A method and system for identifying drill string vibration conditions through surface and downhole data collaboration

Abstract

This invention discloses a method and system for identifying drill string vibration conditions through surface and downhole data collaboration, relating to the field of intelligent drilling technology. The method includes: inputting drilling engineering parameters, including multi-node monitoring data from both the surface and downhole, and analyzing the vibration distribution pattern of the drill string throughout the well section; employing modal decomposition to decouple drill string vibration into independent single-order modal components, including mode shapes and modal coefficients; and building a multi-input multi-output machine learning model, using surface and downhole parameters as inputs and modal coefficients as outputs, to establish data collaborative correlation. This invention, through a full-well section vibration physical model, modal decomposition, and machine learning technology, achieves accurate characterization of the spatial distribution characteristics of drill string vibration throughout the well section, compared to traditional downhole single-point monitoring or surface single-parameter inference. Traditional methods can only monitor vibration near the drill bit, while this invention can achieve real-time monitoring of the vibration distribution throughout the entire well section, enabling early prevention at the drilling site.

Description

Technical Field

[0001] This invention relates to the field of intelligent drilling technology, specifically to a method and system for identifying drill string vibration conditions through surface and downhole data collaboration. Background Technology

[0002] As oil and gas exploration and development advances into deeper, ultra-deep, and complex geological formations, drilling operations face increasingly complex geological environments. As the core transmission component of drilling operations, the drill string's vibration problem is becoming more and more prominent. Severe drill string vibration can not only lead to premature failures such as accelerated drill string wear, loose joints, and broken drill teeth, but also cause wellbore trajectory deviations, fluctuations in drilling parameters, and in severe cases, even major accidents such as stuck drill or broken drill string, directly reducing drilling efficiency and increasing operating costs.

[0003] Currently, drill string vibration condition identification mainly relies on two types of technical solutions: one is indirect judgment based on a single surface parameter, such as inferring the downhole drill string vibration state by monitoring rotary table torque, riser pressure fluctuations, or surface vibration acceleration. However, this solution has the defect of "one-sided data representation." For example, surface parameters are affected by factors such as drilling fluid damping and drill string elastic deformation, and cannot truly reflect the actual vibration amplitude and frequency of the downhole drill string. Especially in deep well scenarios, the correlation between surface and downhole parameters is greatly weakened, which easily leads to misjudgment or omission. The second is direct analysis based on single-point downhole monitoring, which collects local drill string vibration acceleration data through measurement-while-drilling tools. Although it can obtain real downhole vibration information, it is limited by the deployment location of the tools and cannot achieve spatial distribution representation of drill string vibration throughout the well section. Moreover, single-point data is difficult to reflect the multimodal coupled vibration characteristics of the drill string.

[0004] In addition, traditional identification methods also suffer from the problem of "weak adaptability to complex environments": most existing drill string dynamics models are based on the assumption of ideal working conditions, while the load conditions and environmental parameters in actual drilling are dynamically changing, resulting in a significant decrease in the model's prediction accuracy; although some machine learning methods attempt to improve adaptability through data-driven approaches, they fail to establish a collaborative correlation mechanism between surface and downhole data and rely solely on a single data source to train the model, making it difficult to cover the vibration characteristics changes under complex working conditions and unable to meet the requirements for real-time and accurate identification.

[0005] In summary, existing technologies struggle to overcome the limitations of "single-point data" and "poor adaptability to complex environments." There is an urgent need for a drill string vibration condition identification technology that can integrate surface and downhole data, cover vibration characteristics across the entire well section, and adapt to complex working conditions to support the safe and efficient conduct of drilling operations. To this end, a drill string vibration condition identification method and system based on surface and downhole data collaboration is proposed to address the aforementioned problems. Summary of the Invention

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: Firstly, a method for identifying drill string vibration conditions through surface and downhole data collaboration, comprising the following steps: S1. Input drilling engineering parameters, including multi-node monitoring data from the surface and downhole, and analyze the vibration distribution pattern of the drill string throughout the well section; S2. Using the modal decomposition method, the drill string vibration is decoupled into independent single-order modal components, including mode shapes and modal coefficients; S3. Build a multi-input multi-output machine learning model, using surface and downhole parameters as inputs and modal coefficients as outputs, and establish data collaborative correlation. S4. Train the model using drilling engineering parameters, measured vibration data and modal parameters to learn the nonlinear mapping between input parameters and modal coefficients; S5. Based on the predicted modal coefficients and mode shapes, the vibration response of the drill string throughout the well section is inverted to identify high-incidence vibration areas and assess the risk of resonance. S6. The identified vibration information is transmitted to the client, and the staff adjusts the drilling parameters accordingly to reduce abnormal vibration and improve drilling safety.

[0007] Preferably, S1 specifically includes: The drilling engineering parameters collected during the drilling operation include key surface parameters and downhole multi-node monitoring data, including surface rotary table torque, riser pressure, top drive voltage and current, and downhole vibration acceleration and rotation speed measured while drilling. This provides a comprehensive, accurate, and time-consistent multi-source data foundation for subsequent vibration analysis, ensuring the authenticity of the modeling. Based on the collected data, an axial and torsional coupled dynamic model of the drill string throughout the well section is constructed. The drill string is discretized into a mass-spring-damping system using the lumped mass method. The axial and torsional coupled dynamic model includes an axial vibration model and a torsional vibration model. With physical mechanisms as the core, it achieves accurate numerical characterization of the complex dynamic behavior of the drill string. Based on the axial and torsional coupled dynamic model and drilling engineering parameters, the initial state of vibration distribution of the drill string at different well depths is calculated, providing input boundary conditions for subsequent modal decomposition and providing an initial state that conforms to physical reality, thus avoiding decomposition errors caused by boundary condition mismatch.

[0008] Preferably, S1 further includes: In the axial vibration model, based on the dynamics of the top drive system and the back electromotive force equation of the winch motor, the hook load and the boundary conditions at the top of the drill string are derived, providing accurate top load input for axial vibration and enhancing the model's ability to respond to the dynamics of the actual hoisting system. In the torsional vibration model, the turntable speed and torque are calculated based on the voltage-speed relationship and transmission efficiency of the top drive motor, and used as the top excitation input for torsional vibration, so that the top excitation is closer to the real power output, and the physical realism and accuracy of the torsional vibration simulation are improved. By employing a hybrid boundary condition based on stochastic processes at the bottom of the drill string, the axial-torsional coupling vibration characteristics in the drill bit-rock interaction are characterized, and the initial modeling of the vibration distribution of the drill string throughout the well section is completed. This effectively simulates the randomness and coupling of the complex rock breaking process downhole, and introduces a real nonlinear excitation source for the initial modeling.

[0009] Preferably, S2 specifically includes: Based on the axial and torsional coupled dynamic model of the entire well section drill string, the characteristic equation of the mass-spring-damping system is solved to obtain the mode shape functions of each order of the drill string and their corresponding resonance frequencies. The inherent frequency and mode shape characteristics of the system are obtained, providing a theoretical basis for subsequent modal decomposition and improving the physical accuracy of the identification basis. By adopting the modal superposition principle, the vibration response of the drill string is expressed as a linear combination of the mode shapes and the corresponding time-varying modal coefficients, thereby achieving mathematical decoupling of complex vibrations, facilitating independent analysis of each mode, and enhancing the characterization ability of multimodal coupled vibrations. By fitting numerical methods or measured vibration data, the modal coefficients of each order are obtained through decoupling, forming a set of modal parameters that describe the dynamic characteristics of drill string vibration. A set of modal features that reflects the real-time state of vibration is established, providing high-value training labels for machine learning models and improving recognition accuracy.

[0010] Preferably, S2 further includes: By utilizing downhole multi-node monitoring data and combining it with known mode shapes, the time series of modal coefficients of each order are calculated by least squares inversion, thereby achieving accurate quantification and dynamic tracking of multimodal vibration of the drill string and effectively overcoming the inherent limitation that single-point monitoring cannot reflect the dynamic response of the entire well section. The drill string vibration signal is expanded in modal space, and the vibration components contributed by different frequencies and mode shapes are separated to achieve decoupling of multimodal coupled vibration. This simplifies the mechanism analysis of complex coupled vibration, and enables clear location of the main vibration source and identification of its modal characteristics. Establish a mapping relationship library between modal coefficients and drill string vibration amplitude and frequency to provide modal label data for training and validation of machine learning models, generate high-quality training data with clear physical meaning, and lay a reliable supervised learning foundation for data-driven intelligent recognition models.

