A ship deadweight anti-pushing and rollover risk early warning method and system

Abstract

The application provides a ship load capacity back-stepping and roll-over risk early warning method and system, which is based on the rigid body kinematics fact that small angle roll motion of a ship can be approximated as rotation around a transient rotation axis parallel to the longitudinal axis of the ship body, through two pre-marked synchronous acquisition nodes which do not coincide in the transverse plane projection, the ship transverse acceleration, vertical acceleration and roll angle speed time series data are collected respectively; the projection coordinates of the roll transient rotation center in the transverse plane are directly calculated according to the kinematics relationship, the space-time dispersion characteristics formed by the continuous multiple frames of rotation centers are extracted; a plurality of groups of rotation center dispersion data of the ship type under different load capacities, different load distributions and different sea conditions are collected, and a load capacity back-stepping model and a roll-over risk classification model are constructed; in the process of ship navigation, the instantaneous rotation center is calculated in real time and input into the model, so that the current load capacity and the remaining load capacity can be back-stepped, and the real-time evaluation and early warning of the ship roll-over risk can be realized.

Description

Technical Field

[0001] This invention relates to the fields of ship engineering, load measurement and maritime safety early warning technology, and in particular to a method and system for ship load reverse thrust and capsizing risk early warning. Background Technology

[0002] In scenarios such as ocean shipping, inland waterway shipping, port loading and unloading scheduling, ship stability verification, and maritime safety supervision, the measurement of ship deadweight and the prediction of capsizing risk are fundamental aspects of ship operation. Deadweight is the basic data for optimizing ship loading schemes, port loading and unloading scheduling, freight calculation, and stability boundary verification; its measurement accuracy directly determines the ship's operational economy and loading safety. Ship capsizing risk is the control target for ship navigation stability management, accident prevention, and emergency response; the timeliness and accuracy of risk warning directly determine the inherent safety of ship navigation.

[0003] There are four main types of existing load capacity measurement methods:

[0004] 1. Manual Draft Measurement Method: Crew members manually read the ship's draft at six points and calculate the load capacity using the ship's hydrostatic curve. This method is greatly affected by waves, tides, and hull deformation, and is only suitable for calm water environments such as anchoring or berthing. It cannot be used for dynamic measurements during ship navigation.

[0005] 2. Draft sensor method: Pressure or ultrasonic draft sensors are installed at the bow, midship, and stern of the ship to infer the load capacity from the draft data. However, these sensors are susceptible to damage from marine organisms, siltation, and wave impact. Their accuracy deteriorates significantly with long-term use, and hull deformation can introduce systematic errors. Furthermore, the data fluctuates greatly under adverse sea conditions.

[0006] 3. Ballast Water Tank Level Method: This method calculates the weight of ballast water using level sensors in each ballast tank, and then estimates the total deadweight by combining this with the weight of the empty ship. This method requires the installation of sensors and associated piping in dozens of ballast tanks, resulting in complex wiring, extremely high installation and maintenance costs, and significant systematic errors introduced by changes in ballast water density due to temperature and salinity. Furthermore, it cannot directly measure real-time changes in cargo weight.

[0007] 4. Hull Strain Gauge Method: Strain gauges are attached to key structural parts of the hull, and the load capacity is calculated by analyzing the relationship between structural strain and load. This method is significantly affected by ambient temperature, hull fatigue, and welding residual stress, requires frequent on-site calibration, has poor long-term reliability, and can only measure the overall load, failing to distinguish between the weight changes of cargo and ballast water.

[0008] The existing ship capsizing risk early warning technologies are as follows:

[0009] 1. Traditional Static Stability Verification Method: This method relies on the hydrostatic curves and stability cross curves obtained during the ship's design phase to calculate static stability parameters such as initial metacentric height, stability criterion number, and maximum restoring arm, thereby completing stability verification and risk assessment. This method can only perform static calculations for fixed loading conditions in port / anchorage states and cannot address sudden stability hazards during navigation.

[0010] 2. Roll Motion Parameter Threshold Method: This method collects motion parameters such as roll angle, roll angular velocity, and angular acceleration of the ship using gyroscopes and attitude sensors, and sets fixed thresholds to trigger a capsizing warning. This method suffers from severe warning lag; by the time the roll motion parameters reach the preset threshold, the ship has already entered a critical capsizing state, providing only a very short window for emergency response for the crew. Furthermore, the fixed thresholds are not adaptable to different ship types, different load conditions, and different sea conditions, making false alarms and missed alarms highly likely.

[0011] 3. Multi-sensor fusion stability monitoring method: This method involves deploying numerous sensing devices such as draft sensors, level sensors, hull strain gauges, and weather stations to fuse multi-source data for assessing ship stability and capsizing risk. This approach requires invasive installation, necessitating large-scale modifications to the hull structure and piping system. The system is highly complex, with extremely high installation and maintenance costs. Furthermore, data from most sensors fluctuates significantly under harsh sea conditions.

[0012] In summary, existing technologies struggle to simultaneously achieve continuous dynamic monitoring of load capacity and high-precision early warning under complex sea conditions, lacking non-invasive measurement methods based on the inherent motion characteristics of ships. Ship roll motion is one of the most significant degrees of freedom of motion for a ship excited by waves, and its kinematic characteristics are directly related to the ship's mass and inertia distribution; it has a deterministic quantitative mapping relationship with the ship's load capacity and directly reflects the attenuation trend of the ship's dynamic stability and the level of capsizing risk. However, existing publicly available technologies have not yet established a quantitative mapping relationship between the distribution characteristics of the instantaneous roll rotation center and the ship's load capacity and capsizing risk level. Therefore, there is an urgent need to develop a technical solution for ship load capacity and capsizing risk early warning that is low-cost to modify, highly resistant to interference, and capable of all-weather dynamic monitoring. Summary of the Invention

[0013] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a method and system for ship load reverse thrust and capsizing risk early warning, in order to solve the problem that the prior art is difficult to achieve continuous dynamic monitoring of load and high-precision early warning in complex sea conditions, and lacks non-invasive measurement methods based on the inherent motion characteristics of ships.

