Pipe laying ship dynamic positioning and stinger cooperative control method

Abstract

The invention discloses a pipe laying ship dynamic positioning and stinger cooperative control method, belongs to the technical field of ocean engineering ship automatic control, and aims to solve the problems that an existing separation control strategy easily causes control conflicts and influences operation safety and efficiency due to lack of a unified model. The method comprises the following steps: establishing a unified nonlinear dynamic model of coupling motion of a pipe laying ship, a stinger, a pipeline and a marine environment, and based on a multi-sensor data fusion and layered predictive control architecture, solving a multi-objective optimization problem in an online rolling manner; and generating an optimal thrust instruction set and an optimal attitude control instruction set which meet physical constraints of all execution mechanisms and track the reference sequence, issuing the optimal thrust instruction set to each propeller of the dynamic positioning system, and issuing the optimal attitude control instruction set to each hydraulic execution mechanism of the stinger adjusting system. High-precision and high-robustness cooperative control of the pose of the pipe laying ship and the form of the stinger is achieved, and the overall safety and operation efficiency of deep sea pipeline laying operation are improved.

Description

Technical Field

[0001] This invention belongs to the field of automatic control technology for marine engineering vessels, specifically relating to a method for coordinated control of dynamic positioning and support frame on a pipelaying vessel. Background Technology

[0002] In the development of offshore oil and gas resources, the control performance of the pipelaying vessel's operating system directly affects the efficiency and safety of pipeline laying. The pipelaying vessel's operating system is a complex multi-body coupled dynamic system, comprising the vessel itself, a flexible support structure, and the pipeline being laid, while simultaneously being subjected to continuous environmental loads such as wind, waves, and currents. Current technologies typically design and control the pipelaying vessel's dynamic positioning system and the support structure's attitude control system as two relatively independent subsystems.

[0003] This separate control strategy has certain limitations. Dynamic positioning systems typically aim to maintain the ship's position and heading, while the hull support control system primarily adjusts based on preset shapes or angles. Due to the lack of a unified description of the coupling effects between the hull, hull support, and piping, the control decisions made by each subsystem based on local information may not be optimal from a global perspective, and may even lead to conflicting control actions within the system. For example, adjustments to the hull support may significantly alter the force distribution on the hull, thus affecting its positioning capability, while the ship's motion can in turn affect the hull support and piping, inducing additional dynamic stresses.

[0004] The reason for this limitation lies in the significant difficulty in establishing a unified nonlinear dynamic model that accurately describes the aforementioned multibody coupled motions. The support frame is typically considered a multi-segmented, articulated, flexible structure, and its connection to the hull involves complex force and moment interactions. The pipes extend from the support frame rollers to the seabed, and their shape and stress distribution are influenced by the combined effects of ship motion, support frame morphology, and seabed contact conditions. Furthermore, accurately calculating the hydrodynamic loads on the hull and support frame in waves is also highly complex. Due to the lack of such a unified model, the controller struggles to predict the comprehensive impact of current control actions on the future state of the entire system.

[0005] Furthermore, other challenges arise in achieving coordinated control. System state awareness relies on various sensors, such as ship position reference systems, attitude sensors, angle sensors, and tension and stress sensors. These sensor data differ in accuracy, sampling rate, and reliability, making effective fusion a challenge to provide a consistent and reliable estimate of the system state. On the other hand, the system's actuators, including propellers and hydraulic actuators, are subject to significant physical constraints, such as thrust saturation, rate limits, and torque limitations. These constraints must be fully considered when generating control commands; otherwise, actuator saturation, performance degradation, or even system instability may occur. Existing control methods often suffer from insufficient computational efficiency when dealing with such multi-objective, multi-constraint optimization problems, making it difficult to meet the real-time requirements of offshore operations.

[0006] Therefore, existing technologies are to some extent unable to achieve high-precision and robust coordinated control of the pipelaying vessel's position and the form of the pipe-laying rack, which may affect operational safety and efficiency under complex marine environmental loads. Summary of the Invention

[0007] This invention provides a method for the coordinated control of dynamic positioning of a pipelaying vessel and the support frame. Based on a unified nonlinear dynamic model of the coupled motion of the pipelaying vessel, the support frame, the pipeline, and the marine environment, it achieves real-time coordinated optimization control of the pipelaying vessel's posture and the support frame's shape through hierarchical predictive control and multi-sensor information fusion. Furthermore, it can autonomously trigger an emergency avoidance mechanism when the system malfunctions or exceeds the limits, thereby improving the overall safety, accuracy, and robustness of deep-sea pipeline laying operations.

[0008] To achieve these objectives and other advantages of the present invention, a method for coordinated control of dynamic positioning and pipelaying support for a pipelaying vessel is provided, comprising: S1. Establish a unified nonlinear dynamic model for the coupled motion of the pipelaying vessel, the pipe-laying support, the pipeline, and the marine environment. Specifically, the pipe-laying support is considered as a multi-segment flexible beam structure connected to the hull via a universal hinge mechanism, and the interaction torque and force between the support and the hull are calculated. The pipeline is represented as a beam element model based on the absolute nodal coordinate method, and its shape and stress distribution from the pipe-laying support roller to the seabed contact point are calculated in real time. The hydrodynamic forces on the hull and the pipe-laying support are introduced into the unified nonlinear dynamic model based on the three-dimensional potential flow theory, and environmental loads based on the Morrison formula are also introduced. S2. The central controller deployed on the pipelaying vessel receives and integrates monitoring data in real time from the ship's position reference system, IMU attitude sensor, tilt and torque sensors at each hinge point of the pipe support frame, pipe tensioner tension sensor, and fiber optic stress sensors located at the end of the pipe support frame and key nodes of the pipe. S3. The central controller adopts a hierarchical predictive control architecture, which includes an upper layer based on an optimization model and a lower layer responsible for tracking control. The upper layer, based on a unified nonlinear dynamics model and the fusion sensing system state obtained in step S2, performs forward simulation calculations on the evolution of the system state in the future finite time domain; it uses a sequential quadratic programming algorithm to solve the multi-objective optimization problem online in a rolling manner, and outputs the reference trajectory sequence of the pipelaying vessel and the reference shape sequence of the support frame; the lower layer integrates a state observer to estimate the state variables in the system that cannot be directly measured; the lower layer receives the output sequence of the upper layer and, combined with the solution results of the state observer, generates in real time the optimal thrust command set and the optimal attitude control command set that satisfy the physical constraints of all actuators and track the reference sequence. S4. The optimal thrust command set is sent to each thruster of the dynamic positioning system, and the optimal attitude control command set is sent to each hydraulic actuator of the pipelaying support adjustment system to achieve coordinated control of the pipelaying vessel's position and attitude and the support structure.

[0009] Preferably, in step S3, the central controller adopts a hierarchical model predictive control architecture, which includes an upper-level optimizer and a lower-level tracker. The upper-level optimizer operates in a first control cycle, performs forward simulation calculations based on the unified nonlinear dynamics model, solves a multi-objective optimization problem, and outputs a reference trajectory sequence for the pipelaying vessel and a reference shape sequence for the support frame. The lower-level tracker operates in a second control cycle, which is shorter than the first control cycle, receives the output reference sequence from the upper-level optimizer, and uses a robust model predictive control algorithm to handle model mismatch and dynamic effects not described in the model, and generates in real time the optimal thrust command set and the optimal attitude control command set that track the reference sequence and satisfy the physical constraints of the actuator.

[0010] Preferably, the hierarchical predictive control architecture in step S3 is a hierarchical model predictive control architecture, wherein the upper layer is an upper-layer optimizer and the lower layer is a lower-layer tracker. The state observer is an extended Kalman filter. The state vector of the extended Kalman filter includes the three-degree-of-freedom motion states of the pipelaying vessel (swell, roll, and pitch), the angle and angular velocity of the key hinge point of the support frame, and the environmental disturbance state composed of low-frequency wave drift force and ocean current force. The extended Kalman filter performs recursive calculations based on the unified nonlinear dynamics model and the fusion sensing system output measurement values ​​in step S2 to calculate the system's full state and environmental disturbance torques that cannot be directly measured in real time. The lower-level tracker uses the calculated environmental disturbance torques as feedforward compensation and adds them to the optimal thrust command set it generates.

[0011] Preferably, in step S2, the central controller has a built-in sensor signal fusion and fault diagnosis module. This sensor signal fusion and fault diagnosis module uses an adaptive weighted fusion algorithm for redundant backup data from multiple sensors of the same type deployed at the same monitoring point for key monitoring parameters. The weights of the adaptive weighted fusion algorithm are dynamically allocated according to the historical error statistics, real-time signal-to-noise ratio and health status indicators of each sensor to output the optimal estimate of the system status. The key monitoring parameters include ship position, attitude, hinge point angle, pipe tension and stress. Meanwhile, the sensor signal fusion and fault diagnosis module performs real-time fault diagnosis based on the chi-square test method, calculates the residual sequence between the output value of each sensor and the predicted value of the system state, constructs the residual covariance matrix and calculates its chi-square statistic; when the chi-square statistic exceeds the preset threshold, it is determined that the corresponding sensor has failed. Once a fault is diagnosed, the data weight of the faulty sensor is immediately reduced to zero, and the data stream of the backup sensor at the monitoring point is seamlessly switched. At the same time, the fault identifier and timestamp are recorded and reported to the system log.

