A short-term ship attitude continuous-time prediction method and system under irregular sampling

Abstract

The application discloses a short-term attitude continuous-time prediction method for a ship under irregular sampling, and the method comprises the following steps: acquiring historical observation data of the ship, and reserving real-time interval information and missing observation mark information between adjacent observation time points; based on kinematics of the ship and low-order dynamics, a ship type parameterized physical consistency residual is constructed according to a parameter set of the target ship; the above information is input into a continuous-time state updating model, the real-time interval is explicitly used to drive the internal state evolution, and an initial attitude prediction result is generated; and the final prediction result is obtained through checking and feedback correction based on the physical consistency residual. The application can adapt to the irregular sampling condition of the actual ship-borne sensor, improve the generalization ability under the limited sample through the physical constraint, inhibit the error accumulation through the closed-loop correction, and improve the accuracy and stability of the ship attitude prediction.

Description

Technical Field

[0001] This invention relates to the field of ship motion state prediction technology, and more specifically, to a method and system for predicting the short-term continuous time attitude of a ship under irregular sampling. Background Technology

[0002] Ship attitude prediction is a crucial component of ship motion state perception and auxiliary decision-making systems. It primarily predicts the ship's attitude states, such as roll, pitch, and heave, within a future time window based on historical motion states, environmental excitation information, and relevant operating conditions. During actual navigation and maritime operations, ships are influenced by a combination of factors, including wind, waves, currents, and maneuvering conditions, resulting in attitude motion processes exhibiting significant nonlinearity, coupling, time-varying characteristics, and continuous dynamic evolution.

[0003] Currently, ship attitude prediction methods mainly include those based on dynamic mechanism models and those based on historical data. Methods based on dynamic mechanism models typically rely on relatively accurate ship motion equations, parameter identification results, and environmental modeling conditions, but they suffer from limited adaptability under complex sea states and varying operating conditions. Methods based on historical data primarily utilize historical observation data to establish ship attitude change patterns, but they still have certain limitations under conditions of limited samples, significant sea state variations, irregular sampling, and significant multi-degree-of-freedom coupling characteristics.

[0004] However, existing ship attitude prediction methods have the following shortcomings in practical applications: First, most methods assume a fixed sensor sampling interval, requiring the original observation data to be forcibly resampled into a regular time step sequence. However, in reality, shipborne sensors are affected by the working environment and equipment status, and the collected data often has irregularities such as non-fixed sampling intervals, asynchronous acquisition, and local missing data. Forced resampling will distort the true time rhythm, leading to a decrease in prediction accuracy. Second, existing data-driven methods rely on large-scale, high-quality labeled data, which has insufficient generalization ability under limited sample conditions, and does not fully utilize the coupling characteristics between multiple degrees of freedom such as ship roll, pitch, and heave. Third, existing recursive prediction methods lack effective error constraints and correction mechanisms. When the prediction step size increases or the sea state changes drastically, the error gradually accumulates and drifts, affecting the reliability of the prediction results.

[0005] Therefore, it is necessary to propose a method for predicting the continuous time of short-term ship attitude under irregular sampling to solve the above-mentioned technical problems. Summary of the Invention

[0006] The purpose of this invention is to provide a method for predicting the continuous time of short-term ship attitude under irregular sampling, so as to solve the problem of insufficient adaptability of the existing technology under irregular sampling, multiple operating conditions, limited samples and error accumulation conditions, and improve the accuracy, stability and physical rationality of ship short-term attitude prediction.