[0014] To achieve the above and other related objectives, the present invention provides the following technical solution:

[0015] A method for ship load reverse thrust and capsizing risk early warning, applied to a shipborne data processing terminal, includes the following steps: acquiring ship roll motion related data during navigation from two acquisition nodes installed on the hull simultaneously, wherein the ship roll motion related data includes lateral specific force, vertical specific force, and roll angular velocity; preprocessing the ship roll motion related data, and obtaining the corrected roll angular velocity and the instantaneous velocity vector of the two acquisition nodes in the cross section of the hull based on the preprocessed ship roll motion related data;

[0016] The weighted average turning center coordinates and the standardized turning center dispersion feature vector are obtained based on the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes in the hull cross section. The weighted average turning center coordinates and the standardized turning center dispersion feature vector are then input into the load factor inverse model and the capsizing risk classification model, respectively. The load factor inverse model and the capsizing risk classification model output the current total weight of the ship and the membership probability of each risk level at the current moment, respectively.

[0017] In one embodiment of the present invention, before acquiring the ship's hull roll motion data collected synchronously by two acquisition nodes installed on the hull during navigation, the method further includes: constructing a load-weight reverse thrust model and a capsizing risk classification model.

[0018] In one embodiment of the present invention, the construction of the load capacity reverse model and the roll risk classification model includes: collecting effective roll center coordinate data under different working conditions and sea conditions, forming a sample dataset, preprocessing the sample dataset, and dividing the preprocessed sample dataset into a training set and a test set, wherein the effective roll center coordinate data includes a weighted average roll center coordinate and a standardized roll center scatter feature vector;

[0019] A deterministic mapping relationship is established between the weighted average turning center coordinates and the total weight of the ship. An initial load factor inverse model is obtained based on the established deterministic mapping relationship. The initial load factor inverse model is trained using a training set to obtain the load factor inverse model. An initial capsizing risk classification model is constructed and trained using a training set to obtain the capsizing risk classification model.

[0020] In one embodiment of the present invention, the step of preprocessing the ship's roll motion related data and obtaining the corrected roll angular velocity and the instantaneous velocity vector of the two acquisition nodes in the ship's cross section based on the preprocessed ship's roll motion related data includes: correcting the roll angular velocity, obtaining the corrected roll angular velocity based on the correction result, and filtering the effective segments of the corrected roll angular velocity.

[0021] The lateral and vertical specific forces are compensated for by gravity tilt components, and the compensated net specific forces are preprocessed, including DC removal, filtering, and detrending. The instantaneous velocity of the nodes is calculated axis by axis for the preprocessed net specific force data, and the calculated instantaneous velocity of the nodes is normalized and Gaussian weighted smoothed. The instantaneous velocity vectors of the two acquisition nodes in the cross section of the hull are obtained based on the processing results.

[0022] In one embodiment of the present invention, obtaining the weighted average rotation center coordinates and the standardized rotation center dispersion feature vector based on the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes in the hull cross section includes: obtaining the instantaneous rotation center coordinates of each effective moment of the main path based on the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes in the hull cross section and using the gyroscope direct reading method;

[0023] Based on the instantaneous velocity vectors of the two acquisition nodes within the ship's cross section, and using the double-point velocity perpendicular line method, the instantaneous rotation center coordinates at each effective moment of the redundant path are obtained; based on the instantaneous rotation center coordinates at each effective moment of the main path and the redundant path, the weighted average rotation center coordinates and the standardized rotation center dispersion feature vector are obtained.

[0024] In one embodiment of the present invention, obtaining the instantaneous rotation center coordinates of each effective moment of the main path based on the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes in the hull cross section, and using the gyroscope direct reading method, includes: determining the instantaneous rotation center coordinates of each effective moment of the main path according to the following formula: ;

[0025] in, The instantaneous rotation center coordinates of the main path are: [coordinates of the instantaneous rotation center coordinates of the main path]; the instantaneous rotation center coordinates of the first acquisition node are: [coordinates of the instantaneous rotation center coordinates of the first acquisition node]. The instantaneous rotation center coordinates corresponding to the second acquisition node are: The instantaneous rotation center coordinates corresponding to the first acquisition node and the instantaneous rotation center coordinates corresponding to the second acquisition node are calculated from the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes in the cross section of the hull.

[0026] In one embodiment of the present invention, obtaining the instantaneous rotation center coordinates of the redundant path at each effective moment based on the instantaneous velocity vectors of the two acquisition nodes in the hull cross section and using the double-point velocity perpendicular line method includes: determining the instantaneous rotation center coordinates of the redundant path at each effective moment according to the following formula: ;

[0027] ;

[0028] in, The instantaneous rotation center coordinates of the redundant path; and The coordinates of the two acquisition nodes in the ship's coordinate system after calibration; , The instantaneous velocity vectors of the two acquisition nodes within the cross section of the ship's hull.

[0029] In one embodiment of the present invention, after inputting the weighted average turning center coordinates and the standardized turning center scatter feature vector into the load capacity back-calculation model and the capsizing risk classification model respectively, and outputting the current total ship weight and the membership probability of each risk level at the current moment through the load capacity back-calculation model and the capsizing risk classification model respectively, the method further includes: calculating the current actual load capacity based on the current total ship weight and combined with the fixed empty ship weight of the ship type, obtaining the ship's rated load capacity, comparing the current actual load capacity with the ship's rated load capacity, and confirming whether to immediately trigger the ship overload audible and visual warning based on the comparison result;

[0030] Based on the membership probability of each risk level at the current moment, the level corresponding to the highest membership probability is taken as the current ship capsizing risk level, and the risk evolution trend within the next N seconds is predicted.

[0031] A ship load reverse thrust and capsizing risk early warning system includes: a measuring device, wherein the measuring device includes two synchronous motion acquisition nodes and a shipborne data processing terminal, and the acquisition nodes include a three-axis accelerometer, a single-axis X-axis fiber optic gyroscope, a GPS or Beidou synchronous clock module and a data storage and transmission module;

[0032] The shipborne data processing terminal includes a model building module, a data acquisition module, a preprocessing and cross-sectional velocity vector calculation module, a roll instantaneous rotation center calculation and scattering feature extraction module, and a real-time load and capsizing risk reverse calculation module.

[0033] The model building module is used to build a load capacity reverse model and a capsizing risk classification model; the data acquisition module is used to acquire the ship's roll motion data during navigation, which is synchronously collected by two acquisition nodes installed on the hull; the preprocessing and cross-sectional velocity vector calculation module is used to preprocess the ship's roll motion data and obtain the corrected roll angular velocity and the instantaneous velocity vector of the two acquisition nodes in the cross-section of the hull based on the preprocessed ship roll motion data.