[0012] Preferably, it also includes an independent and parallel-operating emergency avoidance subsystem. This emergency avoidance subsystem has built-in multi-level safety thresholds and continuously monitors the equivalent stress of key pipeline nodes, the rate of change of tensioner tension, and the thrust status of each thruster in real time. When any monitored parameter exceeds its corresponding safety threshold, the emergency avoidance subsystem immediately issues the highest priority interruption, overriding the current output command of the central controller, forcing the system to switch to the preset emergency avoidance control mode. It instructs the support frame adjustment system to perform an upward movement at the maximum safe angular velocity to avoid wave loads. At the same time, it instructs the dynamic positioning system to control the tensioner to relax the pipeline tension at a given rate and calls the robust control algorithm of the lower tracker to switch the ship control mode from trajectory tracking to fixed-point position holding mode. The system control is returned to the central controller only after all parameters have returned to normal and become stable.

[0013] Preferably, in the online rolling solution process of step S3, the saturation constraints of the action rate and torque of each hydraulic actuator of the support frame adjustment system and the saturation constraints of the thrust of each propeller of the dynamic positioning system are directly embedded into the solution framework of the sequential quadratic programming algorithm as inequality constraints.

[0014] Preferably, a model prediction error compensation mechanism is introduced in the upper-level optimizer in step S3. This model prediction error compensation mechanism constructs a model error estimator based on the deviation between the output of the state observer in the lower-level tracker and the actual system state, and performs online correction on the unified nonlinear dynamic model to reduce the accumulation of prediction errors caused by model mismatch.

[0015] Preferably, the sensor signal fusion and fault diagnosis module is further configured with a virtual sensor reconstruction unit. When all physical sensors at a certain monitoring point are diagnosed as faulty, the virtual sensor reconstruction unit reconstructs the state estimate of the monitoring point based on a unified nonlinear dynamic model and the data of the remaining effective sensors through a Kalman filter algorithm, and temporarily replaces the output of the faulty sensor until the physical sensor recovers or manual intervention is required.

[0016] Preferably, the emergency avoidance subsystem further includes a smooth switching mechanism for a smooth transition when the system switches from emergency avoidance mode back to central controller control; the smooth switching mechanism specifically includes: After the conditions for exiting emergency avoidance mode are met, the system does not switch immediately, but instead initiates a preset transition time window. During this transition time window, the upper-level optimizer and lower-level tracker of the central controller begin to run synchronously, but the control commands they output are not executed immediately. Instead, they are compared and synchronized with the current commands of the emergency avoidance subsystem. A linear interpolation algorithm is used as the transition function to generate a time-varying mixing coefficient from 0 to 1; During the transition period, the final composite control command issued to the actuator is a weighted synthesis of the emergency avoidance command and the central controller command based on the time-varying mixing coefficient, which starts from 0 and gradually increases to 1 over time. When the time-varying mixing coefficient increases to 1, the composite control command is entirely composed of outputs from the central controller. Once the control is transferred, the emergency avoidance subsystem immediately switches to a dormant monitoring state.

[0017] Preferably, in the solution framework of the sequential quadratic programming algorithm, different priority weights and slack variable penalty coefficients are configured for different types of inequality constraints; Among them, the thrust saturation constraint of each thruster of the dynamic positioning system and the tension safety constraint of the pipeline tensioner are set as the first priority and their corresponding first penalty coefficients; the action rate constraint of the support frame adjustment system is set as the second priority, which is lower than the first priority and its corresponding second penalty coefficient, and the second penalty coefficient is less than the first penalty coefficient.

[0018] The present invention has at least the following beneficial effects: First, this invention establishes a unified nonlinear dynamic model of the coupled motion of the pipelaying vessel, the support frame, the pipeline, and the marine environment, providing a precise predictive basis for collaborative control and fundamentally solving the control conflict problem caused by model deficiencies. Through a hierarchical predictive control architecture, it achieves a combination of long-term optimization and short-term rapid response, ensuring the real-time nature and forward-looking nature of the control. The fusion of multi-sensor data provides the controller with comprehensive system state information. Ultimately, this method can output an optimal instruction set that satisfies the physical constraints of all actuators, achieving high-precision and robust collaborative control of the pipelaying vessel's posture and the support frame's shape, significantly improving the safety and efficiency of pipeline laying operations.

[0019] Secondly, by setting up a two-layer structure with different control cycles, this invention effectively balances the contradiction between complex optimization calculations and rapid response. The upper-layer optimizer has ample time for accurate long-term optimization, while the lower-layer tracker can handle model uncertainties and perform robust control at a higher frequency. This ensures that the system maintains good tracking performance and control stability when facing model mismatch and dynamic effects not described in the model, thereby enhancing the system's adaptability in complex and variable marine environments.

[0020] Third, by employing an extended Kalman filter as a state observer, this invention can efficiently and accurately estimate the system's full state and environmental disturbances, which cannot be directly measured. The estimated environmental disturbance torque is introduced as a feedforward compensation quantity into the lower-level controller, significantly improving the control system's ability to resist environmental disturbances, making thrust commands more precise, reducing steady-state errors in ship position and attitude, and enhancing the accuracy of dynamic positioning and the overall system's anti-interference performance.

[0021] Fourth, this invention, through an adaptive weighted fusion algorithm, fully utilizes redundant sensor data and dynamically optimizes weight allocation, thereby outputting more accurate and reliable estimates of the system's critical states, providing a high-quality data foundation for advanced control algorithms. Combined with a real-time fault diagnosis mechanism based on the chi-square test, it can quickly identify sensor faults and seamlessly switch to backup data streams, greatly improving the perception robustness and fault tolerance of the control system, and avoiding the risk of system misjudgment or loss of control due to single-point sensor failure.

[0022] Fifth, this invention provides a reliable safety layer for the entire collaborative control system by establishing an independent and highest-priority emergency avoidance subsystem. It bypasses the complex computational processes of the central controller, responding in milliseconds to emergency situations and executing the most effective pre-set protective actions. This maximizes the protection of pipelines and ships, effectively preventing accidents from occurring or escalating, and providing crucial safety assurance for high-risk operations.

[0023] Sixth, this invention directly embeds the physical saturation constraints of the actuator into the solution framework of the optimization problem, ensuring from the source that any optimal control command generated is physically executable. This avoids the generation of commands that are theoretically optimized but practically unachievable, ensuring the consistency between the actual effect of the control system and the theoretical design, and improving the effectiveness of control and the stability of the system.

[0024] Seventh, by introducing a model prediction error compensation mechanism, this invention can correct the unified nonlinear dynamic model online, making it continuously approximate the dynamic characteristics of the real system. This effectively suppresses the accumulation of prediction errors caused by inaccurate initial models or time-varying system characteristics, maintains the long-term accuracy of the prediction model, and thus ensures the continuity and reliability of the rolling optimization control effect based on model prediction.

[0025] Eighth, this invention, through a virtual sensor reconstruction unit, can still provide a state estimate of the monitoring point based on the model and information from the remaining sensors, even in the extreme case where all physical sensors fail, ensuring the integrity of the system state information. This prevents the control system from completely failing due to a complete lack of information, maintains the system's ability to operate in degraded mode, and improves task continuity and system availability.

[0026] Ninth, this invention avoids potential command jumps and system shocks when control is handed back from the emergency avoidance subsystem to the central controller through a smooth switching mechanism. By employing a weighted synthesis method with time-varying hybrid coefficients, a smooth transition of control commands is achieved, ensuring the continuity and stability of system state and ship motion, facilitating smooth operation recovery, and improving the overall operational experience and safety of the system.

[0027] Tenth, this invention incorporates engineering safety principles into the algorithm's solution process by configuring priorities and differentiated penalty coefficients for optimization constraints. It ensures that in the event of constraint conflicts, the solver prioritizes satisfying critical constraints related to core safety, even at the cost of minor performance. This design makes the system's behavior more consistent with safety-first engineering logic, further enhancing the system's inherent security.

[0028] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating the collaborative control method of the pipelaying vessel dynamic positioning and the pipe-laying support frame of the present invention. Detailed Implementation

[0030] The present invention will now be described in further detail so that those skilled in the art can implement it based on the description.

