A method for computing a dynamic threshold based on an MLP network for evaluating a risk of a stuck pipe
By combining real-time logging data and soft rod models, and using MLP neural networks to predict the dynamic threshold of stuck pipe risk, the problems of strong subjectivity and high false alarm rate in traditional methods are solved, achieving efficient and accurate assessment of stuck pipe risk and improving the safety and efficiency of drilling operations.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CNOOC ENERGY TECHNOLOGY & SERVICES LTD
- Filing Date
- 2026-03-09
- Publication Date
- 2026-06-05
AI Technical Summary
Traditional stuck pipe risk assessment methods rely on mechanistic models and field experience, which are highly subjective and cannot reflect changes in downhole conditions in real time and accurately, resulting in a high false alarm rate and difficulty in effectively identifying stuck pipe risks during the drilling process.
A dynamic threshold assessment method based on MLP network is adopted, which combines real-time logging data and soft rod model to calculate the theoretical values of hook load and torque, and predicts dynamic threshold through MLP neural network, and performs real-time assessment in combination with risk factors.
It enables efficient and accurate assessment of stuck drill bit risks, improves the safety and efficiency of drilling operations, and provides a scientific basis for decision-making.
Smart Images
Figure CN122155419A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of oil and gas well drilling technology, and in particular relates to a stuck pipe risk assessment method based on dynamic threshold calculation using MLP network. Background Technology
[0002] In the field of oil and gas exploration and development, drilling operations are a complex and high-risk activity. Especially in deep wells, ultra-deep wells, and offshore drilling, factors such as uncertain geological conditions, abnormal formation pressure, and poor wellbore stability can easily lead to serious accidents such as stuck pipe. This can not only result in huge economic losses but also endanger the lives of the workers.
[0003] Stuck drill string refers to the phenomenon where the drill string cannot move freely due to excessive mechanical resistance during drilling or tripping. The mechanism of stuck drill string is complex and has many causes. The formation rules of stuck drill string vary depending on the well type and well depth.
[0004] Traditional methods for assessing stuck pipe risk primarily rely on mechanistic model analysis and subjective judgment from field personnel. Both methods assess the presence of stuck pipe risk based on the changing trends of real-time logging parameters, such as hook load and torque. However, the current parameter risk thresholds still require subjective determination by field experts based on historical well data, which is highly subjective, has poor transferability, and cannot accurately reflect changes in downhole conditions in real time. Therefore, these methods have a high false alarm rate.
[0005] In recent years, with the development of computer technology and artificial intelligence algorithms, data-driven intelligent prediction methods have gradually become a research hotspot. Multilayer Perceptron (MLP) neural networks, as a classic machine learning algorithm, possess complex nonlinear mapping capabilities. By learning parameter mapping relationships from a large amount of historical data, they can be used to establish threshold prediction models for stuck drill risk parameters, enabling dynamic assessment of stuck drill risk and further improving the effectiveness of stuck drill assessment. Summary of the Invention
[0006] The problem this invention aims to solve is to provide a method for assessing stuck pipe risk based on dynamic thresholds calculated using an MLP network. This method collects various parameters during the drilling process in real time, analyzes the stress state of the drill string using a soft rod model, calculates the theoretical values of hook load and torque under no stuck pipe risk, combines the actual values of hook load and torque from logging parameters, and uses an MLP neural network to predict the dynamic thresholds of hook load and torque under the current state. This achieves efficient and accurate assessment of stuck pipe risk, significantly improving the safety and efficiency of drilling operations, and has important theoretical significance and broad application prospects.
[0007] To solve the above-mentioned technical problems, the technical solution adopted by this invention is: a stuck drill risk assessment method based on dynamic threshold calculation using an MLP network, comprising the following steps: S1: Collect real-time logging data on-site; S2: Calculate the theoretical values of the hook load and torque using the soft rod model; S3: Using an MLP neural network model, combined with the logging data, the theoretical value of the hook load, and the theoretical value of the torque, predict the dynamic threshold of the hook load and the dynamic threshold of the torque. S4: Calculate the risk factor for stuck drill bit; S5: Monitor drilling status in real time and issue timely warnings based on the values of the risk factors to provide scientific decision-making basis for on-site operators.