[0007] The objective of this invention can be achieved through the following technical solutions: A method for predicting the continuous time of short-term ship attitude under irregular sampling includes the following steps: Acquire historical observation data of the ship; the historical observation data is organized according to the original sampling time, and retains the true time interval information between adjacent observation times and the missing measurement mark information; Based on ship kinematics and low-order dynamics, and according to the parameter set of the target ship, a parameterized physical consistency residual for the ship form is constructed. The historical observation data, the real time interval information, the missing measurement marker information, and the ship type parameterized physical consistency residual are input into the continuous time state update model. The continuous time state update model explicitly uses the real time interval information to drive the internal state evolution and generate the initial attitude prediction result within the future prediction window. The initial attitude prediction result is verified based on the parametric physical consistency residual of the ship type. If the verification fails, the initial attitude prediction result is corrected, and the correction information is fed back to the continuous time state update model to obtain the final attitude prediction result.

[0008] Furthermore, the acquisition of historical observation data of the ship specifically includes: Collect ship attitude data, environmental excitation data, and operating condition data; Outlier handling, unit standardization, and time reference standardization are performed on the collected raw data; The true time interval between adjacent observation times in the unified data is preserved, and a missing data marker is generated to indicate the validity of each input variable at the corresponding observation time. The processed data is organized into a joint input sequence according to the original sampling time, without mandatory resampling into a regular sequence with a fixed time step.

[0009] Furthermore, the construction of the parametric physical consistency residuals for the ship type specifically includes: Based on the fundamental kinematic relationship between predicted displacement and predicted velocity, and based on the actual time interval between adjacent observation times, a kinematic consistency residual is constructed. Based on the inertial, damping, and recovery terms corresponding to each degree of freedom, as well as the cross-inertial, cross-damping, and cross-recovery terms between different degrees of freedom, a low-order dynamic consistency residual is constructed. Obtain the parameter set of the target ship, which includes the ship's length, beam, draft, displacement, initial metacentric height, and moment of inertia; The parameter mapping module maps the parameter set of the target ship to the equivalent parameters corresponding to the inertia term, the damping term, the recovery term, and each cross term, thus obtaining the parameterized physical consistency residual of the ship type.

[0010] Furthermore, the explicit use of the real time interval information to drive the internal state evolution by the continuous-time state update model specifically includes: Define internal state variables, which are used to compress and express historical attitude observations, environmental excitations, operating conditions, time rhythms, physical constraints, and target ship type information. The internal state variables are updated using a continuous-time state update function. The continuous-time state update function explicitly depends on the actual time interval between adjacent observation times. When the sampling interval is short, the state change is small, and when the sampling interval is long, the state change increases accordingly. The continuous-time state update function adopts the following discrete update form: in, Indicates time The internal state, Indicates adjacent observation times and The actual time interval between them Indicates time Input features, This represents the function that changes the internal state.

[0011] Furthermore, the generation of the initial pose prediction result within the future prediction window specifically includes: Based on the updated internal state variables, the attitude prediction results for multiple target times within the future prediction window are generated at once through the output mapping function; The output mapping function is expressed as: Where G() is the output mapping function, This indicates the time corresponding to the current moment. The sequence of initial attitude prediction results for multiple target times within the future prediction window.

[0012] Furthermore, the initial attitude prediction result is verified based on the parametric physical consistency residual of the ship type. If the verification fails, the initial attitude prediction result is corrected, and the correction information is fed back to the continuous-time state update model. This specifically includes: Calculate the physical residual corresponding to the initial attitude prediction result to obtain the residual norm; The residual norm is compared with a preset threshold. When the residual norm does not exceed the preset threshold, the initial attitude prediction result is output as the final attitude prediction result. When the residual norm exceeds the preset threshold, output correction and internal state feedback correction are performed. The output correction is expressed as: in, This is the initial prediction result. For correction factors, For deviation mapping function, For physical residuals; The internal state feedback correction is expressed as: in, For feedback coefficients, For functions mapped to the state space, the modified internal state is used for state initialization of the next prediction window.

[0013] Furthermore, after obtaining the final attitude prediction result, the method also includes a result evaluation and output step, specifically including: The final attitude prediction results are subjected to physical consistency assessment and usability determination to generate comprehensive evaluation indicators; The comprehensive evaluation index is compared with a preset threshold to determine whether the prediction result is usable. When the prediction result is unavailable, generate an error message, warning message or control trigger signal; Output the final attitude prediction result and the comprehensive evaluation index.