[0011] Preferably, S3 specifically includes: A multi-input-multi-output machine learning model was constructed, using drilling engineering parameters that include key surface parameters and downhole multi-node monitoring data as input feature vectors. A multi-dimensional feature space covering all working conditions was established, providing a data foundation for the model to fully perceive the dynamic state of the drill string. Using the modal coefficients of each order as the model output target, an end-to-end prediction mapping relationship from surface-downhole data to modal characteristics is constructed, opening up a direct bridge from raw data to the underlying physical mechanism, and realizing intelligent analysis of the complex and essential characteristics of vibration; By adopting a BP neural network as the basic model structure and introducing a nonlinear activation function and ADAM optimization algorithm, the model's ability to fit and generalize to complex working conditions is enhanced. This significantly improves the model's accuracy in capturing variable nonlinear relationships during drilling and its convergence stability, ensuring prediction robustness.

[0012] Preferably, S4 specifically includes: Collect historical datasets containing drilling engineering parameters, measured vibration data and corresponding modal coefficients, perform data cleaning, normalization and feature extraction processing to achieve noise suppression and feature enhancement, and improve data quality and model input consistency; The preprocessed data is input into the multi-input multi-output machine learning model, the predicted output is calculated through forward propagation, and the loss function is calculated in combination with the real modal coefficients to complete the initial mapping of the modal coefficients, quantify the prediction error, and provide a clear target for reverse optimization. The model weights are iteratively updated using backpropagation and gradient descent algorithms until the model converges, thus completing the learning of the nonlinear mapping relationship between input parameters and modal coefficients, solidifying high-precision prediction capabilities, and realizing stable real-time calculation of modal coefficients under complex working conditions.

[0013] Preferably, S5 specifically includes: The drilling engineering parameters collected in real time are input into a pre-trained multi-input multi-output machine learning model to predict the modal coefficients of each order, thereby achieving millisecond-level accurate mapping from multi-source data from the surface to the core dynamic characteristics of the drill string. By combining known mode shapes, the vibration response of the entire well section is reconstructed through modal superposition, the spatial distribution of vibration amplitude at different well depths is calculated, and the distribution of vibration amplitude along the well depth is visualized. The dominant frequency components in the modal coefficients are extracted and compared with the natural frequency of the drill string to identify high-vibration areas and assess resonance risk. These high-vibration areas are the high-vibration well sections, enabling early warning of resonance risk and accurate location of well sections with concentrated vibration energy, providing a targeted basis for proactive control.

[0014] Preferably, S6 specifically includes: The vibration identification results, including vibration amplitude distribution, high-incidence areas of vibration, and resonance risk information, are transmitted to the client calculator and monitoring interface in real time, enabling the visualization of vibration information and helping monitoring personnel to quickly grasp the vibration status of the entire well section. Based on the identification results, the staff judged the vibration status of the drill string and adjusted drilling parameters such as rotation speed, drilling pressure or pump displacement based on the vibration amplitude and frequency information, making the parameter adjustment more targeted, significantly improving the vibration suppression efficiency and reducing the operational risk. By adjusting parameters, active suppression of drill string vibration is achieved, forming a closed-loop control of perception, identification, and regulation, which improves the safety and efficiency of the drilling process, effectively avoids downhole failures caused by vibration, extends drill string life, and improves overall drilling efficiency.

[0015] Secondly, a drill string vibration condition identification system based on surface-to-downhole data collaboration is provided to implement the aforementioned drill string vibration condition identification method based on surface-to-downhole data collaboration. The system includes a remote engineering monitoring center, which is communicatively connected to the following modules: The multi-source data acquisition and synchronization module is used to acquire drilling engineering parameters, including key surface parameters and multi-node monitoring data from downhole, in real time through a deployed ground and downhole sensor network, and to achieve synchronous data transmission at a sampling rate of not less than 1000 Hz. The dynamic modeling and boundary reconstruction module establishes an axial and torsional coupled dynamic model of the entire well section drill string based on the lumped mass method, and accurately calculates the top and bottom boundary conditions through motor backpropagation and stochastic process theory. The modal decomposition and coefficient inversion module is used to call finite element software to perform system modal analysis, obtain natural frequencies and mode shapes, and use downhole multi-node monitoring data to invert the time-varying modal coefficients of each order through the least squares method, decoupling complex vibrations into independent modes and realizing the mapping from physical space to modal space. The collaborative machine learning prediction module is used to build a multi-input multi-output machine learning model. It takes drilling engineering parameters as input and modal coefficients as output, learns the nonlinear mapping between input parameters and modal coefficients, and inverts the vibration response of the drill string throughout the well section to achieve real-time and high-precision prediction of the intrinsic characteristics of vibration under complex working conditions. The intelligent diagnosis and closed-loop control module reconstructs the vibration response of the entire well section based on the predicted modal coefficients, identifies high-incidence areas of vibration and resonance risks, and pushes control suggestions through a visual interface, forming a perception-diagnosis-control closed loop.

[0016] This invention provides a method and system for identifying drill string vibration conditions through surface and downhole data collaboration. It offers the following advantages: (i) The method and system for identifying drill string vibration conditions through ground and downhole data collaboration, through the whole-well section vibration physical model, modal decomposition method and machine learning technology, achieves accurate characterization of the spatial distribution characteristics of drill string vibration throughout the whole well section compared with traditional downhole single-point monitoring or ground single parameter inference. Traditional methods can only monitor vibration near the drill bit, while this invention can realize real-time monitoring of vibration distribution throughout the whole well section, and provide early prevention for drilling site.

[0017] (II) The drill string vibration condition identification method and system based on surface and downhole data collaboration constructs an axial-torsional coupled dynamic model of the entire well section drill string and uses modal decomposition technology to decouple the complex vibration response into a series of modal components with clear physical meaning. Based on the modal superposition principle and combined with the predicted modal coefficients, the vibration displacement, velocity and acceleration field from the wellhead to the bottom of the well is reconstructed in real time, accurately identifying well sections with vibration amplitude exceeding the limit. This breaks through the limitations of traditional single-point data and realizes accurate perception of the spatial distribution of vibration throughout the well section, providing a global perspective for vibration source location and risk area assessment.

