Real-time monitoring control system for hinge gasket construction based on multi-sensor fusion

Abstract

This application relates to the field of automation technology in marine engineering construction, and discloses a real-time monitoring and control system for hinged raft construction based on multi-sensor fusion. Through a hydrodynamic field inversion module, strain measurement data and a parameterized forward dynamic model are used to invert the near-field hydrodynamic environment acting on the hinged raft. A state fusion estimation module, within an extended Kalman filter framework, fuses sparse absolute position measurement data and high-density morphological measurement data, and introduces morphological manifold projection correction to accurately estimate the raft's position, attitude, and morphology. A predictive control decision module uses the estimated system state and the inverted near-field hydrodynamic environment as inputs, and solves an optimization problem to generate the optimal control sequence under multiple constraints considering structural safety and the physical limitations of the actuators. This invention achieves active suppression of hydrodynamic disturbances, improves the robustness and accuracy of state estimation, and significantly enhances the automation level and efficiency of hinged raft construction.

Description

Technical Field

[0001] This invention relates to the field of automation technology in marine engineering construction, specifically to a real-time monitoring and control system for hinged submerged raft construction based on multi-sensor fusion. Background Technology

[0002] Hinged submersible rafts, as an important underwater revetment, bottom protection, and pipeline protection structure, are widely used in marine and hydraulic engineering projects. Their construction typically involves lowering the raft from a laying vessel using winches or other actuators, gradually and precisely laying it to a predetermined underwater position. Due to the inherent flexibility and high slenderness ratio of the hinged submersible raft, it undergoes complex bending and torsional deformations during the lowering process.

[0003] Existing methods for monitoring the construction of hinged submerged rafts typically rely on global navigation satellite systems to obtain the position of the laying vessel and underwater acoustic positioning systems to measure sparse positional information of a few key points on the raft. However, such sparse measurement data cannot comprehensively and in real time reflect the complete bending and torsional morphology of the hinged submerged raft as a continuous underwater structure.

[0004] Meanwhile, the hinged submerged raft is subjected to complex and dynamically changing water flow underwater. This hydrodynamic disturbance, which cannot be directly observed, is a major factor affecting construction accuracy. Existing control methods are mostly passive responses, only able to provide feedback correction after the water flow disturbance has caused the submerged raft's position and attitude to deviate. This control process has a significant lag, affecting laying accuracy.

[0005] Due to a lack of precise perception of the integrity of the submerged structure and external environmental disturbances, the current construction process relies heavily on the experience of the operators. Operators make manual adjustments based on limited monitoring data and changes in winch load. This subjective judgment-based operating mode not only makes it difficult to guarantee the accuracy of the construction trajectory but also fails to systematically avoid safety risks such as excessive structural bending or overloading of actuators due to improper control, posing challenges to construction efficiency and project quality. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a real-time monitoring and control system for hinged submerged raft construction based on multi-sensor fusion. This system solves the problems of existing hinged submerged raft construction monitoring and control methods, which are unable to cope with complex and ever-changing underwater dynamic environments, lack sufficient accuracy in real-time state perception of the flexible and highly deformable submerged raft structure, and rely on human experience in the control process. Furthermore, they lack systematic guarantees for structural safety and the physical limitations of the actuators, resulting in low construction efficiency and high safety risks.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solution: a real-time monitoring and control system for hinged submerged raft construction based on multi-sensor fusion, comprising: a data acquisition and preprocessing module, a hydrodynamic field inversion module, a state fusion estimation module, a predictive control decision module, and an execution mechanism.

[0008] The hydrodynamic field inversion module is used to obtain the near-field hydrodynamic environment, which is not directly measurable and constitutes the main disturbance to the control system. This module establishes a near-field hydrodynamic environment parameter... A forward dynamic model parameterized for hinged sinkers with unknown variables.

[0009] This module further defines the cost function. The cost function characterizes the error between the strain predicted by the positive dynamic model under given hydrodynamic parameters and the strain measurement data provided by the data acquisition and preprocessing module. In order to efficiently solve for the optimal hydrodynamic parameters that minimize the cost function, this module adopts a gradient-based optimization algorithm.

[0010] In one embodiment, the module employs the adjoint method to efficiently compute the gradient of the cost function with respect to unknown hydrodynamic parameters. This significantly improves the efficiency of inversion calculations while ensuring accuracy.

[0011] The state fusion estimation module is used to provide accurate, complete and physically consistent state information of the hinge sinker in the global coordinate system.

[0012] In one embodiment, the module will convert the system state vector of the hinge sinker. Defined as:

[0013] ;

[0014] in, and These are the global position and attitude quaternions of the hinge pad at the integration starting point, respectively. A vector of morphological parameters describing its bending and torsional forms. and These are the linear velocity and angular velocity at the integration starting point, respectively. Within an extended Kalman filter framework, this module fuses sparse but accurate absolute position measurement data with high-density, continuous morphological measurement data converted from strain measurement data by periodically performing prediction and update steps.

[0015] To address the issue of estimation results deviating from physical reality due to measurement noise or model uncertainty, this module introduces an additional manifold projection correction step after the update step of the extended Kalman filter. This step utilizes a morphological manifold learned in advance from historical data or high-fidelity simulation data to construct a projection operator on the morphological manifold. For any original morphological parameter vector The projection result is as follows:

[0016] ;

[0017] in It is the morphological mean vector. It is the principal component basis matrix containing the dominant variant mode.

[0018] This projection correction step ensures that the final output state estimate always lies within a physically reasonable shape space, significantly enhancing the robustness and reliability of the estimate.

[0019] The predictive control decision module is used to generate an optimal control sequence that can accurately track the target trajectory while ensuring the safety of the construction process. This module adopts a model predictive control method, and in each control cycle, it solves an optimal control sequence online within a finite prediction time domain. An internally defined optimization problem, the objective function of which is... The tracking error between the future output and the reference trajectory, as well as the stability of the control input, are taken into account.

[0020] To proactively address water flow disturbances, this module integrates a hydrodynamic disturbance feedforward model into the prediction model. This model uses the near-field hydrodynamic environment output by the hydrodynamic field inversion module as a known time-varying disturbance term and performs pre-compensation in the control decision.

[0021] In addition, the optimization problem is subject to multiple constraints, including: structural safety constraints to ensure structural safety, such as the maximum curvature or maximum strain limit of the hinge sinker; and physical constraints of the actuators to ensure system feasibility, such as the control amplitude limit and control change rate limit of the winch tension or release speed.