[0034] The instantaneous roll center of rotation calculation and dispersion feature extraction module is used to obtain the weighted average coordinates of the roll center of rotation and the standardized dispersion feature vector of the roll center of rotation based on the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes in the hull cross section. The real-time load capacity and capsizing risk back-calculation module is used to input the weighted average coordinates of the roll center of rotation and the standardized dispersion feature vector of the roll center of rotation into the load capacity back-calculation model and the capsizing risk classification model, respectively, and output the current total weight of the ship and the membership probability of each risk level at the current time through the load capacity back-calculation model and the capsizing risk classification model, respectively.

[0035] As described above, the ship load reverse thrust and capsizing risk early warning method and system of the present invention have the following beneficial effects: 1. The present invention is a non-invasive design with low engineering implementation cost: The present invention only requires two synchronous motion acquisition nodes to realize the dual functions of load measurement and capsizing early warning. It does not require modification of the hull structure, deployment of a large number of liquid level sensors in the ballast tank, or modification of the ship's piping system. It is convenient to install on existing ships, has a short installation cycle, and the modification cost is reduced by more than 70% compared with the traditional two independent systems. It is suitable for large-scale application of various ocean-going and inland waterway vessels.

[0036] 2. This invention employs redundant calculations, resulting in strong stability under complex sea conditions: Based on the inherent rigid body kinematics characteristics of ship roll motion, this invention does not rely on the ship's static hydrostatic curve and is unaffected by environmental factors such as hull deformation, marine organism attachment, and tidal waves; through gyroscope direct reading and dual-path redundancy verification of dual-point velocity vertical lines, combined with a full-process noise suppression algorithm, it can still achieve effective early warning of capsizing risk under Beaufort scale 6 and below, with environmental adaptability far superior to existing technologies;

[0037] 3. This invention has high measurement accuracy and good dynamic response performance: By suppressing random noise through algorithms such as angular velocity weighted averaging and normalized Gaussian smoothing, the relative error of a single back-calculation of load capacity can be controlled within the order of 1%, which is far better than the 3%~5% error of the traditional manual draft measurement method; the data update frequency can reach 1Hz, which can reflect the changes in the ship's load capacity and the trend of stability decay in real time, realize continuous dynamic monitoring of the ship throughout the entire voyage, and solve the pain point of traditional methods that can only be measured statically when in port;

[0038] 4. This invention achieves advanced early warning of capsizing risk based on dispersion characteristics: This invention can capture the nonlinear decay trend of ship stability in advance by using the spatiotemporal evolution characteristics of the dispersion of the turning center. The early warning advance is 15-20 seconds longer than the traditional roll angle threshold method, which provides the crew with sufficient emergency response window, fundamentally improves the inherent safety level of ship navigation, and solves the pain point of delayed early warning in traditional methods.

[0039] 5. The model of this invention has strong generalization ability and can guarantee accuracy throughout the entire life cycle: This invention establishes a complete incremental update mechanism for the model, which can continuously optimize the model based on the ship's navigation history data, adapt to the changes in hull characteristics caused by the increase in ship age, structural modification, and equipment replacement, and ensure the accuracy and reliability of the system in long-term operation; for different ship types, only one sea trial sample collection is required to build a dedicated model, which can be adapted to mainstream ship types such as bulk carriers, container ships, and inland waterway cargo ships.

[0040] 6. The data of this invention is traceable throughout the entire process, meeting the requirements of maritime supervision: This invention establishes a complete encrypted data storage system, and the entire process of load change, stability status, and early warning records is traceable. It can not only provide data support for ship operation management, but also meet the law enforcement needs of maritime authorities for ship overload supervision, stability control, and accident tracing, thus possessing extremely strong engineering practical value. Attached Figure Description

[0041] Figure 1 The flowchart shown is a method for ship load reverse thrust and capsizing risk warning disclosed in an embodiment of the present invention.

[0042] Figure 2 The diagram shows the overall flowchart of the ship load reverse thrust and capsizing risk warning method disclosed in the embodiments of the present invention.

[0043] Figure 3 The diagram shows the geometric principle of calculating the instantaneous rotation center of roll at dual acquisition nodes in the ship load reverse thrust and capsizing risk early warning system disclosed in this embodiment of the invention.

[0044] Figure 4 The diagram shows the various modules within the shipborne data processing terminal of the ship's deadweight thrust reverse and capsizing risk early warning system disclosed in this embodiment of the invention. Detailed Implementation

[0045] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. It should be noted that, unless otherwise specified, the following embodiments and features described herein can be combined with each other.

[0046] The first embodiment of the present invention relates to a method for ship deadweight reverse thrust and capsizing risk early warning, the process of which is as follows: Figure 1 As shown, the details are as follows:

[0047] Step 101: Construct a load capacity reverse calculation model and a rollover risk classification model.

[0048] Specifically, firstly, effective roll and swerve center coordinate data under different operating conditions and sea states are collected to form a sample dataset. The sample dataset is preprocessed and divided into a training set and a test set. The effective roll and swerve center coordinate data includes weighted average swerve center coordinates and standardized swerve center scatter feature vectors. Then, a deterministic mapping relationship between the weighted average swerve center coordinates and the total weight of the vessel is established. Based on the established deterministic mapping relationship, an initial load factor inverse model is obtained. The initial load factor inverse model is trained using the training set to obtain the load factor inverse model. Finally, an initial capsizing risk classification model is constructed and trained using the training set to obtain the capsizing risk classification model.

[0049] It should be noted that in this embodiment, a second-order polynomial regression model is used to establish a deterministic mapping relationship between the weighted average turning center coordinates and the total weight of the ship, and a random forest algorithm is used to construct an initial capsizing risk classification model. In practical applications, other models or algorithms can also be selected to establish a deterministic mapping relationship between the weighted average turning center coordinates and the total weight of the ship, and to construct an initial capsizing risk classification model.

[0050] Step 102: Obtain relevant data on the ship's rolling motion during navigation from two data acquisition nodes installed on the hull.

[0051] Specifically, the data related to the ship's rolling motion include the lateral force, vertical force, and rolling angular velocity.

[0052] Step 103: Preprocess the relevant data of the ship's roll motion, and obtain the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes in the cross section of the ship based on the preprocessed relevant data of the ship's roll motion.

[0053] Specifically, firstly, the roll rate is corrected, and the corrected roll rate is obtained based on the correction results. Then, the effective segments of the corrected roll rate are selected. Next, gravity tilt component compensation is performed on the lateral and vertical specific forces, and the net specific force after compensation is preprocessed. The preprocessing includes DC removal, filtering, and detrending. Finally, the instantaneous velocity of the nodes is calculated axis by axis from the preprocessed net specific force data, and the calculated instantaneous velocity of the nodes is normalized and Gaussian weighted smoothed. Based on the processing results, the instantaneous velocity vectors of the two acquisition nodes in the transverse section of the hull are obtained.