[0018] (III) The method and system for identifying drill string vibration conditions through ground and downhole data collaboration constructs a multi-input multi-output machine learning prediction framework with key ground parameters and downhole multi-node monitoring data as inputs and drill string vibration modal coefficients as outputs. Through a data-driven approach, it learns the complex nonlinear mapping between multi-source heterogeneous parameters and intrinsic vibration modal characteristics, deeply integrating measurable engineering parameters on the ground with real local vibration information downhole, significantly improving the system's overall characterization and interpretability of real downhole vibration conditions. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the drill string vibration condition identification system in this invention, wherein 1 is the surface-to-downhole data transmission system during drilling operations; 2 is the central processing computer; 3 is the machine learning model; 4 is the drill string vibration identification result; 5 is the client calculator; and 6 is the monitoring terminal operator. Figure 2 This is a flowchart of the drill string vibration condition identification system in this invention. Figure 3 The diagram shows the axial dynamic model of the drill string constructed in this invention, where (a) represents the model of the drill string discretized into a mass-spring-damping system based on the lumped mass method; (b) represents the force analysis of the first lumped mass element; (c) represents the force analysis of the ith lumped mass element; (d) represents the force analysis of the (N-1)th lumped mass element; and (e) represents the force analysis of the Nth lumped mass element. Figure 4The diagram shows the torsional dynamics model of the drill string constructed in this invention, where (a) represents the model of the drill string discretized into a mass-spring-damping system based on the lumped mass method; (b) represents the force analysis of the first lumped mass element; (c) represents the force analysis of the ith lumped mass element; (d) represents the force analysis of the (N-1)th lumped mass element; and (e) represents the force analysis of the Nth lumped mass element. Figure 5 This is a diagram of the machine learning model framework built in this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1, please refer to Figures 1 to 5 This invention provides a technical solution: a method for identifying drill string vibration conditions through ground and downhole data collaboration, comprising the following steps: S1. Input drilling engineering parameters, including multi-node monitoring data from the surface and downhole, analyze the vibration distribution pattern of the drill string throughout the well section, and collect drilling engineering parameters during the drilling operation, including key surface parameters and multi-node monitoring data from the downhole, such as surface rotary table torque, standpipe pressure, top drive voltage and current, and vibration acceleration and rotation speed measured while drilling. This provides a comprehensive, accurate, and time-consistent multi-source data foundation for subsequent vibration analysis, ensuring the authenticity of the model. Based on the collected data, construct an axial and torsional coupled dynamic model of the drill string throughout the well section. Use the lumped mass method to discretize the drill string into a mass-spring-damping system. The axial and torsional coupled dynamic model includes an axial vibration model and a torsional vibration model. With physical mechanisms as the core, it achieves accurate numerical characterization of the complex dynamic behavior of the drill string. Based on the axial and torsional coupled dynamic model and drilling engineering parameters, calculate the initial state of vibration distribution of the drill string at different well depths, providing input boundary conditions for subsequent modal decomposition and providing an initial state that conforms to physical reality, avoiding decomposition errors caused by boundary condition mismatch. The specific work involves: During drilling operations, a sensor network deployed at multiple key nodes on the surface and downhole is used to collect drilling engineering parameters in real time, encompassing key surface parameters and multi-node monitoring data from downhole. Specifically, key surface parameters include rotary table torque (range 0-50 kN·m), riser pressure (range 0-40 MPa), and input voltage and armature current of the top drive system. Downhole multi-node monitoring data, including axial and radial vibration acceleration and drill string rotation speed, is collected via measurement-while-drilling (MWD) tools. These drilling engineering parameters are continuously transmitted to the central processing unit at a set sampling frequency, providing a complete and synchronous multi-source data foundation for subsequent dynamic modeling and state analysis. Based on the collected drilling engineering parameters, a coupled axial and torsional dynamic model of the entire drill string is established using the lumped mass method to numerically characterize the dynamic behavior of the drill string system. During the modeling process, based on the actual structural parameters of the drill string assembly, the drill string is discretized into a series of lumped mass units along the well depth direction. The mass units are connected by equivalent springs (axial stiffness is calculated based on the material's elastic modulus of 210 GPa and cross-sectional area) and damping elements. A drilling fluid viscous damping coefficient (valued at 0.1-1.0 N·s / m) is introduced to simulate the influence of the downhole fluid environment. This axial and torsional coupled dynamic model includes both axial vibration and torsional vibration models to describe the longitudinal and rotational vibrations of the drill string, respectively. The two are coupled through the bottom drill bit-rock interaction boundary. Using the constructed axial and torsional coupled dynamic model and real-time input drilling engineering parameters, the initial dynamic state of the drill string system is solved. Under given top boundary conditions and bottom random boundaries, the initial displacement, velocity, and acceleration response of the drill string at each discrete mass node are calculated using numerical integration methods. This yields a preliminary estimate of the vibration distribution throughout the well section, accurately characterizing the force and motion state of the drill string under the current operating conditions. This provides the necessary, physically accurate input boundary conditions and initial excitation for subsequent modal decomposition analysis. Furthermore, S1 further includes: In the axial vibration model, based on the dynamics of the top drive system and the back electromotive force equation of the winch motor, the hook load and the boundary conditions at the top of the drill string are derived, providing accurate top load input for axial vibration and enhancing the model's ability to respond to the dynamics of the actual hoisting system. For the construction of the axial vibration model: Based on the actual drill string assembly method, the drill string is discretized using the lumped mass method. This involves discretizing the drill string mass along its axis into a finite number of lumped mass elements, with adjacent elements connected by equivalent springs. The spring stiffness coefficients are calculated using the stiffness method, and a Rayleigh damping model is introduced to characterize the drilling mud viscosity effect in the system. This discretizes the drill string into a mass-spring-damping system model. This model comprehensively considers key factors such as the frictional resistance mechanism under wellbore constraints and the viscous damping dissipation effect of drilling mud. At the bottom of the drill string, a hybrid boundary condition based on a stochastic process is used to quantitatively characterize the coupling characteristics of axial and torsional fluctuations during the drill bit-rock interaction process. The top boundary condition is constructed based on the dynamics of the top drive system, which consists of a winch motor for drill string lifting and a top drive motor providing torsional excitation. Its dynamic characteristics directly affect the boundary response at the top of the drill string. The axial vibration model of the drill string and the force analysis of each lumped mass element are as follows: Figure 3 As shown, the forces acting on each lumped mass element conform to Newton's second law. In the top drive system, based on the motor back electromotive force equation, the winch motor voltage is: ; In the formula: This refers to the voltage of the winch motor. For armature current, For armature resistance, The back electromotive force constant is... Let be the motor speed; therefore, the expression for the winch motor speed is: ; The output torque of the winch motor is: ; In the formula: For the output torque of the winch motor, For mechanical efficiency; Taking into account the equivalent radius of the pulley and the transmission ratio, the tension of the wire rope is calculated using the mechanical transmission formula of the pulley system as follows: ; In the formula: For the tension of the wire rope, Let the equivalent radius of the pulley be . The transmission ratio; The overall efficiency expression for a multi-stage pulley system is: ; In the formula: The total efficiency of the pulley system is... For the efficiency of a single pulley, This represents the total number of pulleys; The load on the hook is calculated as follows: ; In the formula: For the large hook load, The pulley ratio; Figure 3 In (b) of the diagram, the axial force balance equation for the first concentrated mass element of the drill string can be expressed as: ; In the formula: The mass of the first lumped mass unit. Let be the acceleration of the first lumped mass element (the second derivative of the displacement of the first lumped mass element with respect to time). The spring connection stiffness coefficient between the first and second lumped mass elements. The structural damping coefficient of the drill string between the first and second lumped mass units. and These are the velocities of the first and second lumped mass elements, respectively (the displacements of the first and second lumped mass elements are the first derivatives with respect to time). This represents the drilling fluid viscosity damping coefficient experienced by the first mass block drill string segment. It is the acceleration due to gravity; Figure 3 In (c), the axial force balance equation for the i-th drill string concentrated mass element can be expressed as: ; In the formula: , Let the mass be the mass of the i-th lumped mass unit. Let be the spring connection stiffness coefficient between the i-th and (i+1)-th lumped mass elements. Let be the structural damping coefficient between the i-th and (i+1)-th lumped mass elements. The drilling fluid viscosity damping coefficient is the i-th lumped mass element. Figure 3 In (d), the axial force balance equation for the (N-1)th concentrated mass element of the drill string can be expressed as: ; Figure 3 In equation (e), the axial force balance equation for the Nth concentrated mass element of the drill string can be expressed as: ; In the formula: The mass of the drill bit node (the mass of the Nth lumped mass unit). This is the force exerted by the drilling fluid pressure at the bottom of the well on the bottom of the drill string, and its value is approximately equal to the buoyancy of the entire drill string when it is submerged in the drilling fluid. This represents the displacement of the drill bit node. The drilling fluid viscosity damping coefficient is the factor applied to the drill bit node. It is the axial force exerted on the drill bit by the rock. According to Newton's third law, it reflects the magnitude of the instantaneous drilling pressure exerted by the drill bit on the rock surface at the bottom of the well, and its value is unknown. Expanding the dynamic equations at each concentrated mass element to construct a matrix equation, the matrix form of the dynamic equations for axial vibration is expressed as: ; During the stage where the drill bit jumps off the bottom of the well, the above formula is: The matrix equations are given by the following formulas, where the coefficient matrix is ​​represented by the following formulas; and the mass matrix [M] of axial vibration is represented by: (12); In the formula: the red dashed line of the mass matrix of axial vibration is used to distinguish the mass of the drill bit node in the drill string vibration model from the mass of other nodes discretized from the drill string; The stiffness coefficient matrix [K] is represented as follows: (13); In the formula: the elements above / to the left of the dashed line correspond to the stiffness connection between the discrete segments of the drill string itself; the elements below / to the right of the dashed line correspond to the stiffness terms of the drill bit nodes. The structural damping coefficient matrix [C] is expressed as follows: (14); In the formula: the area above / to the left of the dashed line corresponds to the structural damping of each discrete segment of the drill string itself; the area below / to the right of the dashed line corresponds to the structural damping term of the drill bit node; Drilling fluid damping effect ] is represented as: (15); In the formula: the upper left side of the dashed line corresponds to the drilling fluid damping of the drill string body; the lower right side of the dashed line corresponds to the drilling fluid damping of the drill bit node. The load vector is represented as: ; Let the displacement vector be represented as: ; In the formula, T represents the transpose of the load vector and displacement vector, and the same applies below; During the contact stage between the drill bit and the bottom of the well, since the drill bit mass point satisfies the geometric relationship of equation (11) above, the dynamic equation of the drill string is expressed as follows: The matrix equation is as shown in equation (10) above, and the coefficient matrix is ​​as shown in equations (12-15) above. lines and Column; at this time, the load vector is represented as: ; At this point, the displacement vectors at each concentrated mass element of the drill string are expressed as: ; In the torsional vibration model, the turntable speed and torque are calculated based on the voltage-speed relationship and transmission efficiency of the top drive motor, and used as the top excitation input for torsional vibration, so that the top excitation is closer to the real power output, and the physical realism and accuracy of the torsional vibration simulation are improved. For the construction of the torsional vibration model: Torsional dynamics model of drill string and stress analysis of each lumped mass element, as follows: Figure 2 As shown, the forces acting on each lumped mass element conform to Newton's second law. In a top drive system, based on the motor back electromotive force equation, the top drive motor voltage is: ; In the formula: This is the voltage for the top drive motor. For armature current, For armature resistance, The back electromotive force constant is... Let be the motor speed; therefore, the expression for the top drive motor speed is: ; Taking into account both the transmission ratio and the mechanical efficiency of the top drive motor, the formula for calculating the turntable speed is: ; In the formula: The rotational speed of the turntable is numerically equivalent to the rotational speed of the first lumped mass element in the torsional vibration model. That is, the first derivative of the angle with respect to time. For the mechanical efficiency of the top drive motor. The transmission ratio; The top drive motor output torque is: ; In the formula This is the output torque of the top drive motor; The torque at the turntable is: ; In the formula: The torque at the turntable; Figure 4In (b) of the diagram, the torsional equilibrium equation for the first concentrated mass element of the drill string can be expressed as: ; In the formula: Let the moment of inertia be the first lumped mass element. Let be the angular acceleration of the first lumped mass element (the second derivative of the rotation angle of the first lumped mass element with respect to time). The spring connection stiffness coefficient between the first and second lumped mass elements. The structural damping coefficient of the drill string between the first and second lumped mass units. and These are the rotational speeds of the first and second lumped mass units, respectively (the rotation angles of the first and second lumped mass units are the first derivatives with respect to time). The drilling fluid viscosity damping coefficient experienced by the first mass block drill string section, indicated by the superscript. The coefficients representing stiffness, structural damping, and drilling fluid viscous damping in the torsional force balance equation are used to distinguish them from the coefficients in the axial force balance equation. Figure 4 In (c) of the equation, the torsional equilibrium equation for the i-th concentrated mass element of the drill string can be expressed as: ; In the formula: , Let be the moment of inertia of the i-th lumped mass element. Let be the spring connection stiffness coefficient between the i-th and (i+1)-th lumped mass elements. Let be the structural damping coefficient between the i-th and (i+1)-th lumped mass elements. The drilling fluid viscosity damping coefficient is the i-th lumped mass element. Figure 4 In (d), the torsional force balance relationship of the (N-1)th drill string concentrated mass element can be expressed as: ; Figure 4 In (e), the torsional force balance relationship of the Nth drill string concentrated mass element can be expressed as: ; In the formula: Let be the moment of inertia of the drill bit node. This represents the torque experienced at the drill bit nodal point. The angle of the drill bit node. The rotational speed of the drill bit node. Let be the angular acceleration at the drill bit node. This is the drilling fluid viscosity damping coefficient experienced at the drill bit node; Therefore, the equilibrium equations for the torsional vibration of the system can be written in matrix form as follows: ; The torque vector is represented as: ; Let the turning vector be: ; The coefficient equations of matrix equation (29) are given by equations (31-34), where the moment of inertia matrix is... Represented as: (31); In the formula: the left side above the dashed line corresponds to the moment of inertia of each discrete segment of the drill string itself; the right side below the dashed line corresponds to the moment of inertia term of the drill bit node. Torsional stiffness coefficient matrix Represented as: (32); In the formula: the area above / left of the dashed line corresponds to the torsional stiffness between discrete segments of the drill string body; the area below / right of the dashed line corresponds to the torsional stiffness term of the drill bit node. Torsional structure damping coefficient matrix Represented as: (33); In the formula: the area above / left of the dashed line corresponds to the torsional damping of each discrete segment of the drill string itself; the area below / right of the dashed line corresponds to the torsional damping term of the drill bit node. Torsional drilling fluid damping effect Represented as: (34); In the formula: the left side above the dashed line corresponds to the torsional drilling fluid damping of the drill string body; the right side below the dashed line corresponds to the torsional drilling fluid damping of the drill bit node. Hybrid boundary conditions based on stochastic processes are used at the bottom of the drill string to characterize the axial-torsional coupling vibration characteristics in the drill bit-rock interaction, complete the initial modeling of the vibration distribution of the drill string throughout the well section, effectively simulate the randomness and coupling of the complex rock breaking process downhole, and introduce a real nonlinear excitation source for the initial modeling. The specific work involves the following: In the surface system of drilling operations, the winch motor, as the core power unit of the hoisting system, directly affects the axial boundary conditions at the drill string tip. According to the principles of electrical machinery, the back electromotive force (EMF) of the winch motor is proportional to its rotational speed. By real-time monitoring of the input voltage (typical operating voltage is 600V / 690V / 3300V three-phase AC) and armature current (measurement range 0-1500A) at both ends of the armature winding, combined with the armature resistance (approximately 0.02-0.1Ω) and the back EMF constant (determined by the motor design, unit V·s / rad), the real-time motor speed is accurately calculated. Furthermore, based on the mechanical efficiency of the transmission system (0.92-0.1Ω), the real-time speed of the motor is determined. .96), the pulley block ratio (commonly 8, 10, or 12) and the effective radius of the wire rope on the drum (approximately 0.4-0.6m), are used to calculate the hook load using the pulley system mechanical transmission formula. This load is directly used as the boundary force input at the top of the drill string axial vibration model, reflecting the real-time constraint and excitation of the drill string axial motion by the ground lifting system. Regarding the determination of torsional vibration excitation, the top drive motor provides the main power for the drill string rotation. Based on the motor's steady-state voltage balance equation, the actual output speed of the motor is calculated by collecting the input line voltage and three-phase current of the top drive motor, combined with the armature circuit resistance (approximately 0.01-0.05Ω) and the inherent back electromotive force coefficient of the motor. Then, the calculation is further performed. Taking into account the gearbox or chain box transmission ratio (ranging from 3:1 to 6:1) and the overall mechanical efficiency of the transmission system (generally 0.85-0.93), the motor speed is converted into the actual rotational speed of the turntable. Furthermore, based on the motor output torque formula and transmission ratio, the input torque acting on the drill string tip is calculated. This torque value is directly used as the rotational excitation boundary condition at the top of the drill string torsional vibration model, and its fluctuation directly reflects the input energy of the ground drive system to the drill string's torsional vibration. For characterizing the boundary conditions at the bottom of the drill string, considering the strong nonlinearity and randomness of the interaction between the drill bit and the underground rock, a hybrid boundary condition model based on stochastic process theory is used to model the axial force acting on the drill bit. Axial force (drill pressure fluctuation) and torsional resistance torque (torque fluctuation) are characterized as stochastic processes with specific statistical properties. Axial force is modeled as a random sequence fluctuating around a set drill pressure value, and its fluctuation amplitude and frequency characteristics are related to lithology, drill bit type, and mechanical drilling speed. Torsional resistance torque is related to the shear strength of the rock, the friction coefficient, and the dynamic penetration depth of the drill bit's cutting teeth, and is manifested as a stochastic excitation. This hybrid boundary condition simultaneously couples axial and torsional degrees of freedom, physically reproducing the complex axial-torsional coupled vibration caused by rock heterogeneity, skipping, stick-slip, and other phenomena during the rock breaking process. It provides bottom-end dynamic constraints that conform to the actual downhole working conditions for the initial modeling of drill string vibration throughout the well section. S2. Using modal decomposition, drill string vibration is decoupled into independent single-order modal components, including mode shapes and modal coefficients. Based on the axial and torsional coupled dynamic model of the entire well section drill string, the characteristic equation of the mass-spring-damping system is solved to obtain the mode shape functions of each order of the drill string and their corresponding resonance frequencies. This provides the inherent frequency and mode shape characteristics of the system, providing a theoretical basis for subsequent modal decomposition and improving the physical accuracy of the identification basis. Using the modal superposition principle, the drill string vibration response is expressed as a linear combination of each order of mode shape and the corresponding time-varying modal coefficients, realizing the mathematical decoupling of complex vibrations. This facilitates independent analysis of each mode and enhances the characterization ability of multimodal coupled vibrations. Through numerical methods or fitting of measured vibration data, the modal coefficients of each order are obtained through decoupling, forming a set of modal parameters describing the dynamic characteristics of drill string vibration. A set of modal features reflecting the real-time state of vibration is established, providing high-value training labels for machine learning models and improving identification accuracy. The specific work involves: based on the established axial and torsional coupled dynamic model of the entire well section drill string, performing modal analysis to obtain its inherent dynamic characteristics; in actual engineering operation, solving the characteristic equations of the mass-spring-damped system using the mature numerical calculation library (ANSYS Mechanical); during modeling, the drill string material density is set to 7850 kg / m³, the elastic modulus to 210 GPa, and the Poisson's ratio to 0.3; and the cross-sectional properties of each discrete element are determined according to the specifications of drill pipe and drill collar in the API standard; in the calculation, the system damping matrix is ​​set to Rayleigh damping, and its mass proportionality coefficient and stiffness proportionality coefficient are based on measured values. After the attenuation curve is fitted and solved, the output is the first M natural frequencies (M is 10 to 20, sufficient to cover the effective frequency band excited by drilling operations) arranged in ascending order of frequency, along with their corresponding normalized mode shape vectors. After obtaining the mode shapes, the dynamic response of the drill string is characterized and reconstructed based on the modal superposition principle. In practical applications, the total vibration response is considered as the result of linear superposition of each mode shape with specific weights (i.e., i.e., variable modal coefficients). Mathematically, this is represented by a matrix operation: the displacement, velocity, or acceleration response vector in the physical coordinate system is projected onto the modal coordinate system through a modal matrix composed of the mode shape vectors, thus obtaining a set of dynamic response vectors. Time-varying modal coordinates, i.e., modal coefficients, decouple complex, coupled multi-degree-of-freedom system vibrations into a superposition of a series of independent single-degree-of-freedom modal vibrations, simplifying subsequent analysis, prediction, and control strategy design. During implementation, ensure that the extracted modal order M sufficiently covers the main vibration energy of the system under drilling excitation, using a cumulative modal effective mass participation factor exceeding 90% as the selection criterion. Combine numerical methods or downhole measured data for decoupling calculations to obtain the time-varying modal coefficients required for the superposition of driving modes. In specific implementations, if numerical methods are used, apply the same boundary conditions and excitations (top load and bottom load) as the actual operating conditions to the constructed dynamic model. Random forces are used to perform time-domain transient dynamic analysis to obtain the response time history of each physical node. Then, the time series of each modal coefficient is obtained by modal coordinate transformation. If based on downhole measured data, triaxial accelerometers deployed at key downhole locations (near the drill bit, at multiple stabilizers) provide the necessary vibration response. Using the measured acceleration signal and the known system mode shapes obtained through modal analysis, the modal coefficients of each order are calculated by least squares inverse fitting. Finally, the modal coefficients, along with the corresponding natural frequencies and mode shapes, are stored together to form a complete modal parameter dataset that can characterize the vibration dynamic characteristics of the current drill string system under specific working conditions. Furthermore, S2 further includes: utilizing downhole multi-node monitoring data, combined with known modal shapes, and calculating the time series of modal coefficients of each order through least squares inversion, to achieve accurate quantification and dynamic tracking of drill string multimodal vibration, effectively overcoming the inherent limitation that single-point monitoring cannot reflect the dynamic response of the entire well section, unfolding the drill string vibration signal in modal space, separating the vibration components contributed by different frequencies and modes, achieving decoupling of multimodal coupled vibration, simplifying the mechanism analysis of complex coupled vibration, clearly locating the main vibration source and identifying its modal characteristics, establishing a mapping relationship library between modal coefficients and drill string vibration amplitude and frequency, providing modal label data for training and verification of machine learning models, generating high-quality training data with clear physical meaning, and laying a reliable supervised learning foundation for data-driven intelligent recognition models; The specific work involves: based on obtaining the mode shapes of each order of the drill string system, using downhole measured vibration data to invert and calculate the time series of each mode coefficient, and determining its time-varying participation level. In practice, triaxial vibration accelerometers deployed at key downhole locations (near the drill bit, drill collar section, and stabilizer) acquire axial, radial, and tangential acceleration vibration signals in real time at a sampling frequency of no less than 1000Hz. During data processing, the raw signals are filtered (using a 0.5-200Hz bandpass filter to remove noise and extremely low-frequency drift) and zero-point corrected. Subsequently, the known normalized mode shapes obtained through finite element modal analysis are used... The modal response is inverted using the least squares method, represented by solving an overdetermined set of equations. By minimizing the sum of squared residuals between the measured response and the theoretical response reconstructed from the modal shapes and the coefficients to be determined, the magnitude of each modal coefficient at each sampling moment is uniquely determined, resulting in a time-varying coefficient sequence that characterizes the dynamic participation of each modality over time. After obtaining the time series of each modal coefficient, the overall vibration response of the drill string can be expanded in modal space. According to the modal superposition principle, the physical vibration response at any position and in any direction on the drill string can be expressed as a linear combination of each modal shape and its corresponding time-varying modal coefficients. Based on this, the original... Coupled multi-degree-of-freedom vibration signals are decomposed into a series of independent single-degree-of-freedom modal vibration components. Each component corresponds to a specific natural frequency and spatial mode shape, and its dynamic characteristics are completely described by the time series of the corresponding modal coefficients. By observing the amplitude, phase, and spectral characteristics of the modal coefficients of each component, the modal order of the dominant vibration energy distribution can be clearly identified, the vibration sources of different frequency components can be analyzed, and the strength of shaft-torsional coupled vibration can be evaluated. This effectively isolates the mutual interference between different mode shapes and frequency components, making it possible to analyze the mechanism of complex coupled vibrations, locate abnormal vibration sources, and quantitatively assess vibration energy. The inverted modal coefficients and their mapped physical properties are integrated. Vibration characteristics: Construct a high-quality dataset suitable for training machine learning models. For each time slice, the modal coefficients of each order are associated and labeled with the main physical characteristics of the overall vibration of the drill string at that moment. The characteristics include: the maximum vibration amplitude of the entire well section calculated based on the modal coefficient amplitude and mode shape reconstruction, the well depth range where the main vibration energy is concentrated, and the dominant vibration frequency identified by modal coefficient spectrum analysis. Finally, a structured dataset is formed with multi-source working condition parameters from the surface and downhole as input features and modal coefficient sequences and their derived physical vibration characteristics as output labels. The data consistency principle is followed to ensure that the input and output are synchronized in time. S3. Build a multi-input multi-output machine learning model, using surface and downhole parameters as inputs and modal coefficients as outputs, and establish data collaborative correlation. S4. Train the model using drilling engineering parameters, measured vibration data and modal parameters to learn the nonlinear mapping between input parameters and modal coefficients; S5. Based on the predicted modal coefficients and mode shapes, the vibration response of the drill string throughout the well section is inverted to identify high-incidence vibration areas and assess the risk of resonance. S6. The identified vibration information is transmitted to the client, and the staff adjusts the drilling parameters accordingly to reduce abnormal vibration and improve drilling safety.