[0022] This invention provides a real-time monitoring and control system for hinged submerged embankment construction based on multi-sensor fusion. It has the following beneficial effects:

[0023] 1. This invention constructs a hydrodynamic field inversion module, which uses easily measurable structural strain data to invert the near-field hydrodynamic environment that constitutes the main disturbance to the control in real time, and integrates it as a feedforward term into the predictive control decision module. This enables the control system to know in advance about the upcoming water flow disturbance and compensate in advance, realizing the transformation from passive feedback correction to active disturbance suppression. Thus, in complex and time-varying underwater environments, it significantly improves the accuracy of hinged raft attitude control and the tracking stability of target trajectory.

[0024] 2. In the state fusion estimation module, this invention not only integrates sparse absolute position measurement data with high-density morphological data, but also innovatively introduces a data-driven morphological manifold projection correction step. This step projects the filtered and updated morphological estimate into a pre-learned effective morphological space determined by physical laws, ensuring that the system's estimation of its own state always conforms to physical reality. This effectively suppresses the estimation divergence problem caused by measurement noise or model inaccuracy, providing a highly reliable and robust state input for subsequent precise control.

[0025] 3. The model predictive control method adopted in this invention incorporates structural safety constraints and physical limitations of the actuator as hard constraints into the online optimization problem of each control cycle. By solving the optimization problem under these multiple constraints, the control commands generated by the system seek optimal tracking performance while ensuring that they do not exceed the safety boundaries of the structure and equipment. This systematically ensures the safety of the entire construction process, reduces the reliance on the experience of operators, and avoids engineering accidents caused by overload or excessive deformation. Attached Figure Description

[0026] Figure 1 This is a system architecture diagram of the present invention;

[0027] Figure 2 This is a flowchart of the method of the present invention.

[0028] The module includes: 10. Data acquisition and preprocessing module; 20. Continuous morphology reconstruction module; 30. Hydrodynamic field inversion module; 40. State fusion estimation module; 50. Predictive control decision module; and 60. Execution instruction generation module. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Example:

[0031] Please see the appendix Figure 1 This invention provides a real-time monitoring and control system for hinged submerged embankment construction based on multi-sensor fusion, comprising:

[0032] The system includes a data acquisition and preprocessing module 10, a continuous morphology reconstruction module 20, a hydrodynamic field inversion module 30, a state fusion estimation module 40, a predictive control decision module 50, and an execution instruction generation module 60.

[0033] The data acquisition and preprocessing module 10 connects to the aforementioned hardware sensing devices to acquire multi-source heterogeneous data in real time. Specifically, this module acquires raw signals from the distributed fiber optic network to demodulate time-varying data, acquires GNSS-RTK and USBL positioning data, and acquires ADCP velocity profile data. This module also performs preprocessing operations such as timestamp synchronization, noise filtering, and coordinate system registration on the acquired raw data to form a standardized data stream for use by other modules in the system.

[0034] The continuous morphology reconstruction module 20 takes strain data stream provided by the data acquisition and preprocessing module 10 as its input. Based on a preset strain-curvature mapping model, this module converts the one-dimensional strain distribution into spatial curvature vectors of the submerged structure at each discrete point. Subsequently, by integrating the curvature vectors along the arc length path of the submerged structure, the continuous geometric morphology of the hinged submerged structure in three-dimensional space is solved. This morphology is in the relative coordinate system, and its absolute position in the global coordinate system is undetermined.

[0035] The hydrodynamic field inversion module 30 takes as input the strain data stream provided by the data acquisition and preprocessing module 10. The module has a forward model of the structural dynamics of the hinged sump. It uses an adjoint optimization algorithm to solve the near-field hydrodynamic distribution around the sump structure that causes the strain result, with the measured strain distribution as the target constraint. The output of the module is a dynamic vector field describing the velocity and pressure of the water near the sump.

[0036] The state fusion estimation module 40 receives data input from multiple modules, including: sparse absolute positioning point coordinates provided by the data acquisition and preprocessing module 10, the relative geometric shape of the raft provided by the continuous morphology reconstruction module 20, and near-field hydrodynamic field information provided by the hydrodynamic field inversion module 30. This module runs a manifold Kalman filter algorithm under physical constraints, using high-frequency, high-resolution relative morphology as a strong constraint, and fuses low-frequency absolute positioning data to obtain the optimal state estimate of the hinged raft in the global coordinate system, including its precise position, shape, and velocity.

[0037] The predictive control decision module 50 takes as input the current optimal state of the system output by the state fusion estimation module 40 and the near-field hydrodynamic field output by the hydrodynamic field inversion module 30. Based on the model predictive control framework, this module uses the system dynamics model and the predicted hydrodynamic disturbances to solve an optimal control problem with the objectives of construction accuracy and structural safety online. Its output is the optimal control sequence within a short time window in the future.

[0038] The execution instruction generation module 60 receives the optimal control sequence output by the predictive control decision module 50. This module extracts the first control instruction from the sequence and converts it into a specific instruction that can be directly executed by the physical actuator, such as the winding and unwinding speed of each cable or the target tension.

[0039] See attached document Figure 2 This invention provides a real-time monitoring and control method for hinged submerged embankment construction based on multi-sensor fusion, comprising the following steps:

[0040] S100 performs multi-source heterogeneous data acquisition and preprocessing. Through distributed optical fiber sensor network, GNSS / USBL positioning equipment and ADCP and other sensors, it synchronously acquires the internal strain distribution of the hinged sinker, the external sparse absolute position and the surrounding far-field water flow profile data, and performs preprocessing operations such as filtering, time synchronization and coordinate system integration on the data.

[0041] S200. Real-time reconstruction of the continuous morphology of the sinking purlin is performed. Based on the strain data processed in step S100, the continuous three-dimensional geometric morphology of the hinged sinking purlin in the relative coordinate system is solved by the strain curvature mapping model and the space curve integral algorithm.

[0042] S300. Perform dynamic inversion of the near-field hydrodynamic environment. Based on the strain data processed in step S100, use the structural dynamics model of the hinged submerged raft and solve the near-field hydrodynamic vector field acting on the surface of the submerged raft structure in reverse using the adjoint optimization method.

[0043] S400. Perform multi-source information fusion and optimal system state estimation. Establish a Kalman filter with fiber reconstruction morphology as manifold constraint. Fuse the sparse absolute position obtained in step S100, the relative morphology reconstructed in step S200, and the near-field hydrodynamic information inverted in step S300 to calculate the optimal state estimate of the hinged submerged raft in the global coordinate system. The state includes position, attitude, morphology, and motion velocity.

[0044] S500. Solve the optimal control sequence based on disturbance feedforward. Take the optimal state obtained in step S400 as the initial condition and the hydrodynamic field inverted in step S300 as the feedforward information of future disturbances. Within a model predictive control framework, solve a finite-time optimal control problem that satisfies multiple constraints such as construction accuracy, structural safety and actuator capability, and obtain a set of optimal control command sequences.