[0031] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0032] like Figure 1 As shown, this embodiment of the invention provides a method for coordinated control of dynamic positioning and pipe-laying support frame on a pipe-laying vessel, including: S1. Establish a unified nonlinear dynamic model for the coupled motion of the pipelaying vessel, the pipe-laying support, the pipeline, and the marine environment. Specifically, the pipe-laying support is considered as a multi-segment flexible beam structure connected to the hull via a universal hinge mechanism, and the interaction torque and force between the support and the hull are calculated. The pipeline is represented as a beam element model based on the absolute nodal coordinate method, and its shape and stress distribution from the pipe-laying support roller to the seabed contact point are calculated in real time. The hydrodynamic forces on the hull and the pipe-laying support are introduced into the unified nonlinear dynamic model based on the three-dimensional potential flow theory, and environmental loads based on the Morrison formula are also introduced. S2. The central controller deployed on the pipelaying vessel receives and integrates monitoring data in real time from the ship's position reference system, IMU attitude sensor, tilt and torque sensors at each hinge point of the pipe support frame, pipe tensioner tension sensor, and fiber optic stress sensors located at the end of the pipe support frame and key nodes of the pipe. S3. The central controller adopts a hierarchical predictive control architecture, which includes an upper layer based on an optimization model and a lower layer responsible for tracking control. The upper layer, based on a unified nonlinear dynamics model and the fusion sensing system state obtained in step S2, performs forward simulation calculations on the evolution of the system state in the future finite time domain; it uses a sequential quadratic programming algorithm to solve the multi-objective optimization problem online in a rolling manner, and outputs the reference trajectory sequence of the pipelaying vessel and the reference shape sequence of the support frame; the lower layer integrates a state observer to estimate the state variables in the system that cannot be directly measured; the lower layer receives the output sequence of the upper layer and, combined with the solution results of the state observer, generates in real time the optimal thrust command set and the optimal attitude control command set that satisfy the physical constraints of all actuators and track the reference sequence. S4. The optimal thrust command set is sent to each thruster of the dynamic positioning system, and the optimal attitude control command set is sent to each hydraulic actuator of the pipelaying support adjustment system to achieve coordinated control of the pipelaying vessel's position and attitude and the support structure.

[0033] In the above embodiments, high-precision collaborative operation between the pipelaying vessel and the support frame system is achieved by establishing a unified nonlinear dynamic model, multi-sensor information fusion, hierarchical predictive control, and command issuance and execution. First, in step S1, a unified nonlinear dynamic model is proposed to establish the coupled motion of the pipelaying vessel, support frame, pipeline, and marine environment. This unified nonlinear dynamic model treats the support frame as a multi-segment flexible beam structure connected to the hull via a universal joint mechanism, fully considering the interaction forces and moments between it and the hull, thus more realistically reflecting the dynamic response of the support frame caused by ship motion during actual operations. Pipeline modeling adopts a beam element model based on the Absolute Nodal Coordinates (ANCF) method. This method can accurately capture the nonlinear deformation and stress distribution of the pipeline from the support frame rollers to the seabed contact point, and is particularly suitable for dynamic simulation of flexible bodies with large displacements and rotations. Furthermore, the unified nonlinear dynamic model also introduces hydrodynamic loads on the hull and support frame calculated based on three-dimensional potential flow theory, as well as environmental loads such as wind, waves, and currents calculated according to the Morrison formula, thereby achieving a comprehensive description of the coupling effects of multiple physics fields within a unified framework. The establishment of this unified nonlinear dynamic model is the foundation of the entire control method, and its high accuracy and completeness provide reliable mathematical model support for subsequent prediction and control.

[0034] In step S2, the central controller deployed on the pipelaying vessel receives and fuses data from multiple types of sensors in real time. These sensors include a vessel position reference system (such as DGPS or acoustic positioning systems), IMU attitude sensors, tilt and torque sensors at each hinge point of the pipe support frame, pipe tensioner tension sensors, and fiber optic stress sensors located at the ends of the pipe support frame and key pipe nodes. The central controller acquires the outputs of these heterogeneous sensors at a high-frequency sampling rate through its built-in data acquisition and communication module, and uses signal conditioning and time synchronization techniques to ensure data consistency. In the fusion processing, an adaptive weighted fusion algorithm or Kalman filtering technique can be used to integrate data from different sources and with different levels of precision to form a comprehensive and high-confidence estimate of the system state (such as vessel position, attitude, hinge point angles, pipe tension and stress, etc.). The key to this step is to improve the robustness and accuracy of state perception through multi-source information fusion, providing reliable input for control decisions.

[0035] Step S3 details the hierarchical predictive control architecture adopted by the central controller. This architecture consists of two layers. The upper layer, based on the unified nonlinear dynamic model established in step S1 and the fused system state obtained in step S2, performs forward simulation calculations within a finite future time domain. It then uses optimization algorithms such as Sequential Quadratic Programming (SQP) to solve the multi-objective optimization problem online, outputting a reference trajectory sequence for the pipelaying vessel and a reference shape sequence for the support frame. The lower layer integrates a state observer (such as an extended Kalman filter) to estimate state variables in the system that cannot be directly measured (such as some environmental disturbances and articulated point angular velocities). It receives the output sequence from the upper layer and, combined with the observer's solution, uses Model Predictive Control (MPC) or robust control methods to generate, in real time, the optimal thrust command set and optimal attitude control command set that satisfy the physical constraints of all actuators. This hierarchical structure effectively coordinates the contradiction between long-term optimization and short-term response. The upper layer focuses on global optimization and forward-looking decision-making, while the lower layer is responsible for high-frequency command generation and disturbance suppression, jointly ensuring the system's control performance under complex sea conditions.

[0036] Step S4 describes the execution process of the control commands: the optimal thrust command set generated in step S3 is sent to each thruster of the dynamic positioning system, and the optimal attitude control command set is sent to each hydraulic actuator of the pipelaying support adjustment system. The thrusters adjust the magnitude and direction of thrust according to the commands to achieve precise control of the ship's position and heading; the hydraulic actuators adjust the pitch and roll angles of each section of the pipelaying support according to the commands, ensuring that its shape always tracks the reference sequence. This process achieves fast and reliable transmission of commands through a high-bandwidth control bus, and ensures the accuracy of the actions through the closed-loop control of the actuators themselves. The entire command issuance and execution process achieves real-time coordinated control of the pipelaying vessel's position and attitude and the shape of the pipelaying support, improving the quality and safety of pipeline laying operations.

[0037] The control method proposed in this embodiment achieves integrated control of the pipelaying vessel's dynamic positioning system and the support frame adjustment system by organically combining key technologies such as unified modeling, multi-sensor fusion, hierarchical prediction, and collaborative execution. Compared with the existing technology that typically controls the two independently, this invention effectively solves the problem of decreased control performance caused by model deficiencies and control conflicts, improves the system's adaptability, robustness, and control accuracy in complex marine environments, and is suitable for challenging operation scenarios such as deep-water pipelaying.

[0038] In one specific implementation, in step S3, the central controller adopts a hierarchical model predictive control architecture, which includes an upper-level optimizer and a lower-level tracker. The upper-level optimizer operates in a first control cycle, performs forward simulation calculations based on the unified nonlinear dynamics model, solves a multi-objective optimization problem, and outputs a reference trajectory sequence for the pipelaying vessel and a reference shape sequence for the support frame. The lower-level tracker operates in a second control cycle, which is shorter than the first control cycle, receives the output reference sequence from the upper-level optimizer, and uses a robust model predictive control algorithm to handle model mismatch and dynamic effects not described in the model, and generates in real time the optimal thrust command set and the optimal attitude control command set that track the reference sequence and satisfy the physical constraints of the actuator.

[0039] In the above implementation, the hierarchical model predictive control frame comprises an upper-level optimizer and a lower-level tracker, which are functionally distinct yet closely coordinated. The upper-level optimizer is responsible for forward simulation and optimization calculations based on the aforementioned unified nonlinear dynamics model. It operates with a relatively long first control cycle, which may range from several seconds to tens of seconds, such as 5 seconds, 10 seconds, or 20 seconds, with the specific value adjustable according to the computational resources and model complexity of the actual system. Within this cycle, the upper-level optimizer has sufficient time to perform computationally intensive future state predictions and multi-objective optimization solutions. Its output is a sequence of reference trajectories for the pipelaying vessel over a future period, such as planned track points, heading angles, and a sequence of reference morphological sequences for the pipelaying frame, such as the target angles of each articulated segment. This process is essentially a forward-looking planning based on the model, providing global optimization guidance for the system.

[0040] The lower-level tracker operates with a second control cycle significantly shorter than the first control cycle of the upper-level tracker, typically on the order of hundreds of milliseconds to one second, such as 100, 200, or 500 milliseconds, to accommodate the rapid response requirements of the actuators and high-frequency disturbances. It receives a reference sequence from the upper-level optimizer as its tracking target. To effectively handle the inevitable model mismatch in real-world systems—the difference between the mathematical model and the actual physical system—and dynamic effects not described in the model, such as high-frequency wave forces and unmodeled hydrodynamic effects, the lower-level tracker employs a robust model predictive control (Robust MPC) algorithm. This algorithm, when generating the optimal thrust command set and attitude control command set, considers model uncertainties and the possible range of external disturbances. By optimizing a minimum-maximum or robust performance index, it ensures that even under worst-case disturbances, the generated control commands satisfy the physical constraints of all actuators, such as thrust saturation and rate limits, and drive the system state to stably track the reference target provided by the upper level.