[0014] Compared with existing technologies, the advantages of this method are as follows: 1. It retains the true time interval information and missing measurement markers in the original observation data, and does not force resampling to a fixed time step sequence, so that the continuous time state update model can explicitly use the actual time difference between adjacent observation times to drive the internal state evolution, effectively adapting to irregular situations such as non-fixed sampling intervals, asynchronous acquisition and local missing data in actual shipborne sensor data. 2. Construct a parameterized physical consistency residual for the ship type, which includes kinematic consistency constraints based on real time intervals and low-order dynamic consistency constraints considering multi-degree-of-freedom coupling effects. The residual is then mapped to equivalent parameters corresponding to the specific ship type through the target ship parameter set, so that the prediction process can achieve the synergistic effect of observation information and physical laws, and maintain good generalization ability under limited sample conditions. 3. After obtaining the initial prediction results, the physical consistency residual of the ship type parameterization is used for verification. When the physical residual exceeds the preset threshold, output correction and internal state feedback correction are performed to form a closed-loop mechanism of "prediction-verification-feedback correction", which effectively suppresses the accumulation of errors and drift in the recursive prediction process. 4. While outputting prediction results, a comprehensive evaluation index is generated by combining prediction error risk and physical consistency residuals. The availability of prediction results is also determined. When the prediction results are unavailable, abnormal prompts, early warning information or control trigger signals are automatically generated. It can directly serve engineering application scenarios such as ship motion simulation, wave compensation auxiliary control, and offshore operation safety assessment.

[0015] This application also provides a system for predicting the continuous time of short-term attitude of a ship under irregular sampling, including: a data acquisition module for acquiring historical observation data of the ship; the historical observation data is organized according to the original sampling time and retains the true time interval information and missing measurement mark information between adjacent observation times; The physical consistency residual construction module is used to construct parametric physical consistency residuals of the ship type based on the ship's kinematic relationships and low-order dynamic relationships, and according to the parameter set of the target ship. The continuous-time state update prediction module is used to input the historical observation data, the real time interval information, the missing measurement marker information and the ship type parameterized physical consistency residual into the continuous-time state update model. The continuous-time state update model explicitly uses the real time interval information to drive the internal state evolution and generate the initial attitude prediction result within the future prediction window. The residual verification and feedback correction module is used to verify the initial attitude prediction result based on the parametric physical consistency residual of the ship type. When the verification fails, the initial attitude prediction result is corrected and the correction information is fed back to the continuous time state update model to obtain the final attitude prediction result. The result evaluation and output module is used to evaluate and output the final attitude prediction result.

[0016] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of a method for predicting the continuous time of short-term attitude of a ship under irregular sampling.

[0017] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for predicting the continuous time of short-term ship attitude under irregular sampling.

[0018] The beneficial effects and methods of the ship short-term attitude continuous time prediction system, electronic equipment and computer-readable storage medium provided in this application are as described above. Attached Figure Description

[0019] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart of a method for predicting the continuous time of short-term ship attitude under irregular sampling provided in an embodiment of the present invention; Figure 2 A block diagram of a ship short-term attitude continuous time prediction system under irregular sampling provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a continuous-time prediction verification feedback correction closed loop based on the physical consistency residual of ship type parameterization provided in an embodiment of the present invention. Detailed Implementation

[0020] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] Reference Figure 2 The ship short-term attitude continuous-time prediction system under irregular sampling provided in this application mainly consists of a data acquisition module, a joint state input sequence construction module, a physical consistency residual construction module, an offline training and parameter determination module, a continuous-time state update prediction module, a residual verification and feedback correction module, and a result evaluation and output module. The modules cooperate and connect with each other to complete the ship short-term attitude prediction and result post-processing.