[0045] S600: Control commands are issued and the system undergoes rolling optimization. The first command of the optimal control command sequence obtained in step S500 is extracted and issued to the winch and other actuators of the construction mother ship. After completing this control cycle, the system returns to step S100 to begin the calculation and control for the next cycle, thus achieving rolling optimization.

[0046] The data acquisition and preprocessing module 10 acquires and corrects distributed strain data, specifically including the following steps:

[0047] The data acquisition and preprocessing module 10 acquires raw measurement data via a fiber optic demodulator connected to a distributed fiber optic sensor network. In one embodiment, the sensor network employs a technique based on Brillouin optical time-domain reflectometry or Brillouin optical time-domain analysis. The fiber optic demodulator sends probe light pulses along the fiber optic path and receives Brillouin scattered light backscattered at various points along the fiber. This module records the center frequency offset of the scattered light relative to the incident light, forming the position along the fiber arc length. At any moment The original Brillouin frequency shift data.

[0048] A preliminary conversion from frequency shift to physical quantity is performed, and the Brillouin frequency shift is collected. Simultaneously affected by strain at the location of the optical fiber and temperature changes The combined effects of these factors can be described by the following formula:

[0049] ;

[0050] in:

[0051] To be at the fiber arc length position ,time Measured Brillouin frequency shift;

[0052] This represents the axial strain experienced by the optical fiber at that point.

[0053] This represents the temperature change of the optical fiber relative to a reference temperature.

[0054] The strain sensing coefficient of the optical fiber;

[0055] This represents the temperature sensing coefficient of the optical fiber.

[0056] Strain sensing coefficient With temperature sensing coefficient It is an inherent physical property of the optical fiber used, and its value can be obtained through prior calibration experiments.

[0057] To correct for the temperature-strain cross-sensitivity effect and accurately separate the mechanical strain caused by structural deformation from the mixed frequency shift signal, one embodiment of the present invention employs a temperature compensation mechanism. The specific implementation of this mechanism can be as follows:

[0058] When deploying sensing fibers on a hinged, slumped structure, one or more additional, relaxed temperature-sensing fibers, unaffected by structural strain, are deployed in parallel along its path. The Brillouin frequency shift measured by this temperature-sensing fiber is... This will be caused solely by temperature changes:

[0059] ;

[0060] The data acquisition and preprocessing module 10 can use the above formula to calculate the temperature change distribution along the line. Subsequently, the frequency shift component caused by temperature is subtracted from the total frequency shift measured by the main sensing fiber to obtain the frequency shift component caused purely by strain. :

[0061] ;

[0062] Ultimately, high-precision strain distribution data with temperature correction can be obtained. :

[0063] ;

[0064] In another embodiment, the correction of the temperature-strain cross-sensitivity effect can also be achieved by using a special optical fiber that is insensitive to temperature, or by simultaneously measuring multiple parameters of the Brillouin scattered light to decouple temperature and strain at the algorithm level.

[0065] Spatial registration and signal filtering are performed to convert the one-dimensional fiber arc length coordinates... Mapped to the three-dimensional physical space of the submerged structure, the precise positions of several feature points along the line in the coordinate system of the submerged structure are recorded during the fiber optic deployment stage, thereby establishing a mapping relationship or lookup table of arc length spatial positions. The data acquisition and preprocessing module 10 uses this mapping relationship to assign each strain data point its three-dimensional spatial coordinates on the structure.

[0066] To improve the signal-to-noise ratio, digital filtering can be applied to the corrected time-varying data sequence. For example, Savitzky-Golay filters or one-dimensional median filters can be used along the spatial or temporal dimensions to smooth the signal and suppress random noise. Such signal processing methods are well-known in the field and will not be elaborated further here.

[0067] After the above steps, the data acquisition and preprocessing module 10 finally outputs a high-precision, temperature-corrected, spatially registered, and spatiotemporally continuous dynamic strain field data. This is for use by the continuous morphology reconstruction module 20 and the hydrodynamic field inversion module 30.

[0068] The data acquisition and preprocessing module 10 acquires and transforms coordinates of sparse absolute positioning data, specifically including the following steps:

[0069] This module acquires heterogeneous positioning data by receiving and parsing data from different types of positioning sensors through a communication interface.

[0070] To establish a unified global reference coordinate system for integrating heterogeneous positioning data, the system first defines a unified global reference coordinate system that does not change over time. In one embodiment, this coordinate system can be selected as a local Cartesian coordinate system of the construction area, such as the Northeast-Sky coordinate system. The origin of this coordinate system is usually set on a known, precisely mapped benchmark point within the construction area. All subsequent positioning data must be converted to this unified coordinate system.

[0071] Coordinate transformation of GNSS-RTK data is performed. For position data in geodetic coordinate systems such as WGS-84 output by the GNSS-RTK receiver, the data acquisition and preprocessing module 10 applies a standard coordinate transformation algorithm to convert it into three-dimensional Cartesian coordinates in a unified global reference coordinate system. This transformation process involves information such as ellipsoid parameters and reference point coordinates. The specific algorithm is well-known in the field and will not be described in detail here.

[0072] The data acquisition and preprocessing module 10 acquires all the input quantities required by the above formula in real time and performs the solution, thereby converting the relative positioning information of USBL into global absolute coordinates.

[0073] Output a standardized sparse absolute position point set. After the coordinate transformation steps described above, all positioning points from GNSS-RTK and USBL are unified to the same global reference coordinate system. The data acquisition and preprocessing module 10 organizes these timestamped 3D coordinate points into a sparse absolute position point set. This point set will serve as a key measurement input for the state fusion estimation module 40, used to globally locate and orient the relative shape output by the continuous shape reconstruction module 20.

[0074] To ensure the accuracy of subsequent fusion and control calculations, the data acquisition and preprocessing module 10 implements a timestamp alignment and synchronization mechanism for multi-source data streams to address issues such as inconsistent data generation times, different transmission delays, and varying sampling frequencies among the sensors. This mechanism specifically includes the following steps:

[0075] A unified system time base is established, with the central data processing and control unit within the system establishing a high-precision system master clock. In one embodiment, this master clock is synchronized with an external high-precision time source via a network time protocol or a precision time protocol. This system master clock serves as the sole time base for timestamp alignment throughout the entire system.

[0076] Clock synchronization is performed on all data acquisition terminals. For all sensor terminals with network capabilities or their data acquisition computers, they are configured as clients of the time synchronization protocol to periodically synchronize with the system master clock established above. In this way, the local timestamps attached by each sensor when generating data can maintain consistency with the system master clock at the sub-millisecond level or higher, thereby ensuring the accuracy of the timestamps from the source.