[0041] The upper layer performs low-frequency, strategic optimization planning, while the lower layer performs high-frequency, tactical tracking and disturbance suppression. This division of labor effectively balances the time-consuming nature of complex optimization calculations with the system's real-time requirements for control commands. The application of robust model predictive control algorithms enables the lower-level controller to no longer rely entirely on the absolute accuracy of the model, but rather to cope with uncertainties, thereby enhancing the reliability of the entire control system and its adaptability in complex and variable marine environments. The entire architecture operates in a cyclical manner: the upper layer periodically updates the optimization scheme based on the latest system state, while the lower layer continuously receives reference commands at a high frequency, estimates the current state, solves the robust optimization problem, and outputs the final executable control commands.

[0042] The dual-cycle structure and robust control method defined in this implementation effectively solve the contradiction between the time consumption of complex system optimization calculation and the real-time control requirements by combining long-cycle optimization with short-cycle fast robust tracking. It also improves the control system's ability to cope with model uncertainties and unknown environmental disturbances, making the entire pipelaying vessel's collaborative control system exhibit stronger adaptability, stability, and reliability.

[0043] In one specific implementation, the hierarchical predictive control architecture in step S3 is specifically a hierarchical model predictive control architecture, wherein the upper layer is an upper-layer optimizer and the lower layer is a lower-layer tracker. The state observer is an extended Kalman filter. The state vector of the extended Kalman filter includes the three-degree-of-freedom motion states of the pipelaying vessel (swell, roll, and pitch), the angle and angular velocity of the key hinge point of the support frame, and the environmental disturbance state composed of low-frequency wave drift force and ocean current force. The extended Kalman filter performs recursive calculations based on the unified nonlinear dynamics model and the fusion sensing system output measurement values ​​in step S2 to calculate the system's full state and environmental disturbance torques that cannot be directly measured in real time. The lower-level tracker uses the calculated environmental disturbance torques as feedforward compensation and adds them to the optimal thrust command set it generates.

[0044] In the above implementation, the specific structure of the hierarchical predictive control architecture is further clarified, indicating that the upper layer is an upper-level optimizer and the lower layer is a lower-level tracker. Based on this, a specific implementation method for a state observer is introduced. The state observer is specifically defined as an Extended Kalman Filter (EKF). The EKF is an advanced state estimation algorithm suitable for nonlinear systems. Its core principle is to locally linearize the nonlinear system model, and then apply the prediction and update steps of the standard Kalman filter to recursively fuse the system model predictions with the actual sensor measurements, thereby obtaining the optimal estimate of the system's internal state. In this cooperative control system, the EKF is not an independent module, but is deeply integrated into the lower-level tracker, providing it with key state information that cannot be directly obtained through sensors, which is a prerequisite for achieving high-precision control.

[0045] The specific composition of the state vector estimated by the extended Kalman filter is further elaborated. This state vector is a comprehensive set that can fully describe the dynamic characteristics of the system. It includes not only the three degrees of freedom of motion of the pipelaying vessel in the horizontal plane, namely sway (forward and backward movement), roll (left and right movement), and pitch (bow turning), which are the core control objects of dynamic positioning; it also includes the angles and angular velocities of the key hinge points of the pipelaying platform, which are directly related to the shape of the platform and its dynamic change trend; more importantly, the state vector also includes the environmental disturbance state composed of low-frequency wave drift force and ocean current force. Estimating environmental disturbances as states is a key application of EKF here, enabling the system to actively "sense" rather than just "respond" to continuous environmental forces. The recursive operation process of EKF is as follows: In each control cycle, which may coincide with the cycle of the lower-level tracker, such as 100 milliseconds or 200 milliseconds, it receives measurement values ​​from the fusion sensing system in step S2, such as GPS position, IMU attitude angle, hinge point angle, etc., and predicts what these measurement values ​​should be based on a unified nonlinear dynamic model; by comparing the predicted values ​​with the actual measured values, and correcting the state estimate based on the calculated Kalman gain, it can solve the full state of the system, including environmental disturbance torque, which cannot be directly measured, in real time.

[0046] The paper further describes how the state information, particularly the environmental disturbance torque, accurately estimated by the EKF, is utilized by the lower-level tracker to improve control performance. When generating the optimal thrust command set, the lower-level tracker does not rely solely on feedback calculations based on the deviation between the reference trajectory and the current ship position. Instead, it directly superimposes the environmental disturbance torque calculated by the EKF as a feedforward compensation onto its calculated thrust command. This is a composite control strategy combining feedforward and feedback. The principle is that feedforward control is responsible for offsetting predictable or estimable disturbances, such as continuous current forces and low-frequency wave drift forces, thus significantly reducing the burden on feedback control; feedback control focuses on handling residual, unmodeled, or random high-frequency disturbances. This combined approach makes the final generated thrust command more accurate and smooth, proactively counteracting environmental disturbances rather than correcting them after they have caused a positional shift, thereby improving the control accuracy and stability of the dynamic positioning system.

[0047] This implementation employs an extended Kalman filter as the core state observer, and carefully designs its state vector to encompass ship motion, rack dynamics, and environmental disturbances. Then, through a feedforward compensation mechanism, the estimated disturbance torque is directly used for the control output, constructing an advanced perception-estimation-compensation control loop. This technical solution improves the control system's perception depth and estimation accuracy of internal states and external environments, effectively enhancing the lower-level tracker's ability to resist continuous environmental disturbances. This results in the entire cooperative control system exhibiting superior robustness, smaller steady-state error, and higher overall control quality.

[0048] In one specific embodiment, in step S2, the central controller has a built-in sensor signal fusion and fault diagnosis module. This sensor signal fusion and fault diagnosis module uses an adaptive weighted fusion algorithm to perform redundant backup data from multiple sensors of the same type deployed at the same monitoring point for key monitoring parameters. The weights of the adaptive weighted fusion algorithm are dynamically allocated according to the historical error statistics, real-time signal-to-noise ratio and health status indicators of each sensor to output the optimal estimate of the system status. The key monitoring parameters include ship position, attitude, hinge point angle, pipe tension and stress. Meanwhile, the sensor signal fusion and fault diagnosis module performs real-time fault diagnosis based on the chi-square test method, calculates the residual sequence between the output value of each sensor and the predicted value of the system state, constructs the residual covariance matrix and calculates its chi-square statistic; when the chi-square statistic exceeds the preset threshold, it is determined that the corresponding sensor has failed. Once a fault is diagnosed, the data weight of the faulty sensor is immediately reduced to zero, and the data stream of the backup sensor at the monitoring point is seamlessly switched. At the same time, the fault identifier and timestamp are recorded and reported to the system log.

[0049] In the above implementation, the sensor signal fusion and fault diagnosis module processes redundant backup data from multiple similar sensors deployed at the same monitoring point for key monitoring parameters such as ship position, attitude, articulation point angle, and pipe tension and stress. It employs an adaptive weighted fusion algorithm, which has the advantage of dynamically and intelligently allocating weights to each sensor's data. The weight allocation is based on a comprehensive consideration of multiple factors: historical error statistics for each sensor, such as the root mean square error over a past period, reflecting its long-term accuracy and stability; real-time signal-to-noise ratio, used to assess data quality under the current measurement environment; and health status indicators, such as sensor self-diagnostic signals or voltage monitoring values. Through this multi-dimensional evaluation, the algorithm can automatically reduce the weight of sensors with temporary accuracy degradation or those subject to interference, while increasing the contribution of more reliable sensors, thereby outputting the optimal estimate of the system's current state. This dynamic fusion process runs continuously, providing the upper-level optimizer and lower-level tracker with system state information far more accurate and reliable than data from a single sensor.

[0050] A real-time fault diagnosis mechanism based on the chi-square test, running in parallel with a fusion algorithm, constitutes a dual guarantee for the system's perception layer. Its working principle is as follows: First, a predicted value of the system state is generated using the system's unified nonlinear dynamic model and state observer. Then, the residual sequence between the output values ​​of each sensor and the corresponding predicted state values ​​is calculated. Under normal system operation, without faults, and with an accurate model, this residual sequence should follow a Gaussian distribution with a mean of zero. Based on this, the algorithm constructs a residual covariance matrix and calculates its chi-square statistic, the magnitude of which reflects the degree of deviation between the actual residuals and the expected distribution. The system presets a fault judgment threshold, the selection of which is related to the required confidence level. For example, a threshold corresponding to 99% or 99.9% confidence level might be used, but this needs to be adjusted according to the specific application's tolerance for false alarms and missed alarms. When the calculated chi-square statistic consistently exceeds this preset threshold, the corresponding sensor is determined to have malfunctioned, such as reading freeze, drift, or severe accuracy degradation.