[0022] Example 2, as Figures 1 to 5 As shown, based on Embodiment 1, the present invention provides a technical solution: S3 specifically includes: constructing a multi-input-multi-output machine learning model, using drilling engineering parameters including key surface parameters and downhole multi-node monitoring data as input feature vectors, establishing a multi-dimensional feature space covering all working conditions, providing a data foundation for the model to fully perceive the dynamic state of the drill string, using modal coefficients of each order as the model output target, constructing an end-to-end prediction mapping relationship from surface-downhole data to modal features, opening a direct bridge from raw data to the intrinsic physical mechanism, realizing intelligent analysis of the complex essential characteristics of vibration, using a BP neural network as the basic structure of the model, introducing nonlinear activation functions and ADAM optimization algorithms, enhancing the model's fitting and generalization ability for complex working conditions, significantly improving the model's accuracy in capturing variable nonlinear relationships during drilling and its convergence stability, and ensuring prediction robustness; The specific work involves: when constructing a multi-input multi-output machine learning model, the input feature vector is defined based on the engineering parameters that can be collected at the drilling site. Key surface parameters include: rotary table torque, riser pressure, and top drive system input voltage, while simultaneously monitoring armature current; downhole multi-node monitoring data comes from the measurement-while-drilling tool, acquiring axial, radial, and tangential vibration accelerations at a sampling frequency of no less than 1000 Hz, while simultaneously monitoring drill string rotation speed; the output target is the modal coefficients of each order obtained through modal analysis decoupling, with the order M selected based on the principle that the cumulative modal effective mass participation coefficient exceeds 90%, ranging from 10 to 20 orders. Both input and output data are processed in 1-second basic time slices. Row-level synchronization and alignment are used to construct one-to-one corresponding sample pairs, ensuring that the data-driven mapping relationship has clear physical timeliness. The model adopts a feedforward BP neural network as the core architecture to construct an end-to-end prediction mapping from multi-source data to modal features. The number of neurons in the input layer is consistent with the dimension of the input features and is dynamically adjusted according to the number of selected surface and downhole parameters. When integrating 9-axis vibration and rotational speed data from 5 types of surface parameters and 3 downhole nodes, the input dimension reaches 32 dimensions. The hidden layer is designed with 3 to 5 layers, with 64 to 256 neurons in each layer. Fully connected layers are used, and the ReLU function is uniformly used for the activation function of the hidden layers to introduce nonlinear transformation capability and alleviate gradient vanishing. The problem is that the number of neurons in the output layer is equal to the order M of the predicted modal coefficients. A linear activation function is used. The model is trained based on a constructed time-series dataset, using mean squared error as the loss function. The ADAM algorithm is used for optimization, with an initial learning rate of 0.001, a first-moment estimation decay factor β1 of 0.9, a second-moment estimation decay factor β2 of 0.999, and a gradient clipping threshold of 1.0 to ensure training stability. The training batch size is set to 128, and the number of training epochs is dynamically determined based on the convergence of the validation set loss. The trained multi-input multi-output machine learning model is embedded into a drilling real-time monitoring system to form an online prediction process, automatically collecting the latest data in 1-second intervals. New key surface parameters and downhole multi-node monitoring data are preprocessed according to the same specifications as the training phase, including data validity verification, dimension normalization, and bandpass filtering from 0.5 to 200 Hz. The processed feature vectors are input in real time to the deployed multi-input multi-output machine learning model. The model performs forward propagation calculations and outputs the predicted values ​​of each modal coefficient in the next time slice in parallel within milliseconds. The predicted modal coefficient sequence is then sent to the subsequent vibration reconstruction and diagnosis module for real-time inversion of the spatial distribution of vibration amplitude across the entire well section, identification of high-vibration well sections, and assessment of resonance risk, forming a closed loop of data acquisition, feature processing, model prediction, and state identification. S4 specifically includes: collecting historical datasets containing drilling engineering parameters, measured vibration data, and corresponding modal coefficients; performing data cleaning, normalization, and feature extraction to achieve noise suppression and feature enhancement, improving data quality and model input consistency; inputting the preprocessed data into a multi-input multi-output machine learning model; calculating the predicted output through forward propagation; calculating the loss function in conjunction with the real modal coefficients to complete the initial mapping of modal coefficients; quantifying the prediction error; providing a clear objective for backpropagation optimization; iteratively updating the model weights using backpropagation and gradient descent algorithms until the model converges; completing the learning of the nonlinear mapping relationship between input parameters and modal coefficients; solidifying high-precision prediction capabilities; and achieving stable real-time estimation of modal coefficients under complex working conditions. The specific work involves: before constructing and training a multi-input multi-output machine learning model, preparing and preprocessing historical datasets. The collected data covers the continuous drilling operation cycle and includes surface engineering parameters (rotary table torque, riser pressure, top drive system input voltage, armature current) as well as triaxial vibration acceleration signals and rotational speeds acquired by downhole measurement-while-drilling tools at a sampling rate of no less than 1000Hz. The corresponding modal coefficients are obtained through inversion from previous modal analysis and measured data. The order M is selected based on the principle that the cumulative modal effective mass participation factor is greater than 90%, and is generally 10. -20th order, data cleaning stage: Missing segments and obvious outliers caused by sensor malfunction or transmission interruption are removed; a bandpass filter of 0.5-200Hz is applied to the vibration acceleration signal to eliminate high-frequency noise and extremely low-frequency drift; subsequently, all parameters are time-aligned, and synchronous samples are extracted in 1-second time slices. During feature extraction, while retaining the original time series points, short-time statistical features can be optionally calculated. Finally, each dimension of features is Z-score normalized to a mean of 0 and a standard deviation of 1, forming a standardized dataset that meets the model input requirements; preprocessing... The sample data was then divided into training, validation, and test sets in a 7:2:1 ratio and input into a multi-input multi-output (MIMO) machine learning model for training. This model employs a feedforward backpropagation (BP) neural network architecture. The number of neurons in the input layer is consistent with the feature dimension. There are three hidden layers with 128, 64, and 32 neurons respectively, using ReLU as the activation function. The number of neurons in the output layer equals the modality coefficient order M, and a linear activation function is used. During training, data is input into the network in batches of 128, and the outputs of each layer are calculated through forward propagation until the modality coefficients are obtained. The model first predicts the modal coefficients; then, it uses the mean squared error between the predicted values ​​and the true modal coefficients as the loss function to measure the current prediction accuracy of the model. The entire training process is carried out in a GPU-accelerated environment, and the forward propagation time of each batch is controlled in milliseconds to ensure training efficiency. The model updates the weights iteratively through the backpropagation algorithm and gradient descent strategy, specifically using the ADAM optimizer, with an initial learning rate set to 0.001, a first-order moment estimation decay factor β1 of 0.9, a second-order moment estimation decay factor β2 of 0.999, and a gradient clipping threshold of 1.To maintain training stability, the gradient of the loss function with respect to each weight is calculated during backpropagation in each batch. The ADAM algorithm then adaptively adjusts the learning rate of each parameter based on the first and second moments of the gradient, and updates the weight matrix and bias terms in the reverse direction of the gradient. During training, the model performance is evaluated on the validation set after each epoch. If the validation set loss does not decrease for 10 consecutive epochs, an early stopping mechanism is triggered, and the model parameters with the minimum validation loss are saved. Finally, when both the training loss and validation loss converge to a stable interval and the mean absolute error on the test set does not exceed 5% of the modal coefficient range, the model training is considered complete. At this point, the nonlinear mapping relationship between the input parameters and modal coefficients has been learned and solidified, and the model can be used for real-time prediction. S5 specifically includes: inputting real-time collected drilling engineering parameters into a pre-trained multi-input multi-output machine learning model to predict modal coefficients of each order, achieving millisecond-level accurate mapping from multi-source data from the surface to the core dynamic characteristics of the drill string, combining known mode shapes of each order, reconstructing the vibration response of the entire well section through modal superposition, calculating the spatial distribution of vibration amplitude at different well depths, visualizing the distribution of vibration amplitude along the well depth, extracting the dominant frequency components in the modal coefficients, comparing and analyzing them with the natural frequency of the drill string, identifying high-incidence vibration areas and assessing resonance risk. Among these, high-incidence vibration areas are high-incidence vibration well sections, enabling early warning of resonance risk and accurate location of well sections with concentrated vibration energy, providing targeted basis for proactive control. The specific work involves the following: During drilling operations, the surface data acquisition system acquires key parameters such as rotary table torque, riser pressure, top drive armature current, and input voltage in real time at a sampling rate of no less than 100Hz. Simultaneously, downhole measurement-while-drilling tools acquire triaxial vibration acceleration and rotational speed signals near the drill bit, drill collar, and stabilizer. All data undergoes bandpass filtering and normalization within the 0.5-250Hz range to form a 35-40 dimension feature vector. This feature vector is input to a pre-deployed, fully trained multi-input multi-output machine learning model. The model performs forward computation under GPU acceleration, outputting real-time predicted values ​​of each modal coefficient within 5ms, with the prediction cycle synchronized with the data acquisition cycle by 1 second. Using the predicted modal coefficients and a pre-obtained normalized modal shape database (containing 12 natural modes within the 0-100Hz frequency band, with the mode shape vector dimension corresponding to 200 discrete nodes at well depth), a modal superposition principle is applied. Vibration response reconstruction is performed, in which, Represents the spatial position of the drill string ,time Vibration response at the location; For the first The specific form of the first-order mode shape is determined by the core coefficient matrix and boundary conditions in the drill string vibration equation. Is with the first The time-varying modal coefficients corresponding to the first-order mode shape function directly reflect the evolution of the contribution of that mode to the overall vibration response of the drill string over time. When the value of a certain time-varying coefficient is large at a certain moment, it means that that mode dominates the vibration response at that moment. This indicates the maximum order of the modal function selected to fully characterize the original vibration response of the drill string under specific accuracy requirements; it automatically calculates the vibration displacement, velocity, and acceleration responses of 200 discrete nodes throughout the entire well section from the wellhead to the bottom, and generates vibration amplitude envelope diagrams at 0.5-meter intervals, thereby identifying well sections where the vibration amplitude exceeds the safety threshold (axial vibration > 3mm, lateral vibration > 2mm) in real time, and highlighting them in the three-dimensional wellbore trajectory model; it performs a Fast Fourier Transform (FFT analysis window length 10 seconds, overlap rate 50%) on the predicted modal coefficient time history to extract the dominant frequency of each mode. The frequency components are analyzed by comparing the real-time frequency with a pre-defined database of natural frequencies of the drill string system (the first 12 natural frequencies range: axial 0.5-35Hz, torsional 0.8-28Hz, lateral 1.2-42Hz). The database of natural frequencies of the drill string system is determined before drilling operations through system modal analysis using a finite element model of the drill string system. If the excitation frequency of any mode deviates from its natural frequency by less than ±5% and the amplitude of the modal coefficient exceeds the threshold of 0.8 for 3 consecutive seconds, a resonance risk is identified and an alarm is triggered. Simultaneously, the energy contribution rate of each mode is calculated. , Indicates the first The energy contribution rate of each mode in the total vibration response; this value is used to identify the dominant vibration mode. Indicates the first First-order mode coefficient vector The square of the 2-norm (i.e., the square of the Euclidean norm). This represents the summation of the vibration energy (i.e., the square of the norm of its modal coefficients) of all M-order modes, and the result represents the total vibration energy of the drill string system during that time period; the dominant vibration modes with an energy share >15% are identified, and their mode shape characteristics are combined to determine the high-vibration well sections (located in the drill collar section and the wellbore dogleg degree >5° / 30m), providing targeted guidance for parameter optimization; S6 specifically includes: transmitting vibration identification results, including vibration amplitude distribution, high-incidence vibration areas, and resonance risk information, to the client calculator and monitoring interface in real time, realizing the visualization of vibration information, assisting monitoring personnel to quickly grasp the vibration status of the entire well section, and allowing staff to judge the drill string vibration status based on the identification results, and adjust drilling parameters such as rotational speed, drilling pressure, or pump displacement based on vibration amplitude and frequency information, making parameter adjustments more targeted, significantly improving vibration suppression efficiency, reducing operational risks, achieving active suppression of drill string vibration through parameter adjustment, forming a closed-loop control of perception-identification-control, improving the safety and efficiency of the drilling process, effectively avoiding downhole failures caused by vibration, extending drill string life, and improving overall drilling efficiency; The specific work involves the following: At the drilling site, the central processing computer deployed in the industrial control room pushes the real-time vibration identification results to the operator console client computer and the monitoring interface of the remote engineering monitoring center via industrial Ethernet at a refresh rate of no less than 1Hz. The transmitted data packet structure is optimized, including timestamps, well depth coordinate indexes, and corresponding vibration state parameters. The vibration amplitude distribution data is displayed at a resolution of 0.5 meters, covering the entire well section from the wellhead to the current well bottom. This data is then simultaneously rendered on the 3D wellbore trajectory model using a color gradient curve (blue-yellow-red representing low-medium-high vibration intensity). High-incidence vibration areas are defined as areas continuously exceeding the safety threshold (axial displacement). For well sections with a displacement > 3.0 mm and a lateral displacement > 2.0 mm) and a length greater than 10 meters, a highlighting flashing border box will be displayed on the interface. Resonance risk information will be provided through a separate alarm panel with text and audio-visual prompts, clearly listing the risk modal order, real-time excitation frequency, the nearest natural frequency value, and the deviation rate, ensuring that monitoring personnel can immediately and unambiguously grasp the core risk points of downhole vibration. After receiving real-time identification information, the shift engineer and driller will make a comprehensive judgment and decision based on drilling operation procedures and the experience database of vibration control experts, prioritizing high-risk items: If the interface triggers a resonance alarm (the deviation between the excitation frequency and the natural frequency is continuously < ±5% and the corresponding modal coefficient amplitude is > 0).If the vibration frequency exceeds 3 seconds, adjust the rotary table speed in increments of 5-10 revolutions per minute to move the excitation frequency away from the natural frequency band and break the resonance condition. For identified high-vibration areas, make targeted adjustments based on the vibration direction: if lateral vibration is dominant, appropriately reduce the drilling pressure within the safety window (adjustment range is generally 10-20 kN) to reduce lateral force; if axial vibration is dominant, fine-tune the top drive speed or pump displacement (adjustment amount is 3%-5% of the rated value) to change downhole pressure fluctuations and drill string dynamics. All parameter adjustments must be carried out within the safe operating range specified in the drilling engineering design, and follow the principle of small steps and frequent observation to avoid new complex downhole conditions induced by parameter mutations. After each parameter adjustment, use the surface and downhole sensor network for 1-2 data acquisition cycles. Feedback is received within approximately 1-2 seconds. The adjusted drilling parameters and the corresponding changes in downhole vibration response (newly predicted modal coefficients and reconstructed vibration amplitudes) are then input back into the system for calculation and evaluation. This forms a closed-loop control flow: real-time data acquisition → multi-source feature extraction → machine learning model prediction → vibration state identification and risk diagnosis → manual or automatic parameter intervention → downhole state feedback. Through continuous iteration of the closed-loop control flow, drilling parameters are guided to dynamically approach the sweet spot region of minimum vibration, achieving proactive suppression of abnormal and severe drill string vibrations. This transforms traditional passive post-event processing into proactive prevention and control based on real-time perception and precise diagnosis, effectively reducing drill string fatigue damage, minimizing non-productive time, and increasing mechanical drilling speed. Mechanistically, this enhances the overall safety and technical and economic efficiency of deep and complex well drilling operations.