[0077] To implement data arrival timestamp compensation, for some sensors that do not directly support time synchronization protocols, the data acquisition and preprocessing module 10 immediately appends an arrival timestamp generated by the system master clock to the data packet upon its arrival at the central data processing and control unit. Simultaneously, the system can pre-calibrate or estimate online the average delay of data transmission from generation to the central unit for this type of sensor. When processing the data, this delay is subtracted from the arrival timestamp to obtain a valid timestamp that is closer to the actual time the data was generated.

[0078] Data interpolation and alignment based on a unified time step are performed. Since the data update frequencies of each sensor differ, the system needs to obtain the measurement values ​​of all sensors at a unified, discrete time point. The data acquisition and preprocessing module 10 sets a fixed data fusion and control cycle. And generate a series of unified timestamps. .

[0079] For any sensor data stream, if in If there is no direct measurement value at any given time, the module will look it up in its data cache. The two most recent measurement points before and after time are used to calculate the result using linear interpolation. The estimated value of the time.

[0080] In another embodiment, for rapidly changing physical quantities, higher-order interpolation methods, such as Lagrange interpolation or spline interpolation, can be used to obtain higher alignment accuracy. Through the above steps, the data acquisition and preprocessing module 10 ultimately outputs a set of time-aligned data snapshots, each containing the same unified timestamp. The measurements from all sensor sources provide high-quality, time-consistent, and data-complete input for subsequent state fusion and control modules.

[0081] Before the continuous morphological reconstruction module 20 performs morphological reconstruction calculations, it requires an established and accurate strain curvature mapping model. The function of this model is to convert the strain physical quantities output by the data acquisition and preprocessing module 10 and measured along the optical fiber into a linear model. This is transformed into a curvature vector describing the degree of geometric bending of the hinge pad at the corresponding position. The method for establishing this model may specifically include the following steps:

[0082] The basic form of the model is determined based on the plane section assumption in mechanics of materials. On a cross-section of a structure, the axial strain at any point is linearly related to its distance from the neutral axis of the section, and this relationship is determined by the curvature of the section. For a slender structure that can be freely bent and stretched in three-dimensional space, its axial strain at the arc length position... The deformation state can be described by a state vector, which includes the bending curvature about two orthogonal local coordinate axes and the axial strain along the neutral axis.

[0083] In one embodiment, at least three non-collinear sensing fibers are required to determine the three-dimensional deformation of a cross-section. Assume a cross-section... There are arranged Root sensing fiber ( If ≥3), then the strain measurements of this section can form a strain vector. .

[0084] The deformation state of this cross section can be represented by a curvature strain state vector. To describe, among which and Local coordinate systems around the cross section shaft and The curvature components of the axis, Let be the axial strain at the neutral axis of the cross section. There exists a linear mapping relationship between them:

[0085] ;

[0086] in: It is A transfer matrix of dimension, which is to be determined and describes the mapping relationship from the structural deformation state to fiber optic strain measurement.

[0087] This matrix depends only on the geometric arrangement of the optical fibers within the cross section and the material properties of the structure, and does not change over time.

[0088] The transfer matrix was determined through experimental calibration. Due to the composite material structure and complex cross-sectional shape of the hinged sinker, the transfer matrix is ​​difficult to obtain accurately through purely theoretical analysis. Therefore, one embodiment of the present invention employs an experimental calibration method to accurately determine the matrix.

[0089] By applying a loading system, a series of different, known deformation states are applied to the sample. When each known deformation is applied, the actual curvature and axial strain of the sample are precisely measured using independent high-precision measuring instruments as the true values. Simultaneously, using the same fiber optic demodulator as the field system, the strain measurement vector corresponding to each sensing fiber is recorded synchronously. .

[0090] To solve for the transfer matrix, after collecting N sets of data pairs, all the data can be organized into a matrix form:

[0091] ;

[0092] in: It is A dimensional strain measurement matrix.

[0093] It is The truth-valued deformable state matrix of dimension .

[0094] This equation is a typical linear system identification problem, with the transfer matrix... This can be obtained by solving a least squares problem, the solution of which is:

[0095] ;

[0096] In another embodiment, to improve numerical stability and noise resistance, singular value decomposition can be used to solve the pseudo-inverse of the above equations, resulting in a more robust solution. estimate.

[0097] The final strain curvature mapping model was established, and the transfer matrix was obtained through experimental calibration. Then, an inverse solution model from measured strain to structural deformation state can be established. When performing real-time calculations, the continuous morphology reconstruction module 20 uses the inverse operation form of this model, that is, obtains the curvature-strain state vector by solving the following formula:

[0098] ;

[0099] in:

[0100] At any moment ,Location The curvature strain state vector is estimated at the location.

[0101] It is a matrix The false rebellion.

[0102] It is a processed real-time strain measurement vector provided by the data acquisition and preprocessing module 10.

[0103] Curvature components in the model and They are then combined into a spatial curvature vector. This serves as the input for the next step, the morphological integration algorithm.

[0104] The continuous shape reconstruction module 20 obtains the curvature vector distributed along the arc length of the hinge sinker. Then, a spatial curve integral algorithm based on the moving frame method is used to convert the curvature distribution into three-dimensional spatial coordinates of the sinkhole's central axis. This algorithm specifically includes the following steps:

[0105] Establish a differential geometric model of the space curve. The central axis of the hinged sinker can be mathematically described as a space curve. In differential geometry, the local shape of a curve is uniquely determined by the rate of change of its tangent vector. A local coordinate system moving along the curve, i.e., a moving frame, has its orientation change directly related to the curvature of the curve. This relationship is described by a system of ordinary differential equations.

[0106] Solve for the tangent vector field. Specifically, the curve at the arc length position... unit tangent vector at point The rate of change and the curvature vector at that point The following relationship exists:

[0107] ;

[0108] in:

[0109] At any moment Arc length position The unit tangent vector at that point indicates the direction of the curve at that point;

[0110] It is the curvature vector obtained from the strain curvature mapping model;

[0111] This represents the cross product operation of vectors.

[0112] Because distributed fiber optic sensor networks provide information along the arc length The curvature values ​​of discrete samples require numerical methods to solve the above ordinary differential equations.

[0113] In one embodiment, a fourth-order Runge-Kutta method can be used for high-precision numerical integration from a known initial tangent vector. Initially, the algorithm proceeds along the arc length direction, starting from the previous discrete point. tangent vector Iteratively calculate the next discrete point tangent vector This iterative process can continue until the entire length of the hinge sinker is covered, thereby obtaining the tangent vector field at all discrete sampling points. For the implementation of the numerical integration algorithm, those skilled in the art can choose as needed, which is a well-known technique in the field and will not be described in detail here.

[0114] In another embodiment, to obtain higher reconstruction accuracy, higher-order numerical integration methods such as the trapezoidal rule or Simpson's rule can be used.