[0051] Once a sensor fault is diagnosed, the system immediately executes a series of automated operations. First, during the data fusion phase, the data weight of the faulty sensor is instantly reduced to zero, ensuring it no longer affects the fusion results. Simultaneously, the system automatically and seamlessly switches to the backup sensor data stream for that monitoring point, ensuring an uninterrupted input source for state estimation. The entire switching process is completed within milliseconds, designed to avoid any disturbance to the control loop. Furthermore, as an important safety and maintenance support function, the module records the faulty sensor's identifier, fault type, and precise timestamp, reporting this information to the system log for subsequent maintenance analysis and troubleshooting. This entire automated process from diagnosis to response greatly ensures that the control system can continuously obtain reliable state information even when facing a single point of failure in a sensor, thereby maintaining the normal operation of the system.

[0052] The sensor signal fusion and fault diagnosis module described in detail in this implementation improves the accuracy and reliability of state estimation through adaptive weighted fusion, achieves rapid and intelligent diagnosis of sensor faults through chi-square test, and ensures the continuity of system perception through a seamless switching mechanism. These technical features work together to construct a highly robust and fault-tolerant perception system, which can effectively resist the risks caused by sensor accuracy degradation or complete failure, providing a solid and reliable data foundation for the entire pipelaying vessel collaborative control system.

[0053] In one specific embodiment, an independent and parallel emergency avoidance subsystem is also included. This emergency avoidance subsystem has built-in multi-level safety thresholds and continuously monitors the equivalent stress of key pipeline nodes, the rate of change of tensioner tension, and the thrust status of each thruster in real time. When any monitored parameter exceeds its corresponding safety threshold, the emergency avoidance subsystem immediately issues a highest-priority interrupt, overriding the current output command of the central controller, forcing the system to switch to a preset emergency avoidance control mode. It instructs the support frame adjustment system to perform an upward movement at the maximum safe angular velocity to avoid wave loads, and simultaneously instructs the dynamic positioning system to control the tensioner to relax the pipeline tension at a given rate. It also calls the robust control algorithm of the lower-level tracker to switch the ship control mode from trajectory tracking to fixed-point position holding mode until the system detects that all parameters have returned to normal and remain stable, and then returns system control to the central controller.

[0054] In the above implementation, an independent and parallel-operating emergency avoidance subsystem is introduced as the highest-level safety protection mechanism of the entire collaborative control system. This emergency avoidance subsystem is an independent module operating in parallel with the central controller at both the hardware and software levels. Its design aims to ensure that even if the central controller experiences delays due to complex calculations or unexpected failures, it can still respond within milliseconds based on the most critical safety indicators. The emergency avoidance subsystem has multiple preset safety thresholds, which are comprehensively set based on historical data from the yield strength of the pipeline material, the working capacity of the tensioner, the stability of the vessel, and the sea conditions. For example, the equivalent stress threshold of critical pipeline nodes may be set at a relatively high proportion of the material's yield strength; the tension change rate threshold of the tensioner may refer to its maximum allowable loading rate; and the thrust status of the propeller is monitored to see if it has reached saturation or overload. These safety thresholds are typically divided into multiple levels, such as early warning, alarm, and emergency avoidance, with different levels corresponding to different system response strategies. The system continuously monitors these parameters that most directly reflect the system's safety status in real time, with extremely short sampling and judgment cycles to ensure timely response.

[0055] When any monitored parameter, such as the stress at a point in the pipeline, suddenly and drastically increases and exceeds its highest safety threshold, the subsystem immediately issues a high-priority interrupt signal. This interrupt signal directly overrides all control commands currently output by the central controller, seizing control of the system and forcibly switching the entire system to a preset emergency avoidance control mode. Immediately afterward, it sends commands to the support frame adjustment system, instructing its hydraulic actuators to drive the support frame to tilt upward at the maximum angular velocity the system can safely execute (this velocity must be set within the allowable range of the support frame's structural strength). The primary purpose is to quickly raise the support frame and critical parts of the pipeline above the water surface or change their stress angle to avoid severe wave impact loads. Almost simultaneously, it sends commands to the dynamic positioning system, controlling the tensioner to relax the pipeline tension at a preset, controllable rate (this rate must avoid the pipeline from generating new dynamic effects due to excessively rapid relaxation), thereby quickly releasing excessive stress within the pipeline. Simultaneously, it invokes the robust control algorithm inherent in the lower-level tracker, but immediately switches the control objective from tracking the optimized trajectory to a simple fixed-point position-holding mode, i.e., maintaining the ship's current position and heading as much as possible, with stability and safety as the top priorities. It should be noted that when the emergency avoidance subsystem executes preset protective actions, the key parameters in its control commands are all finely set based on engineering safety principles. The "maximum safe angular velocity" is not the mechanical limit of the hydraulic actuator, but a safe upper limit set after comprehensively considering the structural strength of the support frame, the flow rate of the hydraulic system, and the dynamic interaction between the pipeline and the seabed. The purpose is to quickly avoid wave loads while avoiding secondary impacts on the structure. Its typical value is usually preset in the range of 0.5 to 2.0 degrees / second. Meanwhile, the "given rate relaxation" refers to the pipeline tension release rate, which is a pre-set controlled parameter. This rate must ensure that the tension release process tends to be quasi-static to avoid dynamic effects such as violent serpentine movement of the pipeline or impact with the seabed caused by excessively rapid relaxation. Its value is usually determined based on the pipeline diameter, water depth, and current tension level, and is sent to the tensioner closed-loop controller for execution in the form of a set value, such as adjusting it at a rate of 5% to 15% of the current tension per minute.

[0056] The system does not immediately exit this mode after executing an emergency action, but continues to monitor all critical parameters. Only after all parameters, such as pipeline stress and tension change rate, have been confirmed to have returned to normal safe ranges and remained stable for a period of time will the emergency avoidance subsystem determine that the danger has been averted. Afterward, instead of abruptly cutting off its own commands, it uses a pre-set, smooth transition process to return control of the system to the central controller. This ensures that a sudden switch in control source will not cause a secondary shock to the system's state. The system then returns to a parallel monitoring dormant state, ready to respond to the next possible emergency.

[0057] The independent emergency avoidance subsystem designed in this implementation provides an extremely reliable final safety barrier for the entire pipelaying operation system by establishing independent parallel monitoring channels, multi-level safety threshold judgment, highest priority interruption coverage, and a series of preset emergency avoidance actions. It greatly enhances the system's ability to respond to sudden extreme situations, enabling it to take the most effective measures at the initial stage of a hazard to maximize the protection of the pipeline, the support structure, and the vessel itself, thereby reducing accident risks and potential losses to a very low level.

[0058] In one specific implementation, during the online rolling solution process in step S3, the saturation constraints of the action rate and torque of each hydraulic actuator of the support frame adjustment system, as well as the saturation constraints of the thrust of each propeller of the dynamic positioning system, are directly embedded into the solution framework of the sequential quadratic programming algorithm as inequality constraints.

[0059] In the above implementation, the constraint handling mechanism of the online rolling solution process has been refined in detail. The purpose is to directly integrate the physical limits of the actuators as hard constraints into the solution framework of the optimization algorithm to ensure the physical feasibility of control commands. These constraints are essentially the inherent physical performance limits of various actuators in the pipelaying vessel's dynamic positioning system and the pipe rack adjustment system. For the pipe rack adjustment system, there are two main types of saturation constraints on its hydraulic actuators: one is the action rate saturation constraint, that is, there is an upper limit to the extension and retraction speed of each hydraulic cylinder. Its value may be set according to the flow rate of the hydraulic pump, the response of the valve, and the system stability requirements. For example, it may be in the range of a few degrees to tens of degrees per second, which limits the speed of pipe rack adjustment; the other is the torque saturation constraint, that is, there is an upper limit to the maximum thrust or torque that the hydraulic cylinder can output. This is determined by the rated pressure of the hydraulic system, the cylinder diameter, etc., and its value needs to be set according to the structural strength requirements to prevent overload damage to the articulation mechanism. For dynamic positioning systems, each thruster is subject to thrust saturation constraints. This means that the maximum forward and reverse thrust that each thruster can provide is fixed, and this value is determined by the thruster's own physical characteristics, such as motor power and propeller size. These constraints are absolute and insurmountable physical boundaries. Any theoretically perfect control command that exceeds these limits cannot be actually executed and may even lead to actuator saturation, system instability, or equipment damage.

[0060] Sequential quadratic programming is a widely used numerical method for solving constrained nonlinear optimization problems. In this invention, a multi-objective optimization problem is solved online in a rolling manner, considering factors such as ship tracking accuracy, maintenance of the support structure, and energy consumption. These physical saturation constraints are directly used as constraints in the form of inequalities (e.g., the absolute value of the command is less than or equal to the maximum value). This means that in each rolling solution process, the algorithm must find the optimal solution within a feasible region bounded by these inequalities, representing all physically realizable commands. Thus, the algorithm is fundamentally forced to generate control commands that match the actual capabilities of the actuator, thereby ensuring a high degree of unity between the theoretical value and engineering practicality of the optimization results.