[0023] Example 3, as Figures 1 to 5 As shown, based on Embodiments 1-2, the present invention also provides a drill string vibration condition identification system based on surface-to-well data collaboration, used to implement the above-mentioned method for drill string vibration condition identification based on surface-to-well data collaboration, including a remote engineering monitoring center, which has the following communication connections: The multi-source data acquisition and synchronization module is used to acquire drilling engineering parameters, including key surface parameters and multi-node monitoring data from downhole, in real time through the deployed surface and downhole sensor network. It achieves synchronous data transmission at a sampling rate of no less than 1000 Hz, establishing a high-sampling, multi-node, cross-space data synchronization acquisition system, providing a unified and complete input basis for subsequent collaborative analysis. The dynamic modeling and boundary reconstruction module establishes an axial and torsional coupled dynamic model of the entire drill string based on the lumped mass method. It accurately calculates the top and bottom boundary conditions through motor backpropagation and stochastic process theory, realizing physical modeling and dynamic boundary reconstruction of the drill string system under real working conditions, and providing a high-fidelity physical simulation environment for modal analysis. The modal decomposition and coefficient inversion module is used to call finite element software to perform system modal analysis, obtain natural frequencies and mode shapes, and use downhole multi-node monitoring data to invert time-varying modal coefficients of each order through the least squares method. This decouples complex vibrations into independent modes, realizes the mapping from physical space to modal space, and provides interpretable and predictable structured features for machine learning. The collaborative machine learning prediction module is used to build a multi-input multi-output machine learning model. It takes drilling engineering parameters as input and modal coefficients as output, learns the nonlinear mapping between input parameters and modal coefficients, and inverts the vibration response of the drill string throughout the well section to achieve real-time and high-precision prediction of the intrinsic characteristics of vibration under complex working conditions. The intelligent diagnosis and closed-loop control module reconstructs the vibration response of the entire well section based on the predicted modal coefficients, identifies high-incidence areas of vibration and resonance risks, and pushes control suggestions through a visual interface, forming a perception-diagnosis-control closed loop to achieve real-time diagnosis and active suppression of drill string vibration, and promotes the transformation of drilling operations from passive response to intelligent prevention.