[0115] To determine the relative form of the output, both integration processes described above require a starting point for integration. The initial conditions, i.e., the initial position. and initial tangent vector Since these initial conditions are unknown in the global reference coordinate system when only internal strain data are available, the continuous morphology reconstruction module 20 can set an arbitrary initial condition in this step.

[0116] The set of spatial coordinate points calculated based on these arbitrary initial conditions constitutes the relative geometric shape of the hinged sinker. This shape accurately reflects the actual bending and stretching shape of the sinker, but its absolute position and orientation in global space are undetermined. This module ultimately outputs point cloud data of this relative shape for the state fusion estimation module 40 to perform subsequent global localization and orientation.

[0117] To address the issue that the morphological result is only a relative morphology due to unknown initial integration conditions, the continuous morphological reconstruction module 20 further performs an initial condition determination and correction calculation. This calculation utilizes the sparse absolute positioning data provided by the data acquisition and preprocessing module 10 to assign the correct position of the relative morphology in the global reference coordinate system. This calculation specifically includes the following steps:

[0118] By establishing a correlation between relative shape and absolute positioning points, the system determines the precise arc length position of each sparse absolute positioning measurement point on the hinged sinker structure based on preset equipment installation information. At the moment of gain After obtaining the relative shape point set, the module extracts the arc length position relative to each absolute positioning point. The corresponding three-dimensional coordinate points are denoted as .

[0119] This results in a set of data pairs. ,in It is the global absolute coordinate measurement value corresponding to that point, output from the above. It is the total number of sparse absolute positioning points.

[0120] An optimization problem involving rigid body transformation is constructed. Since the relative and absolute shapes differ by only one rigid body transformation, this problem can be modeled as solving for an optimal rotation matrix. And an optimal translation vector The goal of this transformation is to minimize the error between the transformed relative shape point and its actual measured absolute coordinate point.

[0121] Solving for the optimal rigid body transformation parameters is a typical absolute orientation problem with a closed-form solution. In one embodiment, this problem can be solved using a singular value decomposition (SVD) method. The specific calculation process of this method is well-known in the art and will not be elaborated here. By solving, the optimal rotation matrix at that moment can be obtained. Translation vector .

[0122] Determine the initial conditions for global integration and correct the shape. After obtaining the optimal transformation parameters, the equivalent initial conditions for the shape integration algorithm in the global reference coordinate system can be determined.

[0123] Furthermore, the continuous morphology reconstruction module 20 applies the obtained optimal transformation to the entire relative morphology point set, thereby obtaining a complete three-dimensional morphology snapshot of the hinge sinker at the current moment in the global reference coordinate system.

[0124] The filtering algorithm in the state fusion estimation module 40 will further smooth and correct these instantaneous solutions in the time dimension, realize iterative optimization in the dynamic process, and thus obtain the final, continuous and robust state estimation results.

[0125] To achieve effective inversion of the near-field hydrodynamic environment, the hydrodynamic field inversion module 30 first needs to establish a forward dynamic model that can accurately describe how the hinged raft moves and deforms under the action of external forces.

[0126] This model is a parameterized mathematical model whose response is not only a function of its own state but also a function of the unknown environmental parameters to be inverted. The process of establishing this model may specifically include the following steps:

[0127] The structure is discretized by spatially representing the continuous hinged sinker structure as a set of finite elements connected by nodes. In one embodiment, considering the slender beam characteristics of the hinged sinker and its large displacement and rotation during construction, beam elements in the absolute nodal coordinate method can be used for discretization. The ANCF method, by using the global position vector and its gradient as nodal coordinates, can accurately describe rigid body motion and large deformation problems, and its mass matrix is ​​a constant matrix, which is beneficial for improving computational efficiency.

[0128] After constructing the system's governing equations and discretizing them, the dynamic behavior of the entire hinged submerged system can be described by a set of second-order ordinary differential equations. The matrix form of its standard Lagrange dynamic equations is as follows:

[0129] ;

[0130] in:

[0131] It is the system's generalized coordinate column vector, containing the position vectors and gradients of all nodes, with dimensions of... , that is, the total number of degrees of freedom of the system;

[0132] and These are the first and second time derivatives of the generalized coordinates, namely the generalized velocity and the generalized acceleration;

[0133] It is the constant mass matrix of the system, which is obtained by performing a standard finite element assembly process on the mass matrices of each element;

[0134] It is the generalized internal force column vector of the system, generated by the elastic deformation of the structure and material damping, and is a generalized coordinate. Nonlinear functions;

[0135] It is the generalized external force column vector of the system, which contains all external forces applied to the structure.

[0136] Determine the specific expressions for each force term.

[0137] Broad internal strength This is primarily the elastic restoring force of the structure. For ANCF beam elements, this force term is obtained by differentiating the element's strain energy with respect to the generalized coordinates. The derivation of its specific expression is well-known in ANCF theory and will not be elaborated here.

[0138] In one embodiment, the hydrodynamic forces acting on the micro-elements of the submerged raft structure can be calculated using the Morison equation, which decomposes the hydrodynamic forces into drag and inertial forces. For any node on the submerged raft, the generalized hydrodynamic vector is obtained by converting the above-mentioned node hydrodynamic expressions into generalized forces and assembling them along the entire structure. Become an explicit dependence on the unknown near-field water flow field With this function, a complete and parameterized forward dynamics model has been established.

[0139] This model can be applied to a given set of near-field water flow fields. Under the given conditions, the dynamic equations above are solved by numerical integration, and the dynamic response of the hinged sinker, including its displacement, velocity, and strain, is predicted in a positive direction.

[0140] After efficiently calculating the gradient of the cost function with respect to unknown hydrodynamic parameters, the hydrodynamic field inversion module 30 applies a gradient-based numerical optimization algorithm to iteratively adjust the hydrodynamic field parameters until the difference between the strain predicted by the forward dynamic model and the measured strain is minimized. This process may specifically include the following steps:

[0141] The optimization problem and initialization are defined, and the hydrodynamic field inversion problem is formalized into an unconstrained or bounded optimization problem, namely, finding the optimal hydrodynamic parameters. Make the cost function Minimize. Before starting the iteration, set an initial guess for the unknown hydrodynamic parameters. This initial value can be obtained by interpolation based on far-field ADCP measurement data, or set to zero flow field when there is no prior information.

[0142] The algorithm executes an iterative optimization loop, starting from the initial guess. In the k-th iteration, the following operations are performed:

[0143] Calculate the gradient at the current point and the cost function at the current parameter point. gradient at .