[0061] During the controller initialization phase, the specific saturation constraint values ​​of all actuators, such as maximum thrust, maximum speed, and maximum torque, are pre-input as prior parameters into the solver settings of the sequential quadratic programming algorithm. Within each cycle of the upper-level optimizer, when the algorithm begins solving a new optimization problem, these constraints are constructed along with the system's dynamic model and objective function. During iterative calculations, the solver continuously verifies whether the current solution satisfies all constraints. If not, it automatically adjusts the search direction to ensure that the control requirements implied by the final output reference trajectory sequence and reference shape sequence are always within the physical capabilities of each actuator. This method fundamentally avoids the problem of control commands being detached from reality, making the optimization calculation closely coupled with physical reality.

[0062] This implementation method embeds key physical constraints into the core of the optimization algorithm, ensuring that all instructions generated by the cooperative control system are physically executable from the outset. This technical feature greatly enhances the practicality and reliability of the entire control system, avoiding performance degradation or system instability risks caused by infeasible instructions, and enabling theoretically optimal control to be seamlessly translated into the safe, stable, and efficient operation of the actual system.

[0063] In one specific implementation, a model prediction error compensation mechanism is introduced in the upper-level optimizer in step S3. This model prediction error compensation mechanism constructs a model error estimator based on the deviation between the output of the state observer in the lower-level tracker and the actual system state, and performs online correction on the unified nonlinear dynamic model to reduce the accumulation of prediction errors caused by model mismatch.

[0064] In the above implementation, the functionality of the upper-level optimizer has been significantly enhanced by introducing a model prediction error compensation mechanism. The aim is to dynamically correct the unified nonlinear dynamic model to address model mismatch issues and maintain long-term prediction accuracy. In practical applications of pipelaying vessel collaborative control, the initially established unified nonlinear dynamic model inevitably contains simplifications, approximations, or inaccurate parameters; this phenomenon is called model mismatch. Furthermore, the time-varying characteristics of marine environmental loads and the slow changes in the characteristics of the vessel and the hosting system over time will cause a gradual deviation between the model's predicted output and the actual system state. If this deviation is ignored, the prediction error will accumulate during the rolling time-domain optimization of the upper-level optimizer, causing the reference trajectory sequence and reference shape sequence calculated based on the model's look-ahead to gradually deviate from the actual optimal path. This will require the lower-level tracker to expend more control energy for correction, and may even lead to system performance degradation or instability. Therefore, online model calibration is crucial to ensuring the long-term effectiveness of predictive control.

[0065] The main idea behind this model prediction error compensation mechanism is to utilize the output of the state observer in the lower-level tracker. The state observer, by fusing sensor measurements, provides a high-confidence estimate of the system's true state. The mechanism calculates a sequence of deviations by continuously comparing the state estimates output by the state observer with the state predictions from the unified nonlinear dynamics model under the same input. This deviation sequence essentially reflects the inaccuracy of the model. Subsequently, a dedicated model error estimator is activated. This estimator can be a parameter estimator used to identify error parameters in the model online, or a disturbance estimator that treats the deviation as an unknown disturbance acting on the system and estimates it. The output of the model error estimator is used to correct the original unified nonlinear dynamics model in real time. Correction methods may include updating certain key parameters of the model, such as hydrodynamic coefficients or damping coefficients, or directly introducing a compensation term into the forward prediction equations to offset the estimated model errors.

[0066] The online calibration process is synchronized with the rolling optimization cycle of the upper-level optimizer. Before or after each optimization cycle, the model error estimator performs a calculation: it collects state observations and model predictions from a short time window, calculates the error, and executes the estimation algorithm. Subsequently, the estimated error parameters or compensation amounts are injected into the dynamic model, updating the model version used for forward simulation prediction in the next optimization cycle. This design makes the prediction model not a static entity, but a dynamic model capable of learning and adapting. It can gradually approximate the dynamic characteristics of the real system, effectively suppressing the accumulation of prediction errors caused by inaccurate initial models or slow time-varying characteristics of the system. This ensures that the decisions made by the upper-level optimizer based on the updated model are always closer to reality, maintaining the long-term accuracy and reliability of predictive control.

[0067] The model prediction error compensation mechanism introduced in this implementation method estimates and corrects model errors online using lower-level state observation information, endowing the control system with valuable adaptive capabilities. This technical feature enhances the realism and reliability of the unified nonlinear dynamic model during long-term operation, effectively curbs the drift of prediction errors, and ensures that model-predictive-based optimization decisions can continuously produce high-performance control effects, thereby enhancing the adaptability and robustness of the entire cooperative control system under different sea states and operational stages.

[0068] In one specific embodiment, the sensor signal fusion and fault diagnosis module is further configured with a virtual sensor reconstruction unit. When all physical sensors at a certain monitoring point are diagnosed as faulty, the virtual sensor reconstruction unit reconstructs the state estimate of the monitoring point based on a unified nonlinear dynamic model and the data of the remaining effective sensors through a Kalman filter algorithm, and temporarily replaces the output of the faulty sensor until the physical sensor recovers or manual intervention is required.

[0069] In the above implementation, a virtual sensor reconstruction unit is further configured on the basis of the sensor signal fusion and fault diagnosis module as a last-line guarantee mechanism to deal with the extreme case of complete sensor failure. This virtual sensor reconstruction unit is not used to handle regular or single sensor failures, but rather is activated in the extreme case where all physical sensors at a certain monitoring point are diagnosed as faulty. Although the probability of such a complete failure scenario is low, once it occurs, it means that all direct measurement data provided by that monitoring point is completely lost. Traditional redundancy design fails at this moment, and the system faces a huge risk of missing critical status information. For example, all tilt sensors at a critical hinge point of a support frame may fail simultaneously due to electrical short circuits, physical damage, or communication interruptions, or all tension sensors at a pipe tensioner may fail. In this case, without additional countermeasures, the control system will be unable to obtain the status information at that location and may be forced to degrade operation or even interrupt operations. The virtual sensor reconstruction unit is precisely the last line of defense designed to deal with such low-probability, high-risk events. Its purpose is to still provide the system with a status estimate of the monitoring point when all physical sensors fail, maximizing the information integrity and functionality of the control system.

[0070] The core function of the virtual sensor reconstruction unit is to infer the state value of the fault monitoring point based on existing information sources in the system through algorithms. Its reconstruction process relies primarily on two pillars of information: first, an established unified nonlinear dynamic model that describes the physical relationships and dynamic evolution of various state variables in the system; and second, measurement data from the remaining effective sensors in the system, which provides the latest state information for other parts of the system. The virtual sensor reconstruction unit typically employs advanced estimation algorithms for state reconstruction; the Kalman filter algorithm is a typical choice, but other state observers or soft measurement techniques can also be used. This algorithm uses the unified nonlinear dynamic model as the basis for state prediction, while simultaneously fusing and updating real-time data from other effective sensors as observations. Through this recursive process of prediction and correction, the state estimate of the fault monitoring point is finally calculated. Essentially, this process utilizes the system's model knowledge and the global sensor network to compensate for the complete lack of local sensing.

[0071] When the sensor signal fusion and fault diagnosis module confirms that all physical sensors at a monitoring point are faulty, it immediately sends an activation signal to the virtual sensor reconstruction unit, specifying the type of state variable to be reconstructed. Once activated, this unit continuously receives data from other valid sensors and the current system's control input at regular intervals, such as once per second or synchronized with the main control cycle, and executes its reconstruction algorithm. The generated state estimate is immediately output, seamlessly replacing the original faulty sensor's data stream, and input to the signal fusion module or directly provided to the controller, thus ensuring the continuity of the system's perception layer and the closure of the control loop. This virtual sensor reconstruction unit maintains this virtual sensing state until external conditions change: either the physical sensors at the faulty monitoring point are repaired and restored to normal operation, and the system automatically switches back to the physical sensor data stream after detecting that their output data has become reliable again; or the operator intervenes manually based on system alarms, confirming the fault or manually selecting another processing method. The virtual sensor reconstruction unit is then disabled, awaiting the next possible call.

[0072] The virtual sensor reconstruction unit described in this implementation provides the control system with fault tolerance depth and operational continuity by reconstructing critical states under extreme conditions using a unified nonlinear dynamics model and other sensor information. This technical feature greatly enhances the system's survivability in the face of the most severe sensor failures, ensuring that the control system's functionality does not completely collapse due to the complete loss of local information, thereby raising the system's reliability and availability to a very high level.