[0022] The data acquisition module is used to acquire historical ship attitude data, environmental excitation information, and operational condition information. It collects data through inertial measurement units, attitude sensors, displacement sensors, environmental monitoring equipment, and shipborne operation monitoring equipment. This information is organized according to the observation time and stored in association with the corresponding timestamps. The joint state input sequence construction module performs basic preprocessing on the acquired raw data, including data cleaning, standardization, time reference unification, and generation of missing measurement markers. It constructs a joint state sequence while retaining the time interval characteristics under irregular sampling conditions. This module does not reconstruct the original data into a regular sequence with a fixed time step; instead, it retains the time interval information and missing measurement markers between adjacent observation times during sequence construction.

[0023] The physical consistency residual construction module is based on the ship's general kinematics, low-order dynamics, and the parameter set of the target ship. It jointly constructs the ship type parameterized physical consistency constraint information and physical residuals for short-term attitude prediction, and transforms the ship's actual motion law into physical consistency residuals that can be used in the training, prediction and correction process.

[0024] The offline training and parameter determination module is used to train, identify, or determine the parameters of the continuous-time state update model, output mapping function, and related parameter mapping module based on historical samples, joint input sequences, and ship type parameterized physical consistency residuals before the system is officially deployed, so as to obtain the model parameters required for subsequent online prediction.

[0025] The continuous time state update prediction module is driven by real time intervals. Based on the joint state input, the ship type parameterized physical consistency residual, and the target ship parameters, it predicts the ship attitude within the future window time. Its internal state update explicitly utilizes the time feature information of adjacent intervals.

[0026] The residual verification and feedback correction module corrects the prediction results based on the initial prediction results and the deviation between the initial prediction results and the residuals. When the deviation is too large, the deviation information is fed back to the continuous time state update process for error suppression, forming a closed loop of "prediction-verification-feedback correction". The result evaluation and output module is used to output the prediction results and evaluate the results in conjunction with indicators such as physical consistency and result usability. It also has output functions such as threshold comparison, prompts, alarms and control triggers.

[0027] Reference Figure 3 , Figure 3 A schematic diagram of a continuous-time prediction verification feedback correction closed loop based on ship type parameterized physical consistency residual is shown. Historical observation data is updated continuously to generate initial prediction results. The initial prediction results are substituted into the physical consistency residual for verification. The deviation generated by the verification is used for output correction on the one hand, and fed back to the continuous-time state update process on the other hand, forming a complete closed-loop prediction framework.

[0028] Reference Figure 1 The method of this application will be described in detail below with reference to specific embodiments. Taking short-term ship attitude prediction as an example, we select three degrees of freedom closely related to ship operations—heave, roll, and pitch—for illustration. In other embodiments, it can be extended to predict six degrees of freedom (pitch, sway, heave, roll, pitch, and bow roll) according to actual needs. The length of the ship's historical attitude observation window is set as L, and the length of the target prediction time window is set as P, where L is the length of the historical input window and P is the length of the prediction window. This is to characterize the differences in physical constraint information under different ship types and loading modes.

[0029] This invention introduces a ship parameter set Its form is as follows in, As captain, For the width of the boat, For drinking water, For drainage volume, For initial stability height, Let be the moment of inertia about each coordinate axis. Parameters used to characterize loading status, ship type, or other additional features.

[0030] The physical consistency residual is constructed by combining general constraints and ship type parameterized correction constraints. Based on general dynamics and low-order dynamics constraints, and based on the target ship parameter set s, equivalent parameters corresponding to the target ship type are generated through parameter mapping module or preset mapping relationship. The ship type parameterized physical consistency residual is constructed in this way, so that the physical consistency residuals constructed under different ship types and different loading conditions are different, which can adapt to different target ship types.