[0144] Determine the descent search direction. Based on the current gradient and historical iteration information, determine an effective descent search direction. In one embodiment, a limited-memory algorithm is employed. This algorithm is a quasi-Newton method, which constructs an approximation of the inverse of the Hessian matrix by storing information from the most recent iterations, thereby obtaining a search direction superior to the steepest descent direction. Search Direction The calculation formula is:

[0145] ;

[0146] in It is in the The approximation of the inverse of the Hessian matrix recursively constructed by the L-BFGS algorithm in the next iteration. The specific implementation of the L-BFGS algorithm is a well-known technique in the field of optimization and will not be described in detail here.

[0147] In another embodiment, the conjugate gradient method or the simpler steepest descent method can also be used to determine the search direction.

[0148] Execution line search determines the optimal step size along the search direction. To find an optimal step size This allows the cost function value to decrease sufficiently. This is achieved by solving a one-dimensional optimization problem:

[0149] ;

[0150] This one-dimensional optimization problem is usually solved using an inaccurate line search method.

[0151] Update hydrodynamic parameters using the calculated optimal step size. Update the hydrodynamic parameters:

[0152] ;

[0153] Perform a convergence check. After each iteration, check if the preset convergence criteria are met. Convergence criteria can include one or more of the following: gradient norm less than a threshold, parameter update amount less than a threshold, parameter update amount less than a threshold, or the decrease in cost function value less than a threshold. If the convergence criteria are met, the iteration terminates. Output the inversion results; the final parameter values ​​output after the optimization algorithm converges. This is the optimal near-field hydrodynamic field obtained within the current time window. This result will be provided as dynamically updated environmental information to the state fusion estimation module 40 and the predictive control decision module 50. The entire inversion process is executed periodically within a rolling short time window to achieve real-time tracking of the dynamically changing hydrodynamic environment.

[0154] To achieve effective fusion of multi-source information, the state fusion estimation module 40 employs a filtering-based estimation method. The first step of this method is to establish a state-space model that comprehensively describes the dynamic characteristics of the hinged sinking system and construct measurement equations that correlate this state with the actual measurements from each sensor. This process may specifically include the following steps:

[0155] Define a system state vector to comprehensively describe the position, attitude, shape, and velocity of the hinged submersible in the global reference coordinate system. Defined as a finite-dimensional vector containing all key degrees of freedom, in one embodiment, this state vector can be constructed as:

[0156] ;

[0157] in:

[0158] It is The position vector represents the hinge sinker at the starting point of integration. The global three-dimensional coordinates at the location.

[0159] It is The unit quaternion represents the orientation of the hinge sinker at the starting point of integration. Using quaternions can avoid the singularity problem.

[0160] It is The morphological parameter vector, since the continuous shape of the hinge sinker is an infinite-dimensional object, is described in a finite-dimensional state space, along the arc length. Curvature vector of distribution It is parameterized as a linear combination of a set of basis functions and their corresponding coefficients.

[0161] For example, ,in It is a pre-selected basis function matrix. This refers to the weight coefficient vector of these basis functions.

[0162] This vector fully describes the bending and twisting morphology of the sinker.

[0163] It is The velocity vector represents the linear velocity at the starting point of the integration.

[0164] It is The angular velocity vector represents the angular velocity at the starting point of the integration.

[0165] Establish nonlinear measurement equations. The function of measurement equations is to abstract the system state vector. Compared with specific sensor measurements obtained from the physical world Connect them. This relationship is usually non-linear, and its general form is:

[0166] ;

[0167] in:

[0168] At any moment A combined vector of all collected measurements;

[0169] It is a nonlinear measurement function that depends on the current state vector. Predict the corresponding theoretical measurement value;

[0170] It is a measurement noise vector, usually assumed to be zero-mean Gaussian white noise.

[0171] Construct specific measurement functions This function will convert the state vector The components are mapped to two main types of measurements:

[0172] The measurement mapping of sparse absolute positions, for positions located at the arc length provided above. The first Absolute position measurement points Its corresponding measurement function It is a state vector The process of calculating the global coordinates of this point. This calculation process reuses the morphological reconstruction algorithm:

[0173] From the state vector Extract morphological parameters And reconstruct the curvature vector distributed along the line. From the state vector Extract the initial position and initial attitude quaternion Convert the quaternion into an initial tangent vector. ,by and As initial conditions, for the curvature vector By performing spatial curve integration, the global coordinates at any position along the line can be obtained. .

[0174] This measurement point The theoretical measured value is Since the space curve integral is a complex nonlinear operation, the measurement mapping is highly nonlinear.

[0175] In one embodiment of the invention, the measurement mapping of fiber reconstructed morphology, for the high-density, continuous morphology measurement results output above, is also integrated into the measurement equation. The morphology measurement results themselves can be parameterized as a set of morphology parameters. Therefore, this part of the measurement equation can be directly expressed as the relationship between the morphological parameter components in the state vector and the measured morphological parameters.

[0176] Through the above steps, the system state fusion estimation module 40 establishes a complete nonlinear state-space model that can describe the system state and correlate it with multi-source heterogeneous measurement data. To further improve the accuracy and robustness of the estimation, the state fusion estimation module 40 introduces a data-driven morphological manifold constraint. This constraint functions by treating the set of all physically possible and reasonable shapes of the hinged arrangement as strong prior knowledge, used to correct instantaneous estimation results that may be generated by the filtering algorithm and do not conform to physical laws. The mathematical expression and implementation of this constraint may specifically include the following steps:

[0177] To construct a dataset for manifold learning, in order to obtain a mathematical description of the morphological manifold, a training dataset that can represent its features is first needed.

[0178] A mathematical representation of the manifold is established using principal component analysis. The actual morphological changes of the hinged sinker occur due to the influence of morphological parameters. Zhang Cheng's high-dimensional space, but its main, energy-concentrated deformation modes usually only occupy a low-dimensional subspace of that space.

[0179] Principal component analysis was used to identify this low-dimensional subspace.

[0180] Select the dominant principal components, sort the eigenvalues ​​in descending order, and select the top ones. The largest eigenvalue and its corresponding eigenvector. The selection criterion is to make this... The proportion of variance explained by each principal component is sufficiently high relative to the total variance. Each feature vector constitutes a dimensional matrix Its column vectors span the shape of the manifold at the mean point. The tangent space nearby. Thus, any physically valid shape vector... All of them can be approximately represented as:

[0181] ;

[0182] in, It is A coordinate vector of dimension 1 represents the coordinates of the shape in a lower-dimensional subspace. This implements the projection operator on the shape manifold. The function of the projection operator is to project any shape parameter vector that may deviate from the manifold. Map back to the point on the manifold that is closest to it. .

[0183] The operator The specific implementation is as follows:

[0184] Center the vector to be projected: .

[0185] Project it onto a lower-dimensional subspace and reconstruct it back to the original space: .