[0073] In one specific embodiment, the emergency escape subsystem further includes a smooth switching mechanism for a smooth transition when the system switches from emergency escape mode back to central controller control; the smooth switching mechanism specifically includes: After the conditions for exiting emergency avoidance mode are met, the system does not switch immediately, but instead initiates a preset transition time window. During this transition time window, the upper-level optimizer and lower-level tracker of the central controller begin to run synchronously, but the control commands they output are not executed immediately. Instead, they are compared and synchronized with the current commands of the emergency avoidance subsystem. A linear interpolation algorithm is used as the transition function to generate a time-varying mixing coefficient from 0 to 1; During the transition period, the final composite control command issued to the actuator is a weighted synthesis of the emergency avoidance command and the central controller command based on the time-varying mixing coefficient, which starts from 0 and gradually increases to 1 over time. When the time-varying mixing coefficient increases to 1, the composite control command is entirely composed of outputs from the central controller. Once the control is transferred, the emergency avoidance subsystem immediately switches to a dormant monitoring state.

[0074] In the above implementation, the smooth switching mechanism does not take effect immediately after the emergency is lifted. Instead, it determines that the conditions for exiting the emergency avoidance mode are met only after the emergency avoidance subsystem has continuously monitored and confirmed that all key parameters have returned to normal and remained stable for a preset time. At this point, the system does not immediately switch control but first initiates a preset transition time window. The length of this time window is a key design parameter, and its value needs to balance the smoothness of the switchover with the recovery efficiency. It may be set in the range of several seconds to tens of seconds, such as 5 seconds, 10 seconds, or 20 seconds. The specific duration needs to be comprehensively determined based on the dynamic response characteristics of the pipelaying vessel and the volatility of the sea conditions. Initiating this window signifies that the system has officially entered the transition procedure from emergency control to normal control.

[0075] Within the transition time window, the upper-level optimizer and lower-level tracker of the central controller are awakened and begin synchronous operation, recalculating the normal optimized trajectory and control commands. However, these newly calculated commands are not immediately issued to the actuators. Simultaneously, the emergency avoidance subsystem maintains its current control command output to ensure system stability. A dedicated synchronization and coordination module compares and synchronizes these two sets of commands—the current commands of the emergency avoidance subsystem and the newly generated commands from the central controller—ensuring they are on the same time reference. Subsequently, the system uses a preset transition function to generate a time-varying mixing coefficient that continuously changes from 0 to 1. Linear interpolation is a simple and reliable choice for implementing this function, allowing the mixing coefficient to linearly increase from an initial value of 0 to an ending value of 1 over time. This time-varying mixing coefficient essentially determines the weight distribution of the two control commands during the transition process.

[0076] Throughout the transition window, the final composite control commands issued to each actuator no longer originate solely from one party. Instead, they are a weighted synthesis of emergency avoidance commands and central controller commands based on a time-varying mixing coefficient. At the start of the transition, the mixing coefficient is 0, meaning the composite commands consist entirely of emergency avoidance commands, identical to before. As time progresses, the mixing coefficient gradually increases, giving central controller commands a larger weight in the composite commands, while the weight of emergency avoidance commands decreases accordingly. This process ensures that the commands received by the actuators are a continuous, smoothly changing quantity, completely avoiding abrupt jumps in thrust, torque, or angle commands. When the time-varying mixing coefficient finally increases to 1, the composite control commands are entirely composed of the central controller's output, marking the successful completion of the control transfer. The emergency avoidance subsystem then stops outputting commands, enters a dormant monitoring state, and continues to fulfill its safety protection responsibilities.

[0077] The smooth switching mechanism designed in this implementation cleverly solves the problems of instruction jumps and system impacts during the handover of control by introducing a transition time window, instruction synchronization comparison, and time-varying hybrid weighted synthesis. This technical feature greatly enhances the smoothness and continuity of the entire system's state switching, completely avoiding secondary interference or instability that may be caused by sudden changes in the control source, ensuring that operations can smoothly return to normal, and improving the overall user experience, safety, and reliability of the system.

[0078] In one specific implementation, in the solution framework of the sequential quadratic programming algorithm, different priority weights and slack variable penalty coefficients are configured for different types of inequality constraints. Among them, the thrust saturation constraint of each thruster of the dynamic positioning system and the tension safety constraint of the pipeline tensioner are set as the first priority and their corresponding first penalty coefficients; the action rate constraint of the support frame adjustment system is set as the second priority, which is lower than the first priority and its corresponding second penalty coefficient, and the second penalty coefficient is less than the first penalty coefficient.

[0079] In the above implementation, within the solution framework of the sequential quadratic programming algorithm, to achieve refined and differentiated processing of different types of inequality constraints, the system assigns different priority weights and corresponding slack variable penalty coefficients to various constraints. This approach stems from the fact that in engineering practice, different constraints have varying degrees of impact on system safety and performance. Therefore, they must be treated differently during the optimization process to ensure that the optimizer can make decisions that conform to engineering safety logic based on preset priorities when constraints conflict or the solution becomes difficult. Specifically, priority weights are used to rank constraints during the solution process, with higher-priority constraints being satisfied first. The slack variable penalty coefficient corresponds to the cost of violating a constraint in the optimization objective function when a constraint cannot be strictly satisfied and a slack variable is introduced. By assigning different penalty coefficients to different constraints, a trade-off mechanism is essentially introduced into the optimization objective, enabling the solver to satisfy all constraints as much as possible while tending to avoid violating those constraints with higher penalty coefficients, i.e., more critical ones.

[0080] Among these constraints, the thrust saturation constraint of each thruster in the dynamic positioning system and the tension safety constraint of the pipeline tensioner are set as the first priority, and corresponding first penalty coefficients are configured. The thrust saturation constraint refers to the maximum thrust limit that each thruster can provide. This value is determined by the physical design of the thruster, and may range from tens to hundreds of kilonewtons per thruster, with the specific value depending on the thruster model and power rating. The tension safety constraint means that the tension maintained by the pipeline tensioner must be within a safe range. Its upper limit needs to take into account factors such as the tensile strength of the pipeline material, the coefficient of friction of the seabed, and the operating water depth. For example, it may be set at 70% to 80% of the pipeline yield tension to avoid excessive stretching that could lead to pipeline damage or breakage. These two types of constraints are set as the first priority because they are directly related to the ship's dynamic positioning capability and the structural integrity of the pipeline; any violation may immediately cause serious safety accidents or equipment damage. In the process of solving sequential quadratic programming, the algorithm will prioritize ensuring that the solution meets these constraints. The first penalty coefficient is usually set to a large value, such as possibly in the range of 1,000 to 10,000. Such a setting means that once the optimization solution needs to slightly violate these constraints (by introducing slack variables), the objective function value will increase sharply, thus forcing the solver to avoid this situation as much as possible and only allowing a very small violation when absolutely necessary.

[0081] In contrast, the motion rate constraint of the rack adjustment system is set as the second priority, below the first priority, and is configured with a corresponding second penalty coefficient. The motion rate constraint refers to the maximum speed limit of the hydraulic actuator driving the rack's shape change. This value is determined by the hydraulic system flow rate, the actuator's mechanical design, and structural vibration considerations, and may be set in the range of a few degrees to tens of degrees per second. While this type of constraint is important, affecting the smoothness of the adjustment process and equipment lifespan, its urgency and direct impact on system safety are generally lower than thrust saturation and tension safety constraints. Therefore, its priority is set to second level, and the second penalty coefficient is also set to a value significantly smaller than the first penalty coefficient, for example, possibly in the range of 100 to 500. This means that during the optimization process, if a constraint conflict is encountered, the solver can allow a relatively large relaxation of the motion rate constraint, i.e., allowing the actuator to operate at a speed slightly exceeding the preset maximum rate, provided that the first priority constraint is satisfied first. Although this is not ideal, this trade-off is acceptable to ensure the more critical safety constraints. Although the penalty coefficient is low, it is still sufficient to ensure that the algorithm will try its best to comply with the constraint when there is no conflict.

[0082] The technical solution designed in this implementation embeds engineering safety principles directly into the mathematical solution process of the optimization problem by configuring differentiated priorities and penalty coefficients for inequality constraints in the sequential quadratic programming algorithm. In actual operation, these configuration parameters are preset during controller initialization. In each upper-level optimization cycle, when the optimization problem is constructed and the sequential quadratic programming solver is invoked, this priority and penalty coefficient information is loaded together. During iterative computation, the solver follows these preset rules for searching: it first strives to find a solution that satisfies all constraints; if this is not possible, it prioritizes satisfying the first priority constraint, even if this comes at the cost of violating the second priority constraint; and, since the first penalty coefficient is much higher than the second penalty coefficient, any slight violation of the first priority constraint will lead to a significant increase in the objective function value, thus being effectively suppressed, while the cost of violating the second priority constraint is relatively small. This mechanism ensures that the control instruction sequence generated by the optimizer always prioritizes the satisfaction of core safety constraints, and only then considers performance optimization objectives, making the behavior of the entire system more in line with the engineering logic of "safety first".

[0083] The number of devices and processing scale described herein are for the purpose of simplifying the description of the invention. Applications, modifications, and variations of the invention will be readily apparent to those skilled in the art.

[0084] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details.