[0031] In step 1, raw data acquisition is performed. Relevant data on the ship's motion are collected and observed using inertial measurement units, attitude sensors, displacement sensors, environmental monitoring equipment, and shipboard operation monitoring equipment. The collected data includes at least attitude state, environmental excitation information, and ship operating condition information, all organized and stored aligned to the observation time and associated with corresponding timestamps. Attitude state information can be written as: in, Indicates heave displacement. Indicates the roll angle. Indicates the pitch angle. Indicates the heave velocity. Indicates the roll rate. This indicates the pitch angular velocity.

[0032] Environmental incentive information can be written as: in, It can represent wave height or wave-related characteristic quantities. Indicates wave direction, Indicates wind speed. Indicates wind direction or current direction. This indicates environmental quantities such as flow rate.

[0033] Operating condition information can be written as: in, Indicates speed, represents the rudder angle, Indicates the thrust status of the propulsion unit. This indicates the loading status, operational status, or other operational parameters related to ship motion. In addition, it is necessary to read the principal dimensional parameters, stability parameters, mass parameters, etc., related to the ship type, to form the target ship type parameter set s.

[0034] In step 2, asynchronous observation feature preservation and data preprocessing are performed. Outlier identification, necessary standardization, and unification of dimensions and time references for the collected raw observation data are conducted. The set of observation times for the raw data is defined as follows: The time interval between adjacent observation times is defined as: For any variable Its standardized form can be expressed as: in, and Variables The mean and standard deviation.

[0035] For locally missing data, this application does not presuppose that all original observation data be forcibly reconstructed into a fixed-time-step regular sequence. Instead, after completing basic cleaning and time reference unification, it retains the true time intervals, asynchronous sampling features, and locally missing data features in the original observation sequence and introduces missing data markers: in, The missing data marker is used to indicate the validity of each input variable at the corresponding observation time and participates in the subsequent joint input construction or state update process to reduce the impact of local missing data on prediction stability.

[0036] In step 3, the input sequence is constructed for continuous-time prediction. After preprocessing, the historical attitude state, environmental excitation, working condition information, time interval information, and missing measurement markers are organized into a unified joint input data vector. The joint input data vector at time ti is: ; The input sequence constructed within the historical time window is as follows: This step not only preserves the information of the ship's attitude state itself, but also preserves the environmental excitation, operating conditions, and the actual sampling time rhythm, so that the subsequent continuous time state update process can evolve the internal state according to the time difference between the actual observation times.

[0037] In step 4, the parametric physical consistency residuals of the ship form are constructed, which is one of the key steps in this invention. First, a universal kinematic consistency residual is constructed: in, Predict pose parameters. This corresponds to the predicted velocity. The kinematic consistency residual is constructed based on the true time interval Δt_i between adjacent observation times, adapting to continuous-time prediction scenarios under irregular sampling conditions. Then, a low-order dynamic consistency residual is constructed, adopting a unified form of "self-term + cross-term," and constraint equations considering the coupling effect between degrees of freedom are established for the three degrees of freedom of heave, roll, and pitch.

[0038] Taking the direction of heave as an example: in, Representing the set of parameters of the target ship respectively The relevant equivalent inertia term, This represents the equivalent damping coefficient. This represents the equivalent recovery coefficient.

[0039] The corresponding cross-damping and cross-restoration coefficients are used to characterize the coupling effect between different degrees of freedom.

[0040] Right end item , , Indicates information from environmental incentives and operating condition information Equivalent external incentives jointly determined.

[0041] The equivalent parameters are determined by both the basic parameters and the ship type correction factor, for example: The same parameterization method can also be used for cross-coupling terms, for example: Other cross terms The corresponding damping and recovery terms can be determined in the same way. Basic parameters, The correction factors corresponding to each cross term are all derived from the target ship parameter set. The determination can be achieved through the parameter mapping module, or by using empirical formulas, system identification, table lookup interpolation, or simulation priors. Based on this, a low-order dynamic consistency residual term is constructed, and the kinematic and dynamic residuals are combined to form the overall physical consistency residual: Among them, These are the weighting coefficients for each constraint term. This is the normalization scaling factor for each residual term, used to eliminate differences in dimensions and orders of magnitude among different residual terms.