[0186] Adding the mean vector, we obtain the final projection result: ;

[0187] The projection operator is applied to the filtering framework and is embedded as a separate correction step after the update phase of the state estimation algorithm.

[0188] Specifically, after each measurement update step of the Kalman filter is completed, the state fusion estimation module 40 obtains a posterior state estimate. .

[0189] Through the above steps, it is ensured that the morphological estimation of each frame output by the state fusion estimation module 40 is physically reasonable, which greatly suppresses the estimation divergence caused by measurement noise or model inaccuracy, and significantly enhances the stability and reliability of the entire state estimation system.

[0190] The predictive control decision module 50 uses model predictive control to calculate the optimal control command for a future period. The core of MPC is to solve an optimization problem, which is performed on a rolling basis in each control cycle. The mathematical paradigm of this MPC optimization problem may include the following steps:

[0191] Defining the prediction time domain and the control time domain, model predictive control first requires setting two key time windows: the prediction time domain and the control time domain. and control time domain .

[0192] Prediction Time Domain This indicates the number of future time steps predicted by the model.

[0193] Control Time Domain This represents the future time steps in which a decision is actually made. Typically, MPC only calculates the control input sequence in the control time domain, and after each rolling optimization, it only applies the first control input of the sequence to the system.

[0194] The objective function is to minimize the performance metric defined in the prediction time domain. This metric typically consists of two parts: a tracking error term and a control input penalty term.

[0195] For the construction process of the hinged sinker, the objective function can be defined as:

[0196] ;

[0197] in: It is the current moment;

[0198] At the current moment Future predicted by the model The system output vector at time t;

[0199] It is the future The target reference trajectory or target state at any given moment;

[0200] It is a positive definite weight matrix used to penalize the deviation between the predicted output and the reference trajectory, and to assign different importance to different output components;

[0201] At the current moment Calculated future Constantly control the changes in input;

[0202] It is a positive definite weight matrix used to penalize drastic changes in control input, aiming to smooth control actions and avoid system jitter and energy waste.

[0203] To ensure the safe and stable operation of the control system, physical constraints need to be imposed on the control inputs and system states.

[0204] Control input constraints: These constraints limit the maximum and minimum values ​​of the winch tension and its rate of change.

[0205] State constraints: These constraints may include the location range of the sinking dam during construction, the minimum radius of curvature, etc.

[0206] Final state constraint: To ensure that the predicted trajectory eventually converges to the target point, a final state constraint or terminal cost function is sometimes imposed.

[0207] In summary, the MPC optimization problem is a nonlinear programming problem that occurs within each control cycle and requires specialized numerical optimization algorithms to solve.

[0208] To improve the robustness and control accuracy of the model predictive control system to environmental disturbances, the predictive control decision module 50 integrates a hydrodynamic disturbance feedforward model. This model can utilize near-field hydrodynamic information provided by the hydrodynamic field inversion module 30 to pre-compensate for the impact of water flow on the hinged submersible before making control decisions. The establishment and integration of this feedforward model may specifically include the following steps:

[0209] Near-field hydrodynamic prediction is obtained from the near-field hydrodynamic field inversion module 30. This is the estimated value at the current moment. To support the optimization of MPC in the prediction time domain, future values ​​are needed. Hydrodynamic field prediction within a time step .

[0210] Using a short-term hydrodynamic forecasting model, in one embodiment, a simplified hydrodynamic forecasting model can be used to analyze the inverted results. Extrapolate to generate the future The prediction model is trained based on historical water flow data, which consists of a prediction sequence at several time steps.

[0211] Spatial and temporal interpolation is necessary because hydrodynamic field inversion is usually performed at discrete time and spatial points, while the MPC prediction model may require continuous or different resolution hydrodynamic inputs. Therefore, it is necessary to perform necessary spatial and temporal interpolation on the inverted and predicted hydrodynamic data to match the requirements of the MPC model.

[0212] To construct a hydrodynamic disturbance feedforward force model, in the MPC prediction model, it is necessary to incorporate the predicted hydrodynamic field. Converted into a generalized disturbance force acting on the hinge sinker This is the same hydrodynamic calculation method used in the established forward dynamics model. That is:

[0213] ;

[0214] in:

[0215] The function represents the Morison equation or other hydrodynamic model used in hydrodynamic calculations, which will predict the current structural state (location). and speed and the predicted near-field water flow velocity. As input, the output is the generalized hydrodynamic force acting on the structure; and The MPC prediction model predicts the future structural state and velocity without control input. It integrates hydrodynamic disturbance feedforward into the MPC prediction model. It describes the evolution of the system's state, including the effects of external forces.

[0216] When integrating feedforward perturbations, the forward dynamic model External force The general explicitly includes this feedforward disturbance force caused by the water flow. Therefore, the dynamic prediction equation in MPC can be written as:

[0217] ;

[0218] in:

[0219] This represents the state evolution of the system under the influence of control forces; This represents the contribution of hydrodynamic disturbance to the state evolution. In practice, this is usually not an independent additive term, but is directly integrated into the state equation of the MPC model. That is, the MPC model will consider the estimated external hydrodynamic disturbance when predicting the state.

[0220] Applications in optimization problem solving: In the process of solving MPC optimization problems, when considering future states... During the prediction process, MPC utilizes near-field hydrodynamic field predictions from the hydrodynamic field inversion module 30. This means that when planning future control actions, MPC already knows the potential disturbances caused by future water flow and can pre-calculate the necessary compensation control quantities.

[0221] In this way, the predictive control decision module 50 can proactively respond to changes in the hydrodynamic environment, enabling the winch control system to maintain the hinged tumbler on the target trajectory more efficiently, especially demonstrating superior performance in complex and ever-changing underwater environments.

[0222] To ensure the safety and feasibility of the hinged submerged dam construction process, the predictive control decision module 50 introduces multiple constraints into the model predictive control optimization problem. These constraints include structural safety limitations and physical limitations of the actuators, restricting the solution space of the optimization problem within an allowable range. The handling of these multiple constraints may specifically include the following steps:

[0223] Integrated structural safety constraints are designed to prevent excessive deformation or stress on the hinged sinker during construction, which could lead to structural damage or failure. These constraints are typically applied to the system's state variables in the form of inequalities.

[0224] The formalization of the MPC optimization problem involves integrating all the aforementioned constraints as inequalities or equality constraints into the defined MPC optimization problem. Solving the optimization problem, this constrained nonlinear programming problem, can be done using interior-point methods, sequential quadratic programming, or other efficient nonlinear optimization algorithms. These algorithms iteratively search for the control input sequence that minimizes the objective function while satisfying all constraints.