[0024] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0025] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for identifying the working condition of drill string vibration in surface downhole data collaboration, characterized in that, Includes the following steps: S1. Input drilling engineering parameters, including multi-node monitoring data from the surface and downhole, and analyze the vibration distribution pattern of the drill string throughout the well section; S2. Using the modal decomposition method, the drill string vibration is decoupled into independent single-order modal components, including mode shapes and modal coefficients; S3. Build a multi-input multi-output machine learning model, using surface and downhole parameters as inputs and modal coefficients as outputs, and establish data collaborative correlation. S4. Train the model using drilling engineering parameters, measured vibration data and modal parameters to learn the nonlinear mapping between input parameters and modal coefficients; S5. Based on the predicted modal coefficients and mode shapes, the vibration response of the drill string throughout the well section is inverted to identify high-incidence vibration areas and assess the risk of resonance. S6. The identified vibration information is transmitted to the client, and the staff adjusts the drilling parameters accordingly to reduce abnormal vibration.

2. The method of claim 1, wherein: S1 specifically includes: Drilling engineering parameters that include key surface parameters and downhole multi-node monitoring data collected during drilling operations include surface rotary table torque, riser pressure, top drive voltage and current, and downhole vibration acceleration and rotational speed measured during drilling. Based on the collected data, an axial and torsional coupled dynamic model of the entire well section drill string is constructed. The drill string is discretized into a mass-spring-damping system using the lumped mass method. The axial and torsional coupled dynamic model includes an axial vibration model and a torsional vibration model. Based on the axial and torsional coupled dynamic model and drilling engineering parameters, the initial state of vibration distribution of the drill string at different well depths is calculated.