[0042] In addition, the average physical consistency loss for historical windows can be defined: Used to measure the degree to which the overall prediction results violate physical consistency constraints.

[0043] In step 5, offline training and parameter determination are performed. A training sample set is constructed using the obtained joint input sequence, the ship type parameterized physical consistency residuals, and the corresponding real future attitude samples. At time [time value missing]... The training input at this location is: The corresponding training output is: The training samples consist of a sequence of "current joint input - future window true output," enabling direct multi-step prediction for future time windows. The total loss function is: in, The mean squared error can be used to represent the data error loss between the predicted and the actual values. The physical consistency loss is composed of kinematic consistency residuals, low-order dynamic consistency residuals, and coupling residuals. To correct the loss between the output and the true value; These are the weighting coefficients.

[0044] The model parameters are iteratively updated using optimization methods such as gradient descent, Adam, and backpropagation until a preset stopping condition is met. After training, the continuous-time state updates of the model parameters, output mapping parameters, and related parameter mapping module parameters are saved. If some equivalent parameters, correction coefficients, or activation mappings are not trained using neural networks, the parameters can also be determined through system identification, empirical formulas, lookup table interpolation, or simulation priors.

[0045] In step six, an initial prediction is made based on a continuous-time state update mechanism, which is one of the core steps of this invention. First, a joint input for the prediction process is established: Its vector representation is: Let the internal state of the model at time ti be h(ti), which represents the comprehensive memory result of historical attitude observations, environmental excitations, operating conditions, time rhythms, physical constraints, and target ship type information.

[0046] The continuous-time state update process can be represented as: For ease of engineering implementation, it can be written in the following form: in, This represents the function of internal state change, and the update quantity is affected by the actual time interval. Modulation: When the sampling interval is short, the state change is small; when the sampling interval is long, the state change increases accordingly. If a continuous-time dynamic expression is used, the internal state satisfies the differential equation: Based on the time interval between actual observation times The continuous dynamics described above are solved by numerical integration or discrete analysis to complete the state update.

[0047] The continuous-time state update function is preferably implemented using a continuous-time cyclic update structure that explicitly depends on the actual time interval. Alternatively, it can be implemented using a liquid neural network, a liquid time constant network, or other continuous-time dynamic networks. This involves obtaining the internal state at the current moment. Then, the output mapping process can be expressed as: in, To output the mapping function, This indicates the time corresponding to the current moment. The initial prediction result sequence for the future prediction window is in the following form: The output vector is represented as: This step does not use point-by-point recursive prediction, but instead directly outputs the attitude results of multiple target times within the future prediction window based on the current internal state. This can reduce the impact of the gradual propagation and accumulation of errors in recursive prediction.

[0048] In step 7, the prediction results are corrected based on the physical consistency residuals. First, the physical residuals corresponding to the initial prediction results are calculated: When the physical consistency residual does not exceed the preset threshold, the initial prediction result can be output directly or only a minor correction can be performed. When it exceeds the preset threshold, output correction and / or internal state feedback correction are performed.

[0049] Define the threshold as , κ is determined based on training set statistics and is a hyperparameter (e.g., 2 or 3). When the initial prediction result is displayed, it is output directly or slightly modified; when At that time, output correction and / or internal state feedback correction are performed.

[0050] The output correction is represented as: in, For correction factors, This is the revised prediction result.

[0051] Internal state feedback correction is represented as: in, For feedback coefficients, This is the deviation mapping function. The corrected internal state is used for state initialization in the next prediction window. The hull-type parameterized physical consistency residual has a dual role: in the offline training phase, it participates in model parameter optimization as a physical consistency constraint; in the online prediction phase, it serves as the basis for prediction result verification and feedback correction. This dual role can suppress error accumulation and drift in the multi-step recursion process.