[0225] By integrating these multiple constraints, the predictive control decision module 50 can generate optimal control commands that not only enable the hinged sinker to accurately track the target trajectory, but also ensure structural safety and the feasibility of the control system. This significantly improves the automation level and safety of the entire construction system.

[0226] After fully constructing the MPC optimization problem with multiple constraints, the predictive control decision module 50 enters the solution and execution loop phase. The goal of this phase is to efficiently calculate the optimal control action within each control cycle and translate it into physical commands for the underlying actuators. This process may specifically include the following steps:

[0227] Solve the constrained nonlinear optimization problem at the current time. The module will estimate the latest state. Hydrodynamic disturbances in the future prediction time domain and target reference trajectory As input, a formalized nonlinear programming problem is solved.

[0228] In one embodiment, this solution process is performed using a numerical optimization solver, such as an algorithm employing sequential quadratic programming or the interior-point method. The current instruction is generated using a rolling time-domain strategy; the core principle of model predictive control is the rolling time-domain strategy. Although the optimal control sequence for the entire future control time domain is calculated... However, the predictive control decision module 50 only extracts and uses the first element of the sequence as the control instruction to be executed at the current moment:

[0229] ;

[0230] In the current control command After execution, the control command is converted into a physical execution signal, and the optimal control command is calculated. It is a mathematical vector. The instruction needs to be converted into a physical signal that the underlying hardware can recognize and execute. This conversion is done by the system's control interface module.

[0231] The process continues until the next moment and repeats. After the current control command is sent to the actuator, the system waits for the next control cycle. At the new moment, the data acquisition and preprocessing module 10, the continuous morphology reconstruction module 20, the hydrodynamic field inversion module 30, and the state fusion estimation module 40 will provide new measurement data and updated system state estimates. .

[0232] The predictive control decision module 50 will take this new state as the starting point, reconstruct and solve the new MPC optimization problem. This rolling optimization mechanism, which replans future control actions based on the latest state information at each time step, enables the control system to continuously provide feedback correction for measurement noise, model errors and unforeseen environmental disturbances, thereby ensuring the robustness and accuracy of the entire control process.

Claims

1. A real-time monitoring and control system for hinged submerged raft construction based on multi-sensor fusion, characterized in that, include: The data acquisition and preprocessing module is used to acquire and preprocess the strain measurement data and sparse absolute position measurement data of the hinge sinker. The hydrodynamic field inversion module, connected to the data acquisition and preprocessing module, is used to invert the near-field hydrodynamic environment of the hinged submerged raft based on the strain measurement data and the parameterized forward dynamic model of the hinged submerged raft. The continuous morphology reconstruction module is connected to the data acquisition and preprocessing module and is used to convert the strain measurement data into high-density morphology measurement data based on the strain measurement data and using a preset strain-curvature mapping model. The state fusion estimation module, connected to the continuous morphology reconstruction module, is used to fuse the sparse absolute position measurement data and the high-density morphology measurement data to estimate a system state vector containing the position, attitude, and morphology of the hinged submerged raft. The predictive control decision module, connected to the hydrodynamic field inversion module and the state fusion estimation module, is used to generate an optimal control sequence for controlling the hinged submerged raft based on the system state vector and the near-field hydrodynamic environment. An actuator, connected to the predictive control decision module, is used to receive and execute the optimal control sequence; The state fusion estimation module specifically adopts an extended Kalman filter framework, which estimates the system state vector by periodically executing prediction and update steps and fusing the sparse absolute position measurement data and the high-density morphology measurement data. The extended Kalman filter framework further includes a manifold projection correction step, which is performed after the update step. This step uses a projection operator on the morphological manifold to project the morphological parameters in the updated system state vector onto a morphological manifold that has been pre-learned from historical or simulation data.

2. The real-time monitoring and control system for hinged submerged raft construction based on multi-sensor fusion as described in claim 1, characterized in that, The hydrodynamic field inversion module is specifically used for: A parameterized forward dynamic model of the hinged submerged raft is established, taking the near-field hydrodynamic environment as an unknown parameter. Define a cost function for the error between the strain predicted by the positive dynamic model and the strain measurement data; A gradient-based optimization algorithm is used to iteratively solve for the near-field hydrodynamic environment that minimizes the cost function.

3. The real-time monitoring and control system for hinged submerged raft construction based on multi-sensor fusion according to claim 2, characterized in that, The gradient-based optimization algorithm uses the adjoint method to calculate the gradient of the cost function with respect to the near-field hydrodynamic environment.

4. The real-time monitoring and control system for hinged submerged raft construction based on multi-sensor fusion as described in claim 1, characterized in that, The predictive control decision module employs a model predictive control method, which generates the optimal control sequence by solving an optimization problem defined in the prediction time domain and subject to multiple constraints within each control cycle.

5. The real-time monitoring and control system for hinged submerged raft construction based on multi-sensor fusion according to claim 4, characterized in that, The model predictive control method integrates a hydrodynamic disturbance feedforward model into a predictive model. The hydrodynamic disturbance feedforward model uses the near-field hydrodynamic environment obtained by the hydrodynamic field inversion module to pre-compensate for the flow disturbance in the optimization problem.

6. The real-time monitoring and control system for hinged submerged raft construction based on multi-sensor fusion according to claim 4, characterized in that, The multiple constraints include structural safety constraints and physical limitations of the actuator; The structural safety constraints include curvature constraints or strain constraints on the hinge sinkers. The physical limitations of the actuator include the amplitude limitation of the control quantity or the rate of change limitation of the control quantity.

7. The real-time monitoring and control system for hinged submerged raft construction based on multi-sensor fusion as described in claim 1, characterized in that, The system state vector includes the global position, attitude, linear velocity, and angular velocity of the hinge sinker at the integration starting point, as well as morphological parameters describing its bending and torsional morphology.

8. A method for real-time monitoring and control of hinged submerged raft construction based on multi-sensor fusion, employing the real-time monitoring and control system for hinged submerged raft construction based on multi-sensor fusion as described in any one of claims 1-7, characterized in that, Includes the following steps: Collect and preprocess strain measurement data and sparse absolute position measurement data of hinge sinkers; Based on the strain measurement data, the strain measurement data is converted into high-density morphological measurement data using a preset strain-curvature mapping model. Based on the strain measurement data and the parameterized forward dynamic model of the hinged submerged raft, the near-field hydrodynamic environment of the hinged submerged raft is obtained by inversion. By fusing the sparse absolute position measurement data and the high-density morphology measurement data converted from the strain measurement data, a system state vector containing the position, attitude, and morphology of the hinge sinker is estimated. Based on the system state vector and the near-field hydrodynamic environment, an optimal control sequence for controlling the hinged submerged raft is generated; According to the optimal control sequence, the control actuator performs construction operations on the hinged sinker.

Images