[0008] Furthermore, in S1, the logging data includes drill bit position, hook height, actual hook load value, drilling pressure, actual torque value, drilling speed, rotary table speed, and standpipe pressure.
[0009] Furthermore, S2 includes the following steps: S21: Based on the wellbore trajectory, well structure, and key parameters of the drill string assembly of the target well, establish the soft rod model that matches the target well; S22: Using the logging data from the site, calculate the theoretical values of the hook load and torque when there is no risk of stuck drill bit using the soft rod model.
[0010] Furthermore, in S21, the flexible rod model is established based on the following differential equation, (1) (2) (3) Where F is the axial force of the drill string, N; s is the well depth, m; q is the buoyancy of the drill string per unit length, N / m; The average wellbore inclination angle at this micro-element segment is expressed in rad. In the diagram, a negative sign indicates that the drill string is pulled up, and a positive sign indicates that the drill string is lowered. This is the axial friction coefficient of the drill string; The lateral force per unit length between the drill string and the wellbore, in N / m; The torque on the drill string is N·m; The coefficient of friction of the drill string in the circumferential direction; Where is the outer diameter of the drill string, in meters; Let N be the axial force at the lower end of this infinitesimal segment; The borehole azimuth variation at the micro-element segment is expressed in rad. The value is rad, representing the change in borehole inclination angle at the micro-element segment.
[0011] Furthermore, the differential equation is transformed into a difference equation using the finite difference method for iterative solution to obtain the theoretical values of the hook load and the torque. The difference equation is as follows: (4) (5) (6) (7) in, , , respectively, represent the upper and lower axial forces of the i-th drill string micro-element, in N; q is the buoyancy of the drill string micro-element, in N / m; The lateral force of the drill string in a micro-element segment is N / m; Let be the length of the drill string micro-segment, in meters; , These are the upper and lower torques of the drill string micro-segment, respectively, in N·m; , The inclination angle between the upper and lower parts of the drill string, in rad; , The azimuth angles between the upper and lower parts of the drill string are in rad.
[0012] Furthermore, in S3, the MLP neural network model includes an input layer, a hidden layer, and an output layer. The input layer is used to input key parameters, and the number of neurons in the input layer is equal to the number of input features. The hidden layer is used for deep computation of the input key parameters and extracts dynamic threshold key features through nonlinear transformation. The number of hidden layers and the number of neurons in each layer depend on the complexity of the task. The optimal combination of hidden layers and the number of neurons is obtained through training with a large number of hyperparameter combinations. The output layer outputs results according to task requirements, and the output layer outputs the dynamic threshold of the hook load and the dynamic threshold of the torque.
[0013] Furthermore, an activation function is added to each neuron to introduce nonlinear factors, enabling the MLP neural network model to learn nonlinear relationships. Mean squared error is introduced as a loss function during the MLP neural network model training process to estimate the difference between predicted and true values, guiding the update of network weights. A time window is introduced into the MLP neural network model, allowing it to further consider temporal properties and better conform to the dynamic evolution characteristics of stuck pipe risk. After the model is built, it is trained with a large number of drilling parameter samples, enabling the MLP neural network model to adaptively output the optimal stuck pipe risk threshold corresponding to the current real-time parameters.
[0014] Furthermore, the input features of the MLP neural network model also include the current well depth. After training, the MLP neural network model automatically learns the mapping relationship between well depth and stuck pipe risk threshold, so that the output hook load dynamic threshold and torque dynamic threshold are dynamically adjusted as the well depth changes.