[0054] Step 104: Based on the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes in the hull cross section, obtain the weighted average coordinates of the gyrocenter and the standardized gyrocenter scatter feature vector.

[0055] Specifically, firstly, based on the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes within the ship's cross section, the instantaneous rotation center coordinates at each effective moment of the main path are obtained using the gyroscope direct reading method; then, based on the instantaneous velocity vectors of the two acquisition nodes within the ship's cross section, the instantaneous rotation center coordinates at each effective moment of the redundant path are obtained using the double-point velocity perpendicular line method; finally, based on the instantaneous rotation center coordinates at each effective moment of the main path and the redundant path, the weighted average rotation center coordinates and the standardized rotation center dispersion feature vector are obtained.

[0056] It should be noted that the instantaneous rotation center coordinates at each valid moment of the main path are determined according to the following formula: ; The instantaneous rotation center coordinates of the main path; the instantaneous rotation center coordinates corresponding to the first acquisition node are... The instantaneous rotation center coordinates corresponding to the second acquisition node are: The instantaneous rotation center coordinates corresponding to the first and second acquisition nodes are calculated from the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes in the hull cross section.

[0057] The instantaneous rotation center coordinates at each effective moment of the redundant path are determined using the following formula: ; ; The instantaneous rotation center coordinates of the redundant path; and The coordinates of the two acquisition nodes in the ship's coordinate system after calibration; , The instantaneous velocity vectors of the two acquisition nodes within the cross section of the ship's hull.

[0058] Step 105: Input the weighted average turning center coordinates and the standardized turning center scatter feature vector into the load capacity inverse model and the capsizing risk classification model, respectively. Output the current total weight of the ship and the membership probability of each risk level at the current time through the load capacity inverse model and the capsizing risk classification model, respectively.

[0059] Specifically, after step 105, the following steps are also included: calculating the current actual load capacity based on the current total weight of the ship and the fixed empty weight of the ship type, obtaining the rated load capacity of the ship, comparing the current actual load capacity with the rated load capacity of the ship, and confirming whether to immediately trigger the ship overload audible and visual warning based on the comparison results.

[0060] Based on the membership probability of each risk level at the current moment, the level corresponding to the highest membership probability is taken as the current ship capsizing risk level, and the risk evolution trend in the next N seconds is predicted.

[0061] In practical applications, one aspect is that the inventive principle of this invention is as follows: the small-angle rolling motion of a ship under the excitation of ocean waves can be approximated as the rotation of a rigid body about an instantaneous axis of rotation parallel to the longitudinal axis of the hull (X-axis, pointing from the ship's length to the bow); the projection of this instantaneous axis of rotation onto the transverse section of the hull (YZ plane, perpendicular to the ship's length direction) is denoted as the instantaneous center of rotation of the rolling motion. ;

[0062] The instantaneous turning center position of a ship during roll is determined by the mass inertia distribution of the hull (including the mass of water attached to the hull), and therefore will exhibit a regular shift with changes in the ship's load capacity and its load distribution. Under different load conditions and sea states, the effective instantaneous turning centers of multiple consecutive frames will form a statistically characteristic scattered point cloud within the hull's cross section, and there is a nonlinear mapping relationship between this point cloud and the ship's load capacity and dynamic stability. This invention utilizes the correspondence between this scattered point cloud and the load capacity and dynamic stability to construct an empirical mapping model, thereby realizing load capacity back-calculation and capsizing risk warning.

[0063] Let the angular velocity vector of the ship's roll be... ,in Let be the scalar of the roll angular velocity about the X-axis; at any measuring point on the hull. The position vector relative to the instantaneous axis of rotation is projected onto the transverse section of the hull as follows: According to the velocity composition theorem for rigid body rotation about a fixed axis, the velocity vector of the roll motion at this measuring point is: (1), Equation (1) is the core kinematic equation of this invention, which describes the deterministic relationship between the velocity of the measuring point and the instantaneous rotation center coordinates during the roll motion; the roll motion will generate Y-axis and Z-axis velocity components at the same time, and the rotation center coordinates can be calculated by the accelerometer and roll gyroscope in the corresponding directions.

[0064] Based on the aforementioned core features, this invention simultaneously achieves ship load factor reverse thrust and capsizing risk early warning through a pre-constructed mapping model: weighted average turning center coordinates. It has a deterministic nonlinear mapping relationship with the ship's total weight and deadweight, which can complete the high-precision back-calculation of deadweight; the spatiotemporal distribution characteristics such as dispersion features, offset rate, and inherent rolling period have a quantitative mapping relationship with the ship's dynamic stability and capsizing risk level, which can complete the real-time assessment and early warning of capsizing risk.

[0065] II. Method and Flowchart, please refer to the following for details. Figure 2 This method requires only two synchronous motion acquisition nodes and a shipborne terminal equipped with a data processing and model inference module. The complete method consists of 7 core steps, as follows:

[0066] Step 1: Measurement System Construction and Hull Coordinate System Definition: The core purpose of this step is to build an integrated hardware measurement system, establish a unified spatial coordinate system and time reference, and provide a standardized basic framework for subsequent data acquisition and calculation.

[0067] 1.1 Definition of the ship's coordinate system: A right-handed rectangular coordinate system is established with the origin O at the intersection of the ship's midships perpendicular, baseline, and midships longitudinal section. The coordinate axes are defined strictly according to general marine engineering standards: the X-axis, pointing towards the bow along the ship's length, is positive; the Y-axis, pointing towards the port side along the ship's beam, is positive; and the Z-axis, perpendicular to the ship's baseline and pointing upwards, is positive. The ship's rolling motion is defined as a small rotation about the X-axis, with its instantaneous axis of rotation parallel to the X-axis. The projection of this axis onto the ship's transverse section is the instantaneous center of rotation for rolling. ;

[0068] 1.2 Hardware System Configuration: The measurement device consists of two synchronous motion acquisition nodes and one shipborne data processing terminal; each acquisition node integrates a three-axis accelerometer, a single-axis X-axis fiber optic gyroscope, a GPS / BeiDou synchronous clock module, and a data storage and transmission module; the core performance indicators must meet the following requirements: sampling frequency. Acceleration measurement range ±10g, measurement accuracy ≤0.001g; angular velocity measurement range ±100° / s, measurement accuracy ≤0.01° / s; inter-node clock synchronization accuracy ≤1µs.