[0052] In step 8, the prediction results are evaluated and output. Let the corrected prediction result sequence be: Define comprehensive evaluation indicators: in, This represents the prediction error risk estimate, which is derived from historical validation error statistics, model output equations, etc. Represents the physical consistency residual term. These are the weighting coefficients.

[0053] Based on preset threshold Determine the usability of the results: in, This indicates that the prediction results are available. This indicates that the prediction result should trigger prompts, warnings, or related compensation and control actions.

[0054] Furthermore, based on the prediction results and evaluation indicators, physical consistency assessment results, result availability judgment results, abnormal prompts or early warning information, operation threshold comparison results, and compensation or control trigger signals can be generated, transforming the prediction results from simple numerical output into engineering application information that can directly serve ship motion simulation, wave compensation auxiliary control, offshore operation safety assessment, and status monitoring.

[0055] Taking the berthing operation scenario of a wind power maintenance vessel as an example, the method and system provided in this application can collect ship attitude data (from IMU), environmental excitation data (from wave radar and anemometer), and operating condition data (from shipborne operation monitoring equipment) in real time. Due to the harsh marine environment, sensor data may have irregular sampling intervals or partial missing data. The method in this application can retain these irregular features and does not force resampling.

[0056] The system constructs a parameterized physical consistency residual based on the target ship's parameter set. It then predicts the ship's heave, roll, and pitch attitudes within the next 0-5 seconds using a continuous-time state update model. The model explicitly uses real time intervals to drive state updates. Initial prediction results are then input into the physical consistency residual for verification. If the physical residual exceeds a preset threshold, the system automatically performs output correction and feeds back the deviation information to the state update model to adjust subsequent predictions.

[0057] After comprehensive evaluation, the revised prediction results are output to the wave compensation platform's control system if the availability determination is passed; otherwise, an early warning signal is triggered to alert the operators. Through the aforementioned closed-loop prediction mechanism, this application can provide accurate, stable, and physically reasonable forward-looking information on ship attitude for berthing compensation control, effectively reducing the risks of berthing operations.

[0058] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0059] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0060] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0061] These computer program instructions can also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting the continuous short-term attitude of a ship under irregular sampling, characterized in that, Includes the following steps: Acquire historical observation data of the ship; the historical observation data is constructed according to the original sampling time, and retains the true time interval information between adjacent observation times and the missing measurement mark information; Based on ship kinematics and low-order dynamics, and according to the parameter set of the target ship, a parameterized physical consistency residual for the ship form is constructed. The historical observation data, the real time interval information, the missing measurement marker information, and the ship type parameterized physical consistency residual are input into the continuous time state update model. The continuous time state update model uses the real time interval information to drive the internal state evolution and generate the initial attitude prediction result within the future prediction window. The initial attitude prediction result is verified based on the parametric physical consistency residual of the ship type. If the verification passes, the initial attitude prediction result is directly output. If the verification fails, the initial attitude prediction result is corrected, and the correction information is fed back to the continuous time state update model to obtain the final attitude prediction result.

2. The method according to claim 1, characterized in that, The acquisition of historical observation data of ships specifically includes: Collect raw data including ship attitude, environmental excitation, and operating conditions; Outlier processing, unit unification, and time reference unification are performed on the collected raw data to obtain unified data; The true time interval between adjacent observation times in the unified data is retained, and a missing measurement marker is generated to indicate the validity of each input variable at the corresponding observation time. The processed data is constructed into a joint input sequence according to the original sampling time, and resampling is not forced into a regular sequence with a fixed time step.