[0015] Furthermore, in step S3, when training the MLP neural network model, the dataset is divided into a training set, a validation set, and a test set. The MLP neural network model is trained using the training set, the hyperparameters are adjusted using the validation set, and the generalization ability of the MLP neural network model is evaluated using the test set.
[0016] Furthermore, in S4, when the difference between the actual value of the hook load and the theoretical value of the hook load exceeds the hook load dynamic threshold, or when the difference between the actual value of the torque and the theoretical value of the torque exceeds the torque dynamic threshold, the risk factor increases by 1; when the difference between the actual value of the hook load and the theoretical value of the hook load exceeds the hook load dynamic threshold, and simultaneously, the difference between the actual value of the torque and the theoretical value of the torque exceeds the torque dynamic threshold, the risk factor increases by 2.
[0017] The advantages and positive effects of this invention are: This invention combines real-time logging data with a flexible rod model to calculate the theoretical values of hook load and torque. Simultaneously, it incorporates an MLP model to automatically predict dynamic thresholds for stuck pipe risk and introduces a stuck pipe risk factor, thereby dynamically assessing the risk. This method boasts advantages such as high calculation accuracy and strong adaptability. Specifically, by dynamically adjusting thresholds using an MLP neural network, it accurately reflects the actual downhole conditions under different well depths and real-time parameters, effectively identifying stuck pipe risks during drilling. This method provides field operators with a scientific basis for decision-making, improving drilling efficiency and safety. Attached Figure Description
[0018] Figure 1 This is a schematic diagram of the overall process of an embodiment of the present invention.
[0019] Figure 2 This is a flowchart of the iterative solution of the hook load and torque using the flexible rod model in an embodiment of the present invention.
[0020] Figure 3 This is a schematic diagram of the force analysis of a certain infinitesimal element in the flexible rod model of this invention.
[0021] Figure 4 This is a schematic diagram of the MLP neural network model structure according to an embodiment of the present invention. Detailed Implementation
[0022] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0023] The embodiments of the present invention will be further described below with reference to the accompanying drawings: like Figure 1 As shown, the method for assessing stuck drill risk based on dynamic threshold calculation using an MLP network includes the following steps.
[0024] S1: Collect real-time logging data. Specifically, the logging data includes drill bit position, hook height, actual hook load, drilling pressure, actual torque, drilling speed, rotary table speed, and standpipe pressure.
[0025] S2: Calculate the theoretical hook load and theoretical torque using a flexible rod model. Specifically, the flexible rod model, as an effective mechanical analysis model for the drill string, can calculate the stress on the drill string under different operating conditions, providing a mechanistic basis for stuck pipe risk assessment. In the construction of the flexible rod model and the solution of theoretical parameters, a suitable flexible rod model for the well is first established based on key parameters such as the wellbore trajectory, well structure, and drill string assembly. Subsequently, using field logging data, the theoretical torque and hook load values are accurately calculated. The differential equation of the flexible rod model is as follows: (1) (2) (3) Where F is the axial force of the drill string, N; s is the well depth, m; q is the buoyancy of the drill string per unit length, N / m; The average wellbore inclination angle at this micro-element segment is expressed in rad. In the diagram, a negative sign indicates that the drill string is pulled up, and a positive sign indicates that the drill string is lowered. This is the axial friction coefficient of the drill string; The lateral force per unit length between the drill string and the wellbore, in N / m; The torque on the drill string is N·m; The coefficient of friction of the drill string in the circumferential direction; Where is the outer diameter of the drill string, in meters; Let N be the axial force at the lower end of this infinitesimal segment; The borehole azimuth variation at the micro-element segment is expressed in rad. The value is rad, representing the change in borehole inclination angle at the micro-element segment.
[0026] Preferably, the finite difference method can be further used to transform the above differential equation into a difference equation: (4) (5) (6) (7) in, , , respectively, represent the upper and lower axial forces of the i-th drill string micro-element, in N; q is the buoyancy of the drill string micro-element, in N / m; The lateral force of the drill string in a micro-element segment is N / m; Let be the length of the drill string micro-segment, in meters; , These are the upper and lower torques of the drill string micro-segment, respectively, in N·m; , The inclination angle between the upper and lower parts of the drill string, in rad; , The azimuth angles between the upper and lower parts of the drill string are in rad.