Claims

1. A method for coordinated control of dynamic positioning and pipelaying support for pipelaying vessels, characterized in that, include: S1. Establish a unified nonlinear dynamic model for the coupled motion of the pipelaying vessel, the pipe-laying support, the pipeline, and the marine environment. Specifically, the pipe-laying support is considered as a multi-segment flexible beam structure connected to the hull via a universal hinge mechanism, and the interaction torque and force between the support and the hull are calculated. The pipeline is represented as a beam element model based on the absolute nodal coordinate method, and its shape and stress distribution from the pipe-laying support roller to the seabed contact point are calculated in real time. The hydrodynamic forces on the hull and the pipe-laying support are introduced into the unified nonlinear dynamic model based on the three-dimensional potential flow theory, and environmental loads based on the Morrison formula are also introduced. S2. The central controller deployed on the pipelaying vessel receives and integrates monitoring data in real time from the ship's position reference system, IMU attitude sensor, tilt and torque sensors at each hinge point of the pipe support frame, pipe tensioner tension sensor, and fiber optic stress sensors located at the end of the pipe support frame and key nodes of the pipe. S3. The central controller adopts a hierarchical predictive control architecture, which includes an upper layer based on an optimization model and a lower layer responsible for tracking control. The upper layer, based on a unified nonlinear dynamics model and the fusion sensing system state obtained in step S2, performs forward simulation calculations on the evolution of the system state in the future finite time domain; it uses a sequential quadratic programming algorithm to solve the multi-objective optimization problem online in a rolling manner, and outputs the reference trajectory sequence of the pipelaying vessel and the reference shape sequence of the support frame; the lower layer integrates a state observer to estimate the state variables in the system that cannot be directly measured; the lower layer receives the output sequence of the upper layer and, combined with the solution results of the state observer, generates in real time the optimal thrust command set and the optimal attitude control command set that satisfy the physical constraints of all actuators and track the reference sequence. S4. The optimal thrust command set is sent to each thruster of the dynamic positioning system, and the optimal attitude control command set is sent to each hydraulic actuator of the pipelaying support adjustment system to achieve coordinated control of the pipelaying vessel's position and attitude and the support structure.

2. The method for coordinated control of pipelaying vessel dynamic positioning and pipe-laying support frame according to claim 1, characterized in that, In step S3, the central controller adopts a hierarchical model predictive control architecture, which includes an upper-level optimizer and a lower-level tracker. The upper-level optimizer operates in a first control cycle, performs forward simulation calculations based on the unified nonlinear dynamics model, solves a multi-objective optimization problem, and outputs a reference trajectory sequence for the pipelaying vessel and a reference shape sequence for the support frame. The lower-level tracker operates in a second control cycle, which is shorter than the first control cycle. It receives the output reference sequence from the upper-level optimizer and uses a robust model predictive control algorithm to handle model mismatch and dynamic effects not described in the model, and generates in real time the optimal thrust command set and the optimal attitude control command set that track the reference sequence and satisfy the physical constraints of the actuators.

3. The method for coordinated control of pipelaying vessel dynamic positioning and pipe-laying support frame according to claim 2, characterized in that, The hierarchical predictive control architecture in step S3 is specifically a hierarchical model predictive control architecture, where the upper layer is an upper-layer optimizer and the lower layer is a lower-layer tracker. The state observer is an extended Kalman filter. The state vector of the extended Kalman filter includes the three-degree-of-freedom motion states of the pipelaying vessel (swell, roll, and pitch), the angle and angular velocity of the key hinge point of the support frame, and the environmental disturbance state composed of low-frequency wave drift force and ocean current force. The extended Kalman filter performs recursive calculations based on the unified nonlinear dynamics model and the fusion sensing system output measurement values ​​in step S2 to calculate the system's full state and environmental disturbance torques that cannot be directly measured in real time. The lower-level tracker uses the calculated environmental disturbance torques as feedforward compensation and adds them to the optimal thrust command set it generates.

4. The method for coordinated control of pipelaying vessel dynamic positioning and pipe-laying support frame according to claim 1, characterized in that, In step S2, the central controller has a built-in sensor signal fusion and fault diagnosis module. This sensor signal fusion and fault diagnosis module uses an adaptive weighted fusion algorithm to perform redundant backup data from multiple sensors of the same type deployed at the same monitoring point for key monitoring parameters. The weights of the adaptive weighted fusion algorithm are dynamically allocated according to the historical error statistics, real-time signal-to-noise ratio and health status indicators of each sensor to output the optimal estimate of the system status. The key monitoring parameters include ship position, attitude, hinge point angle, pipe tension and stress. Meanwhile, the sensor signal fusion and fault diagnosis module performs real-time fault diagnosis based on the chi-square test method, calculates the residual sequence between the output value of each sensor and the predicted value of the system state, constructs the residual covariance matrix and calculates its chi-square statistic; when the chi-square statistic exceeds the preset threshold, it is determined that the corresponding sensor has failed. Once a fault is diagnosed, the data weight of the faulty sensor is immediately reduced to zero, and the data stream of the backup sensor at the monitoring point is seamlessly switched. At the same time, the fault identifier and timestamp are recorded and reported to the system log.

5. The method for coordinated control of pipelaying vessel dynamic positioning and pipe-laying support frame according to claim 1, characterized in that, It also includes an independent and parallel emergency avoidance subsystem. This subsystem has built-in multi-level safety thresholds and continuously monitors the equivalent stress of key pipeline nodes, the rate of change of tensioner tension, and the thrust status of each thruster in real time. When any monitored parameter exceeds its corresponding safety threshold, the emergency avoidance subsystem immediately issues the highest priority interruption, overriding the current output command of the central controller, forcing the system to switch to the preset emergency avoidance control mode. It instructs the support frame adjustment system to perform an upward movement at the maximum safe angular velocity to avoid wave loads. At the same time, it instructs the dynamic positioning system to control the tensioner to relax the pipeline tension at a given rate and calls the robust control algorithm of the lower tracker to switch the ship control mode from trajectory tracking to fixed position holding mode. The system control is returned to the central controller only after all parameters have returned to normal and become stable.

6. The method for coordinated control of dynamic positioning and pipelaying support frame of pipelaying vessel according to claim 1, characterized in that, In the online rolling solution process of step S3, the saturation constraints of the action rate and torque of each hydraulic actuator of the support frame adjustment system and the saturation constraints of the thrust of each propeller of the dynamic positioning system are directly embedded into the solution framework of the sequential quadratic programming algorithm as inequality constraints.

7. The method for coordinated control of pipelaying vessel dynamic positioning and pipe-laying support frame according to claim 1, characterized in that, In step S3, a model prediction error compensation mechanism is introduced in the upper-level optimizer. This mechanism is based on the deviation between the output of the state observer in the lower-level tracker and the actual system state. It constructs a model error estimator to perform online correction on the unified nonlinear dynamic model, thereby reducing the accumulation of prediction errors caused by model mismatch.

8. The method for coordinated control of dynamic positioning and support frame of pipelaying vessel according to claim 4, characterized in that, The sensor signal fusion and fault diagnosis module is also equipped with a virtual sensor reconstruction unit. When all physical sensors at a certain monitoring point are diagnosed as faulty, the virtual sensor reconstruction unit reconstructs the state estimate of the monitoring point based on a unified nonlinear dynamic model and the data of the remaining effective sensors through a Kalman filter algorithm, and temporarily replaces the output of the faulty sensor until the physical sensor recovers or manual intervention is required.

9. The method for coordinated control of dynamic positioning and pipelaying support frame of pipelaying vessel according to claim 5, characterized in that, The emergency escape subsystem also includes a smooth switching mechanism for a smooth transition when the system switches from emergency escape mode back to central controller control; the smooth switching mechanism specifically includes: After the conditions for exiting emergency avoidance mode are met, the system does not switch immediately, but instead initiates a preset transition time window. During this transition time window, the upper-level optimizer and lower-level tracker of the central controller begin to run synchronously, but the control commands they output are not executed immediately. Instead, they are compared and synchronized with the current commands of the emergency avoidance subsystem. A linear interpolation algorithm is used as the transition function to generate a time-varying mixing coefficient from 0 to 1; During the transition period, the final composite control command issued to the actuator is a weighted synthesis of the emergency avoidance command and the central controller command based on the time-varying mixing coefficient, which starts from 0 and gradually increases to 1 over time. When the time-varying mixing coefficient increases to 1, the composite control command is entirely composed of outputs from the central controller. Once the control is transferred, the emergency avoidance subsystem immediately switches to a dormant monitoring state.

10. The method for coordinated control of dynamic positioning and support frame of pipelaying vessel according to claim 6, characterized in that, In the solution framework of the sequential quadratic programming algorithm, different priority weights and slack variable penalty coefficients are configured for different types of inequality constraints; Among them, the thrust saturation constraint of each thruster of the dynamic positioning system and the tension safety constraint of the pipeline tensioner are set as the first priority and their corresponding first penalty coefficients; the action rate constraint of the support frame adjustment system is set as the second priority, which is lower than the first priority and its corresponding second penalty coefficient, and the second penalty coefficient is less than the first penalty coefficient.

Images