3. The method according to claim 1, characterized in that, The constructed ship hull parameterized physical consistency residuals specifically include: Based on the fundamental kinematic relationship between predicted displacement and predicted velocity, and based on the actual time interval between adjacent observation times, a kinematic consistency residual is constructed. Based on the inertial, damping, and recovery terms corresponding to each degree of freedom, as well as the cross-inertial, cross-damping, and cross-recovery terms between different degrees of freedom, a low-order dynamic consistency residual is constructed. Obtain the parameter set of the target ship, which includes the ship's length, beam, draft, displacement, initial metacentric height, and moment of inertia; The parameter mapping module maps the parameter set of the target ship to the equivalent parameters corresponding to the inertia term, the damping term, the recovery term, and each cross term, thus obtaining the parameterized physical consistency residual of the ship type.

4. The method according to claim 1, characterized in that, The process of driving internal state evolution using the real time interval information by the continuous-time state update model specifically includes: Define internal state variables, which are used to compress and express historical attitude observations, environmental excitations, operating conditions, time rhythms, physical constraints, and target ship type information. The internal state variables are updated using a continuous-time state update function. The continuous-time state update function depends on the actual time interval between adjacent observation times. When the sampling interval is short, the state change is small, and when the sampling interval is long, the state change increases accordingly. The continuous-time state update function adopts the following discrete update form: in, Indicates time The internal state, Indicates adjacent observation times and The actual time interval between them Indicates time Input features, This represents the function that changes the internal state.

5. The method according to claim 4, characterized in that, The initial attitude prediction results within the generated future prediction window specifically include: Based on the updated internal state variables, the attitude prediction results for multiple target times within the future prediction window are generated at once through the output mapping function; The output mapping function is expressed as: Where G() is the output mapping function, This indicates the time corresponding to the current moment. The sequence of initial attitude prediction results for multiple target times within the future prediction window.

6. The method according to claim 1, characterized in that, The initial attitude prediction result is verified based on the parametric physical consistency residual of the ship type. If the verification fails, the initial attitude prediction result is corrected, and the correction information is fed back to the continuous-time state update model. Specifically, this includes: Calculate the physical residual corresponding to the initial attitude prediction result to obtain the residual norm; The residual norm is compared with a preset threshold. When the residual norm does not exceed the preset threshold, the initial attitude prediction result is output as the final attitude prediction result. When the residual norm exceeds the preset threshold, output correction and internal state feedback correction are performed. The output correction is expressed as: in, This is the initial prediction result. For correction factors, For deviation mapping function, For physical residuals; The internal state feedback correction is expressed as: in, For feedback coefficients, For functions mapped to the state space, the modified internal state is used for state initialization of the next prediction window.

7. The method according to claim 1, characterized in that, After obtaining the final attitude prediction result, the process also includes a result evaluation and output step, specifically including: The final attitude prediction results are subjected to physical consistency assessment and usability determination to generate comprehensive evaluation indicators; The comprehensive evaluation index is compared with a preset threshold to determine whether the prediction result is usable. When the prediction result is unavailable, generate an error message, warning message or control trigger signal; Output the final attitude prediction result and the comprehensive evaluation index.

8. A system for predicting the continuous time of short-term ship attitude under irregular sampling, characterized in that, include: The data acquisition module is used to acquire historical observation data of the ship; The historical observation data is organized according to the original sampling time, and retains the true time interval information between adjacent observation times and the missing measurement mark information; The physical consistency residual construction module is used to construct parametric physical consistency residuals of the ship type based on the ship's kinematic relationships and low-order dynamic relationships, and according to the parameter set of the target ship. The continuous-time state update prediction module is used to input the historical observation data, the real time interval information, the missing measurement marker information and the ship type parameterized physical consistency residual into the continuous-time state update model. The continuous-time state update model uses the real time interval information to drive the internal state evolution and generate the initial attitude prediction result within the future prediction window. The residual verification and feedback correction module is used to verify the initial attitude prediction result based on the parametric physical consistency residual of the ship type. When the verification fails, the initial attitude prediction result is corrected and the correction information is fed back to the continuous time state update model to obtain the final attitude prediction result. The result evaluation and output module is used to evaluate and output the final attitude prediction result.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1 to 7.

Images