[0027] Through the above iterative solution process, the theoretical values of the ground hook load and torque under no-stuck-drill-risk conditions can be calculated, providing a data foundation for stuck-drill-risk assessment. The iterative calculation process is as follows: Figure 2 As shown, the force analysis of a certain infinitesimal element in the flexible rod model is as follows: Figure 3 As shown.
[0028] S3: Using an MLP neural network, combined with logging data, theoretical hook load, and theoretical torque, dynamically predicts the stuck pipe risk thresholds for hook load and torque, such as... Figure 4 As shown.
[0029] Specifically, MLP is a feedforward artificial neural network model capable of nonlinearly mapping input data to calculate and predict target values. The MLP model includes an input layer, hidden layers, and an output layer. The input layer is used to input key parameters, such as parameters in logging data that are strongly correlated with the dynamic threshold of stuck drill risk. The hidden layer is used to calculate the depth of the input parameters and extract key features of the dynamic threshold through nonlinear transformation. The number of hidden layers and the number of neurons in each layer can be trained and adjusted based on sample data to obtain the optimal combination. The output layer outputs results according to task requirements; in this case, it outputs the dynamic thresholds of hook load and torque. An activation function is added to each neuron to introduce nonlinear factors, enabling the network to learn nonlinear relationships. Mean squared error (MSE) is introduced as a loss function during neural network training to estimate the difference between the predicted and true values, guiding the update of network weights. A gradient optimizer is used to update the weights in the network using the loss function, thereby obtaining model parameters that minimize the error between the predicted and true values. Since stuck drill risk is a time-series problem that evolves dynamically over time, a time window is introduced into the MLP, allowing the model to further consider temporal properties and better reflect the dynamic evolution characteristics of stuck drill risk. Once the model is built, it can be trained with a large number of drilling parameter samples, enabling the MLP to adaptively output the optimal stuck pipe risk threshold corresponding to the current real-time parameters.
[0030] Preferably, the dynamic threshold calculation in this embodiment takes into account the influence of well depth, enabling the threshold to be dynamically adjusted as the well depth changes. Specifically, well depth is used as an input feature of the model, allowing the neural network to automatically learn the relationship between well depth and stuck pipe risk. After model training, the output threshold will include the influence of well depth. By using well depth as an input feature of the neural network, the model automatically establishes a mapping relationship of "well depth → risk probability" under data-driven conditions. This makes the model more sensitive to risk at different well depths—deeper wells are more likely to trigger risk warnings, thereby achieving the effect of "threshold adjustment with well depth".
[0031] In terms of data processing, the raw logging data is first cleaned and normalized to ensure that the data quality meets the training requirements of the MLP neural network model. Key factors affecting the hook load and torque dynamic thresholds are identified through expert experience and correlation testing, thereby selecting model input parameters, including 10 parameters: drill bit position, hook height, actual hook load, theoretical hook load, drilling pressure, actual torque, theoretical torque, drilling speed, rotary table speed, and standpipe pressure.
[0032] In terms of MLP neural network architecture design, the dimension of the input layer is first determined, and the number of neurons in the input layer equals the number of input features. The number of hidden layers and the number of neurons in each layer depend on the complexity of the task, and the optimal combination of hidden layers and the number of neurons can be obtained through training with a large number of hyperparameter combinations. For the activation function, ReLU (Rectified Linear Unit) was used because ReLU is suitable for most intermediate layers and can effectively solve the gradient vanishing problem.
[0033] For model training and testing, the dataset was first divided into training, validation, and test sets to fully evaluate model performance and prevent overfitting. Then, the MLP model was trained using the training set data, and hyperparameters were tuned using the validation set. Next, the model's generalization ability and prediction accuracy were evaluated using the test set. Finally, the trained model was deployed to predict dynamic thresholds for hook load and torque based on real-time logging data.