[0069] 1.3 Node Deployment and Calibration: Two data acquisition nodes were rigidly fixed to the main deck or superstructure of the ship using rigid supports, avoiding installation on flexible, vibrating parts. High-precision calibration of the node coordinates was performed using 3D laser scanning. After calibration, the coordinates of the nodes in the ship's coordinate system were as follows: and Suppose that the following conditions must be met: (a) the two nodes are in The projections of the planes do not coincide, that is... (b) Projected distance between the two nodes in the cross section (c) The Y-coordinate of at least one node To ensure that the roll motion generates a measurable Z-axis velocity component at this node, and to avoid [the following]... lead to Unsolvable problems; please refer to the following for details. Figure 3 , Figure 3 A schematic diagram of the geometric principle for solving the instantaneous rotation center of roll of a dual-acquisition node (ship cross section YZ), where ① is the first synchronous motion acquisition node. ② is the second synchronous motion acquisition node. ; ③ is the instantaneous rotation center M of the roll; ④ is the origin of the coordinate system; ⑤ is the outer contour of the hull cross section; ⑥ is the shipborne data processing terminal.

[0070] Step 2: System-wide synchronous time-series data acquisition (corresponding to step 102): The core purpose of this step is to obtain the original data of the ship's rolling motion under a unified time reference, so as to provide reliable data support for subsequent calculation of the turning center and feature extraction.

[0071] Set a uniform sampling frequency Sampling duration Number of sampling points The synchronous clock module triggers two acquisition nodes to synchronously acquire roll motion-related data. Each node synchronously acquires three types of time-series data: lateral force. Vertical force Roll angular velocity , ;in Corresponding to two data collection nodes, The data is a time series sample. Immediately after the data collection is completed, the data quality is checked. If there is packet loss, exceeding limits, or a synchronization error greater than 10µs, the data collection is retried. The raw data that passes the check is transmitted to the shipborne data processing terminal in real time through the wireless data transmission module.

[0072] Step 3: Data Preprocessing and Cross-Sectional Velocity Vector Calculation (Step 3 corresponds to Step 103 above): The core purpose of this step is to filter valid motion data, eliminate noise, drift, and gravity component interference in the original data, and calculate the instantaneous velocity vectors of the two nodes within the hull cross-section, providing a standardized and valid signal for the calculation of the center of rotation. The complete preprocessing and calculation process is as follows:

[0073] 3.1 Roll Angle and Effective Segment Selection: Gyroscope Zero-Off Correction: Use the average value over a long period of time in still water. As the zero bias of the gyroscope, the corrected angular velocity is: The corrected roll rate is integrated to obtain the instantaneous roll angle. : (2) After integration, least squares detrending is performed to eliminate the accumulated error caused by integration drift; only the following are retained. and The data segment is used as the effective segment, and data with low motion intensity and low signal-to-noise ratio are discarded; the setting of the 0.5° roll angle threshold is to ensure that the acceleration signal of the measuring point generated by the roll motion is significantly higher than the sensor's background noise, so as to ensure the calculation accuracy;

[0074] 3.2 Gravity Tilt Component Compensation: The raw data collected by the accelerometer is specific force, including components of hull motion acceleration and gravitational acceleration. During roll, the components of gravitational acceleration along the Y and Z axes of the body coordinate system change in real time with the roll angle, and this component is at the same frequency as the tangential acceleration generated by the roll rotation. Failure to compensate for this component will directly lead to systematic deviations in velocity and rotation center calculations. During roll, the projection vector of gravitational acceleration in the coordinate system is: (3), where, This represents gravitational acceleration; by subtracting the gravitational tilt component from the original specific force data, we obtain the net specific force generated solely by the ship's motion: (4); This step is necessary because In Components and rotational tangential acceleration If the frequency is the same, failure to compensate will contaminate subsequent calculations;

[0075] 3.3 DC removal, filtering, and detrending processing: For the compensated net specific force data, three operations are performed sequentially: (a) DC component removal: Eliminating the fixed bias caused by the sensor's zero bias, the processing formula is: (5), where, The actual data collection length is processed similarly. (b) 8th order Butterworth bandpass filter: the passband range is 0.1~2Hz, covering the main roll frequency range of most civilian ships; (c) Least squares detrending term processing: eliminates low-frequency trend interference caused by residual low-frequency drift and sensor temperature drift;

[0076] 3.4 Velocity Integration and Drift Suppression: The trapezoidal integration method is used to calculate the instantaneous velocity at each node on the preprocessed acceleration data, with the initial velocity set to 0. (6); (7); among which After integration, the obtained velocity sequence is processed. , Then, high-pass filtering and detrending processing with a cutoff frequency of 0.1Hz are performed again to further suppress integral drift;

[0077] 3.5 Normalized Gaussian Weighted Smoothing: To suppress random noise, normalized Gaussian weighted smoothing is used for lengths of... Smoothing is performed using a sliding window; the window weight is defined as: (8); In the formula As a normalization factor, ensuring To avoid signal amplitude attenuation;

[0078] The smoothed speed is: (9); Finally, the smooth velocity vectors of the two nodes in the cross section of the hull are obtained. , This serves as the basis for subsequent instantaneous rotation center calculations.

[0079] Step 4: Calculation and Scattering Feature Extraction of Instantaneous Rotation Center in Roll (Step 4 corresponds to Step 104): The core purpose of this step is to calculate the instantaneous rotation center coordinates at each effective moment through two paths, which serve as redundant verifications; after verification, the spatiotemporal scattering features of the rotation center are extracted to provide input for subsequent model back-inference.

[0080] 4.1 Path A – Gyroscope Direct Reading Method (Main Solution Path): Deformed from the core kinematic equations of rigid body rotation about a fixed axis, for each node... ,when (angular velocity threshold) To avoid noise amplification at low angular velocities, the instantaneous rotation center coordinates corresponding to that node can be directly calculated. (10); The arithmetic mean of the solution results of the two nodes is used to obtain the instantaneous rotation center coordinates of the main path: (11);

[0081] 4.2 Path B – Two-Point Velocity Perpendicular Line Method (Redundant Verification Path): Based on the core characteristic of rigid body rotation about a fixed axis, the velocity vector of a node is always perpendicular to the line connecting that point to the instantaneous center of rotation, i.e. Within the transverse section of the hull, the following geometric relationship is satisfied: the dot product is zero. (12); The above equation is rearranged into a standard linear system of equations: (13); Determinant of the coefficient matrix of the system of equations: (14) If If the velocity vectors of the two nodes are approximately parallel, skip the data for that moment; otherwise, solve the equations using Cramer's rule to obtain the instantaneous rotation center coordinates of the redundant path: (15); (16);

[0082] 4.3 Dual-path fusion and effective data filtering: Set a consistency threshold for each effective time point. (Adaptable to mainstream sizes of civilian bulk carriers and container ships, balancing calculation accuracy and data utilization), if the norm of the difference between the two path calculation results is sufficient. Then, the arithmetic mean of the two is taken as the effective instantaneous rotation center coordinate at that moment. Otherwise, the solution is deemed invalid, and the data at that moment is discarded.