3. The method for identifying drill string vibration conditions through surface and downhole data collaboration according to claim 2, characterized in that: S1 further includes: In the axial vibration model, based on the dynamics of the top drive system and the back electromotive force equation of the winch motor, the hook load and the boundary conditions at the top of the drill string are derived. In the torsional vibration model, the turntable speed and torque are calculated based on the voltage-speed relationship and transmission efficiency of the top drive motor, and used as the excitation input for the top of the torsional vibration. A hybrid boundary condition based on a stochastic process is used at the bottom of the drill string to characterize the axial-torsional coupling vibration characteristics in the drill bit-rock interaction, thus completing the initial modeling of the vibration distribution of the drill string throughout the well.

4. The method for identifying drill string vibration conditions through surface and downhole data collaboration according to claim 1, characterized in that: S2 specifically includes: Based on the axial and torsional coupled dynamic model of the drill string throughout the well section, the characteristic equations of the mass-spring-damping system are solved to obtain the mode shape functions of each order of the drill string and their corresponding resonance frequencies. Using the modal superposition principle, the drill string vibration response is expressed as a linear combination of mode shapes and corresponding time-varying modal coefficients. By fitting numerical methods or measured vibration data, the modal coefficients of each order are obtained through decoupling, forming a set of modal parameters that describe the dynamic characteristics of drill string vibration.

5. The method for identifying drill string vibration conditions through surface and downhole data collaboration according to claim 4, characterized in that: S2 further includes: Using downhole multi-node monitoring data and combined with known mode shapes, the time series of modal coefficients of each order are calculated by inversion using the least squares method; The drill string vibration signal is expanded in modal space, and the vibration components contributed by different frequencies and mode shapes are separated to achieve decoupling of multimodal coupled vibration. Establish a mapping relationship library between modal coefficients and drill string vibration amplitude and frequency to provide modal label data for training and validation of machine learning models.

6. The method for identifying drill string vibration conditions through surface and downhole data collaboration according to claim 1, characterized in that: S3 specifically includes: Construct a multi-input multi-output machine learning model, using drilling engineering parameters that include key surface parameters and downhole multi-node monitoring data as input feature vectors; Using the modal coefficients of each order as the model output target, an end-to-end prediction mapping relationship from surface-downhole data to modal features is constructed; A BP neural network is used as the basic structure of the model, and a nonlinear activation function and ADAM optimization algorithm are introduced to enhance the model's fitting and generalization ability to complex working conditions.

7. The method for identifying drill string vibration conditions through surface and downhole data collaboration according to claim 1, characterized in that: S4 specifically includes: Collect historical datasets containing drilling engineering parameters, measured vibration data, and corresponding modal coefficients, and perform data cleaning, normalization, and feature extraction processing. The preprocessed data is input into the multi-input multi-output machine learning model, the predicted output is calculated through forward propagation, and the loss function is calculated by combining the real modality coefficients. The model weights are iteratively updated using backpropagation and gradient descent algorithms until the model converges, thus completing the learning of the nonlinear mapping relationship between the input parameters and the modal coefficients.

8. The method for identifying drill string vibration conditions through surface and downhole data collaboration according to claim 1, characterized in that: S5 specifically includes: The drilling engineering parameters collected in real time are input into a pre-trained multi-input multi-output machine learning model to predict the modal coefficients of each order. By combining the known mode shapes, the vibration response of the entire well section is reconstructed through modal superposition, and the spatial distribution of vibration amplitude at different well depths is calculated. The dominant frequency components in the modal coefficients are extracted and compared with the natural frequency of the drill string to identify high-vibration areas and assess resonance risk. The high-vibration areas are the high-vibration well sections.

9. The method for identifying drill string vibration conditions through surface and downhole data collaboration according to claim 1, characterized in that: S6 specifically includes: The vibration identification results, including vibration amplitude distribution, high-incidence areas of vibration, and resonance risk information, are transmitted to the client calculator and monitoring interface in real time. Based on the identification results, the staff judged the vibration status of the drill string and adjusted drilling parameters such as rotation speed, drilling pressure or pump discharge rate based on the vibration amplitude and frequency information. Active suppression of drill string vibration is achieved through parameter adjustment, forming a closed-loop control system of perception, identification, and regulation.

10. A drill string vibration condition identification system based on surface-to-downhole data collaboration, used to implement the drill string vibration condition identification method based on surface-to-downhole data collaboration as described in any one of claims 1-9, comprising a remote engineering monitoring center, characterized in that, The remote engineering monitoring center has the following communication connection modules: The multi-source data acquisition and synchronization module is used to acquire drilling engineering parameters, including key surface parameters and multi-node monitoring data from downhole, in real time through a deployed ground and downhole sensor network. The dynamic modeling and boundary reconstruction module establishes an axial and torsional coupled dynamic model of the entire well section drill string based on the lumped mass method; The modal decomposition and coefficient inversion module is used to call finite element software to perform system modal analysis, obtain natural frequencies and mode shapes, and invert time-varying modal coefficients of each order using downhole multi-node monitoring data through the least squares method. The collaborative machine learning prediction module is used to build a multi-input multi-output machine learning model. It takes drilling engineering parameters as input and modal coefficients as output, learns the nonlinear mapping between input parameters and modal coefficients, and inverts the vibration response of the drill string throughout the well section. The intelligent diagnosis and closed-loop control module reconstructs the vibration response of the entire well section based on the predicted modal coefficients, identifies high-incidence areas of vibration and resonance risks, and pushes control suggestions through a visual interface.

Images