[0034] S4: Define the risk factor based on the hook load and torque.
[0035] Specifically, when the difference between the actual hook load and the theoretical hook load exceeds the dynamic threshold for hook load, or when the difference between the actual torque and the theoretical torque exceeds the dynamic threshold for torque, the risk factor increases by 1. When the difference between the actual hook load and the theoretical hook load exceeds the dynamic threshold for hook load, and when the difference between the actual torque and the theoretical torque simultaneously exceeds the dynamic threshold for torque, the risk factor increases by 2. The calculation of the risk factor considers two key indicators: hook load and torque. It can comprehensively assess the circumferential and axial resistance status of the drill string during drilling, and reflect the potential sticking risk in a timely and accurate manner. Because it combines the dynamic threshold predicted by the MLP model, the risk factor can automatically adjust according to the changing trends of real-time parameters such as well depth and standpipe pressure, improving the accuracy of sticking risk assessment.
[0036] Preferably, the risk factors in this embodiment consider both the axial and circumferential stresses on the drill string, which can effectively identify abnormal situations during the drilling process and improve drilling safety.
[0037] S5: Real-time monitoring of drilling status, timely issuance of early warnings based on risk factor values, providing scientific decision-making basis for on-site operators.
[0038] The advantages and positive effects of this invention are: This invention combines real-time logging data with a flexible rod model to calculate the theoretical values of hook load and torque. Simultaneously, it incorporates an MLP model to automatically predict dynamic thresholds for stuck pipe risk and introduces a stuck pipe risk factor, thereby dynamically assessing the risk. This method boasts advantages such as high calculation accuracy and strong adaptability. Specifically, by dynamically adjusting thresholds using an MLP neural network, it accurately reflects the actual downhole conditions under different well depths and real-time parameters, effectively identifying stuck pipe risks during drilling. This method provides field operators with a scientific basis for decision-making, improving drilling efficiency and safety.
[0039] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.
Claims
1. A method for assessing stuck drill risk based on dynamic threshold calculation using MLP networks, characterized in that: Includes the following steps, S1: Collect real-time logging data on-site; S2: Calculate the theoretical values of the hook load and torque using the soft rod model; S3: Using an MLP neural network model, combined with the logging data, the theoretical value of the hook load, and the theoretical value of the torque, predict the dynamic threshold of the hook load and the dynamic threshold of the torque. S4: Calculate the risk factor for stuck drill bit; S5: Monitor drilling status in real time and issue timely warnings based on the values of the risk factors to provide scientific decision-making basis for on-site operators.
2. The stuck drill risk assessment method based on MLP network dynamic threshold calculation according to claim 1, characterized in that: In S1, the logging data includes drill bit position, hook height, actual hook load, drilling pressure, actual torque, drilling speed, rotary table speed, and standpipe pressure.
3. The method for assessing stuck drill risk based on dynamic threshold calculation using an MLP network according to claim 1 or 2, characterized in that: In S2, Includes the following steps, S21: Based on the wellbore trajectory, well structure, and key parameters of the drill string assembly of the target well, establish the soft rod model that matches the target well; S22: Using the logging data from the site, calculate the theoretical values of the hook load and torque when there is no risk of stuck drill bit using the soft rod model.
4. The stuck drill risk assessment method based on MLP network dynamic threshold calculation according to claim 3, characterized in that: In step S21, the flexible rod model is established based on the following differential equation: （1） （2） （3） Where F is the axial force of the drill string, N; s is the well depth, m; q is the buoyancy of the drill string per unit length, N / m; The average wellbore inclination angle at this micro-element segment is expressed in rad. In the diagram, a negative sign indicates that the drill string is pulled up, and a positive sign indicates that the drill string is lowered. This is the axial friction coefficient of the drill string; The lateral force per unit length between the drill string and the wellbore, in N / m; The torque on the drill string is N·m; The coefficient of friction of the drill string in the circumferential direction; Where is the outer diameter of the drill string, in meters; Let N be the axial force at the lower end of this infinitesimal segment; The borehole azimuth variation at the micro-element segment is expressed in rad. The value is rad, representing the change in borehole inclination angle at the micro-element segment.