[0083] 4.4 Center of Rotation Weighted Average and Scatter Feature Extraction: For samples within the sampling period... The effective instantaneous rotation center coordinates, expressed as the square of the roll angular velocity. Calculate a weighted average for the weights (the greater the roll velocity, the higher the signal-to-noise ratio, and the greater the weighting percentage), and obtain the weighted average rotation center coordinates. The calculation formula is: (17) Simultaneously calculate the standard deviation of the scattering in the Y and Z directions of the rotation center. , Rotation center offset rate Rolling inherent period Based on these features, a standardized feature vector of rotation center distribution is constructed and used as input to the mapping model.

[0084] Step 5: Construction of a mapping model between ship type-specific turning center, ship deadweight, and stability (corresponding to step 101): The core purpose of this step is to complete the collection of sample data and establish a mapping model between the turning center distribution characteristics and the ship deadweight and capsizing risk level, so as to provide a standardized judgment benchmark for subsequent real-time back-inference and early warning.

[0085] 5.1 Sample Data Collection: Sample data will be collected during sea trials or specialized tests before vessel delivery; 1. Under five typical load conditions: unloaded, 1 / 4 rated load, 1 / 2 rated load, 3 / 4 rated load, and fully loaded; 2. Under five typical load distribution conditions: uniform distribution, concentrated at the bow, concentrated amidships, concentrated at the stern, and lateral unilateral offset; 3. Under Beaufort scale 1-6 sea states; 10 sets of effective roll center data will be collected for each condition, with each set including the data obtained from steps 3-4. The corresponding standard deadweight W (obtained by shore-based weighbridge or high-precision draft measurement), and the standard capsizing risk level based on the IMO ship stability code and ship model capsizing test, were collected in a total of no less than 250 valid samples.

[0086] The ship capsizing risk level is based on the ship's stability state and overturning risk, and is divided into four levels according to the International Integrity Code 2008 adopted by the International Maritime Organization (IMO): Level I (Low Risk): The ship's initial stability high GM meets the code requirements, the stability criterion number K≥1, the stability reserve is sufficient, and there is no risk of capsizing; Level II (Medium Risk): The ship's initial stability high GM is close to the lower limit of the code, the stability criterion number 1>K≥0.8, the rolling coupled motion is enhanced, and stability monitoring needs to be strengthened; Level III (High Risk): The ship's initial stability high GM is lower than the code requirements, the stability criterion number 0.8>K≥0.5, the stability is severely weakened, the nonlinear motion is significant, and stability recovery measures need to be taken immediately; Level IV (Critical Risk): The ship's stability criterion number K<0.5, close to the critical state of overturning, the stability reserve is about to be exhausted, and emergency avoidance measures need to be taken immediately.

[0087] 5.2 Data Cleaning and Dataset Partitioning: Using... The criteria and the isolated forest algorithm remove outlier sample data and eliminate invalid samples caused by abnormal sea conditions and operational errors during the sea trial. The cleaned valid sample dataset is divided into training set and test set in a 7:3 ratio for model training and accuracy verification, respectively.

[0088] 5.3 Construction of the deadweight inverse model: A second-order polynomial regression model was adopted to establish the weighted average turning center coordinates and the total weight of the ship. The deterministic mapping relationship is expressed in the model as follows: (19), where The coefficients to be determined for the model are obtained by fitting the training set data using the least squares method. For large ships with a wide load capacity range and significant nonlinear characteristics between the turning center and load capacity, a BP neural network or other nonlinear regression model can be used instead of the multinomial model. Alternatively, the standard deviation of the turning center dispersion can be added to the feature vector. , With the inherent period of roll Additional characteristic quantities are used to further distinguish different distribution conditions under the same load.

[0089] 5.4 Rollover Risk Classification Model Construction: Using the standardized turn center scatter feature vector as input and the rollover risk level as output, a rollover risk classification model is constructed using the random forest algorithm. The model output is the membership probability of each risk level. ( ): In the formula, This is the mapping function for random forests. For ships with sufficient sample size and computing power, Long Short-Term Memory (LSTM) networks can be used instead of random forest algorithms to fully utilize the correlation characteristics of time series data and further improve the lead time and accuracy of risk prediction.

[0090] 5.5 Model Validation and Optimization: Test Set Validation: The average relative error is 0.82%, and the maximum relative error is 1.5%, which meets the accuracy requirements for load capacity measurement; the capsizing risk classification model must meet the early warning accuracy rate of ≥98%. If the model does not meet the accuracy requirements, supplementary sample data for the corresponding working conditions will be provided, and the model structure and hyperparameters will be adjusted until the engineering application requirements for ship operation are met. Finally, a mapping model specific to this ship type will be obtained and deployed to the shipboard data processing terminal.

[0091] Step 6: Real-time load capacity and capsizing risk back-calculation (corresponding to step 105): The core purpose of this step is to simultaneously complete the real-time back-calculation of the ship's load capacity and the graded early warning of capsizing risk based on a pre-trained dual-objective model during the ship's navigation.

[0092] 6.1 Real-time data processing and feature extraction: During ship navigation, real-time data processing is performed according to steps 2-4. The average value of the rotation center coordinates is taken as the input by using a sliding window averaging method and taking the average value of the rotation center coordinates over 10 consecutive sampling periods (60s per period).

[0093] 6.2 Real-time load capacity calculation: Average rotation center coordinates after averaging the sliding window values. Substituting the values ​​into the pre-trained load factor inverse model, the current total weight of the ship is obtained. Combined with the fixed empty weight of this ship type Calculate the current actual load capacity. : (21); combined with the ship's rated deadweight Calculate the remaining loadable weight : (22) If the current load capacity Immediately trigger the ship overload audible and visual warning, and simultaneously record the overload data to the local database;

[0094] 6.3 Real-time capsizing risk assessment and graded early warning: The standardized scatter feature vector calculated in real time is input into the capsizing risk grading model, and the membership probability of each risk level at the current moment is output. The level corresponding to the highest membership probability is taken as the current ship capsizing risk level. At the same time, based on the risk probability change trend of 10 consecutive sliding windows, the ARIMA time series prediction algorithm is used to predict the risk evolution trend in the next 30 seconds, further improving the early warning lead time.

