater than 0, it is determined to be an upward trend; if the slope of the linear expression is less than 0, it is determined to be a downward trend. S4. Normalize each wellhead torque sequence and encode the labeling status of the corresponding wellhead torque sequence; S5. The labeled state of the wellhead torque sequence and the corresponding mechanical drilling speed change trend is used as training data to train the constructed machine learning model to obtain the drill string friction state recognition machine learning model.
2. The method for identifying the frictional state of a torsional directional drilling drill string based on machine learning according to claim 1, characterized in that, The wellhead torque is obtained by the following method: obtaining the inverter torque of the top drive, and correcting the inverter torque to obtain the wellhead torque.
3. The method for identifying the frictional state of a torsional directional drilling drill string based on machine learning according to claim 1, characterized in that, The start and end times of the mechanical drilling rate change trend are fixed periods, and the wellhead torque sequence is sampled once every period.
4. The machine learning-based method for identifying the frictional state of a drill string in torsional directional drilling as described in claim 1, characterized in that, The relationship between the change in drill string friction and the mechanical drilling rate is as follows: as the mechanical drilling rate increases, the drill string friction increases; as the mechanical drilling rate decreases, the drill string friction decreases.
5. The method for identifying the frictional state of a torsional directional drilling drill string based on machine learning according to claim 1, characterized in that, An upward trend in mechanical drilling speed is marked as 1, and a downward trend in mechanical drilling speed is marked as 0.
6. The method for identifying the frictional state of a torsional directional drilling drill string based on machine learning according to claim 1, characterized in that, The status labeling of each wellhead torque sequence in step S3 also includes labeling erroneous data; the method for labeling erroneous data is as follows: each wellhead torque sequence is judged for erroneous data, and if there is no erroneous data, it is marked as 0; If erroneous data exists, it is marked as the sequence number of the erroneous data in the wellhead torque sequence.
7. The method for identifying the frictional state of a torsional directional drilling drill string based on machine learning according to claim 1, characterized in that, The machine learning model mentioned in step S5 is a BP neural network model.
8. The method for identifying the frictional state of a torsional directional drilling drill string based on machine learning according to claim 3, characterized in that, The encoding process is Onehot encoding.
9. The machine learning-based method for identifying the frictional state of a drill string in torsional directional drilling, as described in claim 3 or 8, is characterized in that... The normalization process involves dividing the wellhead torque by the stable wellhead torque; the stable wellhead torque is obtained using the following method: Raise the drill bit from the bottom of the well and rotate the drill string until the inverter torque is stable. Record the stable inverter torque. Correct the stable inverter torque to obtain the stable wellhead torque.