5. The stuck drill risk assessment method based on MLP network dynamic threshold calculation according to claim 4, characterized in that: The differential equation is transformed into a difference equation using the finite difference method for iterative solution to obtain the theoretical values of the hook load and the torque. The equations are as follows: （4） （5） （6） （7） in, , Let N be the upper and lower axial forces of the i-th drill string micro-element segment, respectively. q is the lateral force per unit length between the drill string and the wellbore, N / m; q is the buoyancy of the drill string per unit length, N / m. Let be the length of the drill string micro-segment, in meters; , These are the upper and lower torques of the drill string micro-segment, respectively, in N·m; , The inclination angle between the upper and lower parts of the drill string, in rad; , The azimuth angles between the upper and lower parts of the drill string are in rad.
6. The method for assessing stuck drill risk based on dynamic threshold calculation using an MLP network according to claim 1 or 2, characterized in that: In S3, the MLP neural network model includes an input layer, a hidden layer, and an output layer. The input layer is used to input key parameters, and the number of neurons in the input layer is equal to the number of input features. The hidden layer is used for deep computation of the input key parameters and extracts dynamic threshold key features through nonlinear transformation. The number of hidden layers and the number of neurons in each layer depend on the complexity of the task. The optimal combination of hidden layers and the number of neurons is obtained through training with a large number of hyperparameter combinations. The output layer outputs results according to task requirements, and the output layer outputs the dynamic threshold of the hook load and the dynamic threshold of the torque.
7. The method for assessing stuck drill risk based on dynamic threshold calculation using an MLP network according to claim 6, characterized in that: An activation function is added to each neuron to introduce nonlinear factors, enabling the MLP neural network model to learn nonlinear relationships. Mean squared error is introduced as a loss function during the training process of the MLP neural network model to estimate the difference between the predicted and actual values, guiding the update of network weights. A time window is introduced into the MLP neural network model, allowing it to further consider temporal properties and better conform to the dynamic evolution characteristics of stuck pipe risk. After the model is built, it is trained with a large number of drilling parameter samples, enabling the MLP neural network model to adaptively output the optimal stuck pipe risk threshold corresponding to the current real-time parameters.
8. The method for assessing stuck drill risk based on dynamic threshold calculation using an MLP network according to claim 6, characterized in that: The input features of the MLP neural network model also include the current well depth. After training, the MLP neural network model automatically learns the mapping relationship between well depth and stuck pipe risk threshold, so that the output hook load dynamic threshold and torque dynamic threshold are dynamically adjusted as the well depth changes.
9. The method for assessing stuck drill risk based on dynamic threshold calculation using an MLP network according to claim 6, characterized in that: In step S3, when training the MLP neural network model, the dataset is divided into a training set, a validation set, and a test set. The MLP neural network model is trained using the training set, the hyperparameters are adjusted using the validation set, and the generalization ability of the MLP neural network model is evaluated using the test set.
10. The method for assessing stuck drill risk based on dynamic threshold calculation using an MLP network according to claim 1 or 2, characterized in that: In S4, when the difference between the actual value of the hook load and the theoretical value of the hook load exceeds the dynamic threshold of the hook load, or when the difference between the actual value of the torque and the theoretical value of the torque exceeds the dynamic threshold of the torque, the risk factor increases by 1; when the difference between the actual value of the hook load and the theoretical value of the hook load exceeds the dynamic threshold of the hook load, and simultaneously, the difference between the actual value of the torque and the theoretical value of the torque exceeds the dynamic threshold of the torque, the risk factor increases by 2.