[0095] Based on real-time risk levels and trend predictions, corresponding tiered early warning strategies are implemented: Level I Low Risk: The data processing terminal displays the ship's stability status normally, with no warning actions; Level II Medium Risk: A yellow interface warning is triggered, highlighting the risk level and reminding the crew to strengthen stability monitoring and sea state observation, while simultaneously pushing forward suggestions for course and speed optimization; Level III High Risk: An orange audible and visual warning is triggered, locking the warning interface and simultaneously pushing forward stability restoration measures such as ballast water adjustment, activation of anti-roll devices, and course reversal; Level IV Critical Risk: A red emergency audible and visual warning is triggered, simultaneously pushing forward the ship's emergency maneuvering plan and personnel evacuation procedures, and automatically sending distress signals and the ship's real-time location to the shore-based maritime supervision platform.

[0096] Step 7: Results Output, Data Storage, and Model Update: The core purpose of this step is to visualize the back-inference and early warning results, establish a full lifecycle data management and model optimization mechanism, and ensure the accuracy and reliability of the system in the long term.

[0097] 7.1 Visualized Output of Results: The shipboard data processing terminal displays core operational data in real time, including key information such as current load capacity, remaining load capacity, overload status, average turning center coordinates, capsizing risk level, instantaneous roll angle, and inherent roll period; it supports real-time plotting of instantaneous turning center scatter plots, historical data backtracking, and report export to meet the needs of crew operations and maritime regulatory inspections.

[0098] 7.2 Data storage: All operational data is encrypted and stored in the local database with timestamps, including: raw motion time series data, instantaneous turning center coordinates, scatter feature vectors, load back-calculation results, risk level judgment results, early warning records, ship operating conditions and sea state data. The data storage period is no less than one complete operating voyage of the ship, and a complete historical database is established to meet the needs of accident tracing and data analysis.

[0099] 7.3 Model Update: Establish a model lifecycle optimization mechanism. Every quarter, based on the newly added navigation history data of the ship, complete the model fine-tuning and accuracy optimization. After the ship completes major repair, structural modification, or dock repair, immediately supplement the full operating condition sample data and retrain the model. After the ship experiences a stability hazard, supplement the critical operating condition sample data and update the model.

[0100] The second embodiment of the present invention relates to a ship deadweight thrust reverser and capsizing risk early warning system. Please refer to [link to relevant documentation]. Figure 3 and Figure 4 The device includes a measuring device, which comprises two synchronous motion acquisition nodes and a shipborne data processing terminal. The acquisition nodes include a three-axis accelerometer, a single-axis X-axis fiber optic gyroscope, a GPS or BeiDou synchronous clock module, and a data storage and transmission module.

[0101] Specifically, the shipborne data processing terminal includes a model building module, a data acquisition module, a preprocessing and cross-sectional velocity vector calculation module, a roll instantaneous rotation center calculation and scatter feature extraction module, and a real-time load and capsizing risk reverse calculation module.

[0102] The system comprises the following modules: a model building module for constructing a load factor inversion model and a capsizing risk classification model; a data acquisition module for acquiring ship roll motion data collected synchronously by two acquisition nodes mounted on the hull during navigation; a preprocessing and cross-sectional velocity vector calculation module for preprocessing the ship roll motion data and obtaining the corrected roll angular velocity and instantaneous velocity vectors of the two acquisition nodes within the ship's cross-section based on the preprocessed data; a roll instantaneous turning center calculation and dispersion feature extraction module for obtaining the weighted average turning center coordinates and standardized turning center dispersion feature vectors based on the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes within the ship's cross-section; and a real-time load factor and capsizing risk inversion module for inputting the weighted average turning center coordinates and standardized turning center dispersion feature vectors into the load factor inversion model and the capsizing risk classification model, respectively, and outputting the current total ship weight and the membership probability of each risk level at the current moment through the load factor inversion model and the capsizing risk classification model.

[0103] It is not difficult to see that this embodiment is a system implementation corresponding to the first embodiment, and this embodiment can be implemented in conjunction with the first embodiment. The relevant technical details mentioned in the first embodiment are still valid in this embodiment, and will not be repeated here to reduce repetition. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the first embodiment.

[0104] It is worth mentioning that all modules involved in this embodiment are logical modules. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. Furthermore, to highlight the innovative aspects of this invention, this embodiment does not introduce units that are not closely related to solving the technical problem proposed by this invention; however, this does not mean that other units are absent from this embodiment.

[0105] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. All equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this invention should still be covered by the claims of this invention.

Claims

1. A method for early warning of ship deadweight thrust reverse and capsizing risk, characterized in that, Applied to shipborne data processing terminals, the method includes the following steps: Construct a load capacity inverse model and a rollover risk classification model; The ship acquires data related to its rolling motion during navigation from two acquisition nodes installed on the hull, which are collected synchronously. The data related to the rolling motion includes lateral force, vertical force, and rolling angular velocity. The data related to the ship's roll motion are preprocessed, and the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes in the cross section of the ship are obtained based on the preprocessed data. The weighted average coordinates of the gyrocenter and the standardized gyrocenter distribution feature vector are obtained based on the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes in the cross section of the hull. The weighted average turning center coordinates and the standardized turning center scatter feature vector are input into the load capacity inverse model and the capsizing risk classification model, respectively. The load capacity inverse model and the capsizing risk classification model output the current total weight of the ship and the membership probability of each risk level at the current time, respectively. The construction of the load capacity inverse model and the rollover risk classification model includes: Effective roll and yaw center coordinate data under different working conditions and sea conditions are collected and a sample dataset is constructed. The sample dataset is preprocessed and divided into a training set and a test set. The effective roll and yaw center coordinate data includes weighted average yaw center coordinates and standardized yaw center scatter feature vectors. A deterministic mapping relationship is established between the weighted average turning center coordinates and the total weight of the ship. An initial load factor inverse model is obtained based on the established deterministic mapping relationship. The initial load factor inverse model is then trained using a training set to obtain the load factor inverse model. Using the standardized turn center scatter feature vector as input and the rollover risk level as output, an initial rollover risk classification model is constructed using the random forest algorithm. The initial rollover risk classification model is then trained using the training set to obtain the rollover risk classification model.

2. The method for ship deadweight reverse thrust and capsizing risk early warning according to claim 1, characterized in that: The preprocessing of the ship's roll motion related data, and the obtaining of the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes in the ship's cross section based on the preprocessed ship's roll motion related data, includes: The roll rate is corrected, the corrected roll rate is obtained based on the correction result, and the effective segments of the corrected roll rate are screened. The lateral and vertical specific forces are compensated for by gravity tilt components, and the net specific force after compensation is preprocessed, wherein the preprocessing includes DC removal, filtering and detrending processing. The instantaneous velocity of the nodes is calculated axis by axis from the preprocessed net specific force data, and the calculated instantaneous velocity of the nodes is normalized and Gaussian weighted smoothed. Based on the processing results, the instantaneous velocity vectors of the two acquisition nodes in the cross section of the hull are obtained.

3. The method for ship deadweight reverse thrust and capsizing risk early warning according to claim 1, characterized in that: The process of obtaining the weighted average coordinates of the turning center and the standardized turning center distribution feature vector based on the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes within the hull cross section includes: Based on the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes in the hull cross section, the instantaneous rotation center coordinates of each effective moment of the main path are obtained using the gyroscope direct reading method. The instantaneous rotation center coordinates of each effective moment of the redundant path are obtained based on the instantaneous velocity vectors of the two acquisition nodes in the cross section of the hull and using the double-point velocity perpendicular line method. The weighted average rotation center coordinates and the standardized rotation center distribution feature vector are obtained based on the instantaneous rotation center coordinates at each effective moment of the main path and the redundant path.

4. The method for ship deadweight reverse thrust and capsizing risk early warning according to claim 3, characterized in that: The process of obtaining the instantaneous rotation center coordinates of each effective moment of the main path based on the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes within the ship's cross-section, using a gyroscope direct reading method, includes: The instantaneous rotation center coordinates at each valid moment of the main path are determined using the following formula: ； in, The instantaneous rotation center coordinates of the main path are: [coordinates of the instantaneous rotation center coordinates of the main path]; the instantaneous rotation center coordinates of the first acquisition node are: [coordinates of the instantaneous rotation center coordinates of the first acquisition node]. The instantaneous rotation center coordinates corresponding to the second acquisition node are: The instantaneous rotation center coordinates corresponding to the first acquisition node and the instantaneous rotation center coordinates corresponding to the second acquisition node are calculated from the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes in the cross section of the hull.

5. The method for ship deadweight reverse thrust and capsizing risk early warning according to claim 3, characterized in that: The step of obtaining the instantaneous rotation center coordinates of each effective moment of the redundant path based on the instantaneous velocity vectors of the two acquisition nodes in the ship's cross section and using the double-point velocity perpendicular line method includes: The instantaneous rotation center coordinates at each effective moment of the redundant path are determined using the following formula: ； ； in, The instantaneous rotation center coordinates of the redundant path; and The coordinates of the two acquisition nodes in the ship's coordinate system after calibration; , The instantaneous velocity vectors of the two acquisition nodes within the cross section of the ship's hull.

6. The method for ship deadweight reverse thrust and capsizing risk early warning according to claim 1, characterized in that: After inputting the weighted average turning center coordinates and the standardized turning center scatter feature vector into the load capacity inverse model and the capsizing risk classification model, respectively, and outputting the current total ship weight and the membership probability of each risk level at the current time through the load capacity inverse model and the capsizing risk classification model, the process further includes: The current actual load capacity is calculated based on the current total weight of the vessel and the fixed empty weight of the vessel type. The rated load capacity of the vessel is obtained, and the current actual load capacity is compared with the rated load capacity of the vessel. Based on the comparison result, it is determined whether to immediately trigger the vessel overload audible and visual warning. Based on the membership probability of each risk level at the current moment, the level corresponding to the highest membership probability is taken as the current ship capsizing risk level, and the risk evolution trend within the next N seconds is predicted.

7. A ship load factor reverse thrust and capsizing risk early warning system, characterized in that: include: The measuring device includes two synchronous motion acquisition nodes and a shipborne data processing terminal. The acquisition nodes include a three-axis accelerometer, a single-axis X-axis fiber optic gyroscope, a GPS or Beidou synchronous clock module, and a data storage and transmission module. The shipborne data processing terminal includes a model building module, a data acquisition module, a preprocessing and cross-sectional velocity vector calculation module, a roll instantaneous rotation center calculation and scattering feature extraction module, and a real-time load and capsizing risk reverse calculation module. The model building module is used to build a load capacity inverse estimation model and a rollover risk classification model. The specific process of building the load capacity inverse estimation model and the rollover risk classification model includes: Effective roll and yaw center coordinate data under different working conditions and sea conditions are collected and a sample dataset is constructed. The sample dataset is preprocessed and divided into a training set and a test set. The effective roll and yaw center coordinate data includes weighted average yaw center coordinates and standardized yaw center scatter feature vectors. A deterministic mapping relationship is established between the weighted average turning center coordinates and the total weight of the ship. An initial load factor inverse model is obtained based on the established deterministic mapping relationship. The initial load factor inverse model is then trained using a training set to obtain the load factor inverse model. Using the standardized turn center scatter feature vector as input and the rollover risk level as output, the initial rollover risk classification model is constructed using the random forest algorithm, and the initial rollover risk classification model is trained using the training set to obtain the rollover risk classification model. The data acquisition module is used to acquire data related to the ship's rolling motion during navigation, which are collected synchronously by two acquisition nodes installed on the ship's hull. The preprocessing and cross-sectional velocity vector calculation module is used to preprocess the ship's roll motion related data, and obtain the corrected roll angular velocity and the instantaneous velocity vector of the two acquisition nodes in the ship's cross section based on the preprocessed ship's roll motion related data. The instantaneous roll center of rotation calculation and dispersion feature extraction module is used to obtain the weighted average coordinates of the center of rotation and the standardized dispersion feature vector of the center of rotation based on the corrected roll angular velocity and the instantaneous velocity vectors of the two acquisition nodes in the cross section of the hull. The real-time load capacity and capsizing risk inversion module is used to input the weighted average turning center coordinates and the standardized turning center scatter feature vector into the load capacity inversion model and the capsizing risk classification model, respectively. The load capacity inversion model and the capsizing risk classification model output the current total weight of the ship and the membership probability of each risk level at the current time, respectively.

Images