A very short-term ship motion prediction method, device, medium and product

Abstract

This application discloses a method, device, medium, and product for very short-term ship motion prediction, relating to the field of data processing. The method includes: acquiring ship motion parameters, environmental parameters, and wave spectrum parameters within a very short period to construct a time series; performing dynamic normalization processing on the time series to obtain a dynamically normalized sequence; constructing three sets of convolutional kernels at different scales and combining them with a ship motion attention mechanism and a bidirectional LSTM network to construct an MSA-CNN-LSTM model; and using the MSA-CNN-LSTM model, obtaining the ship motion prediction result based on the dynamically normalized sequence. This application can achieve high-precision prediction at different time scales, thereby improving the safety of autonomous take-off and landing of helicopters under complex sea conditions and the reliability of mission execution.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to a method, device, medium, and product for very short-term ship motion forecasting. Background Technology

[0002] Shipborne helicopters are helicopters deployed on ships, with their primary takeoff and landing platform being the ship itself, and their main operational range being the ocean and land adjacent to it—significantly different from traditional land-based helicopters. Since the advent and large-scale deployment of helicopters, their unique hovering and vertical takeoff and landing capabilities have allowed them to better integrate with the limited deck space on most ships, thus giving rise to shipborne helicopters, which have gradually developed and improved. Shipborne helicopters can perform a variety of missions at sea, and their development has primarily focused on multi-functionality and enhanced survivability. In recent years, with increased emphasis on and development of the ocean, high-performance shipborne helicopters designed to complement increasingly modern ships have emerged, and the demand for safe and efficient shipboard operations in various environments is also rising.

[0003] Unlike land-based helicopters, shipborne helicopters face more diverse and severe challenges when taking off and landing on ships, constrained by complex and ever-changing weather conditions. The unstable marine atmospheric environment, the six-degree-of-freedom motion of the ship under the influence of wind and waves, the geometric limitations of the ship's superstructure and deck, and the unsteady airflow generated by the coupling effect between the ship and waves all significantly increase the difficulty of shipborne helicopter takeoffs and landings. This difficulty is not only borne by the helicopter's rotor and airframe components but also a test of the pilot's operational load. Especially in high sea states, the ship experiences large-amplitude and long-period rolling motions under the excitation of wind and waves, leading to increased unsteady characteristics of its wake field and further complicating takeoffs and landings. Ships experience six-degree-of-freedom rolling motions under the combined influence of wind, waves, and currents, with roll, pitch, and heave being particularly severe. The swaying motion of a ship itself causes the deck position for helicopter take-off and landing to undergo a large-amplitude periodic movement. At the same time, this violent swaying motion has a stronger effect on the unsteady fluctuations of the ship's wake and the aerodynamic interference effect on helicopters.

[0004] In recent years, research has emerged on the characteristics of wake fields in ship rolling motion and their impact on helicopters. For example, the PIV (Pilot-Induced Variants) experiment was used to study the SFS2 wake field under pitching conditions. This method focuses on the influence of parameters such as the ship's pitch angle and pitch frequency on the turbulence distribution and period of the flow field, finding that the ship's pitching motion and attitude significantly affect the formation and development of the wake. Numerical simulations of the full-scale ONRT (On-the-Air Test Reactor) ship wake field considering atmospheric turbulence, ship motion, and wave suction effects were performed. This study analyzed the influence of these factors by decomposing the velocity field and compared it with previous scaled-down wind tunnel experiments under uniform wind conditions. A comparison with MERGEFORMAT demonstrated that the impact of ship motion is the most significant. The study investigated the effects of ONRT ship wake and helicopter control under sea states 3 and 6, considering waves and motion. The study incorporated a nested mesh method for the Black Hawk helicopter rotor, tail rotor, and fuselage. Two-way coupled numerical simulations were performed to study the changes in thrust, pitch, and roll moments of hovering helicopters over time under different wind, wave, and ship motion parameters. The results showed that the helicopter's aerodynamic parameters also fluctuate periodically with the periodic ship wake field, and the impact is more significant under high sea states.

[0005] Other studies have focused on air wake and aerodynamic interference between ships and helicopters during ship motion. For example, using dynamic actuators, a scaled-down model of an amphibious assault ship was measured and displayed in a wind tunnel under rolling conditions. The results showed that rolling alters the deck flow field structure, and the rotor's rolling torque also changes with the ship's rolling. Further research was conducted on the sinusoidal pitching of a ship model in a wind tunnel, varying the pitch angle and period. Changes in the ship's flow field under headwind and port wind conditions revealed a hysteresis effect in the flow field during ship motion and interference with the aerodynamic loads of helicopters taking off and landing at the bow. Numerical simulations were performed on the flow field of a frigate at a certain tilt angle and the aerodynamic loads of different helicopter landing paths, suggesting that helicopters should adjust their landing paths according to different ship tilt states. Numerical simulations were performed on the flow field of the SFS2 frigate under different pitching conditions, including scenarios such as sudden onset of pitching and a halving of the pitch period, analyzing changes in the vortex structure above the deck and the vertical velocity.

[0006] Against this backdrop, ship motion prediction has become a crucial link in ensuring the safe take-off and landing of helicopters in high sea states. Predicting the characteristics of ship motion under various sea states, identifying suitable resting periods for helicopter landing, and enabling the judgment and prediction of whether a helicopter can successfully land under high sea states are all essential. Among these, ultra-short-term prediction mainly targets changes in ship attitude and motion within a 1–20 second time window, providing crucial information for real-time helicopter landing control and dynamic early warning. Because ship motion is significantly affected by environmental disturbances, its dynamic processes exhibit strong randomness, nonlinearity, and multi-scale characteristics. Therefore, how to achieve high-precision prediction at different time scales is one of the core scientific problems currently facing this field. Furthermore, in existing technologies, prediction methods based on hybrid deep learning models combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) have the following drawbacks: 1. A single convolutional kernel cannot capture both the high-frequency fluctuations and low-frequency trends of ship motion, resulting in incomplete feature capture; 2. Lack of targeted weighting of key motion parameters results in weak resistance to wind, wave, and noise interference; 3. Fixed normalization methods cannot adapt to the differences in data distribution under different sea states, and have poor generalization under high sea states; 4. Unidirectional LSTM can only capture historical data dependencies and cannot utilize the correlation between time series data, resulting in insufficient prediction continuity. Summary of the Invention

[0007] The purpose of this application is to provide a method, equipment, medium, and product for very short-term ship motion prediction, which can achieve high-precision prediction at different time scales, thereby improving the safety of autonomous take-off and landing of helicopters and the reliability of mission execution under complex sea conditions.

[0008] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for very short-term ship motion prediction, including: The ship motion parameters, environmental parameters, and wave spectrum parameters are obtained over a very short period of time to construct a time series; the very short period of time refers to 1 second to 20 seconds. The time series is dynamically normalized to obtain a dynamically normalized sequence; Three sets of convolutional kernels of different scales were constructed, and the MSA-CNN-LSTM model was built by combining the ship motion attention mechanism and the bidirectional LSTM network. The ship motion prediction results are obtained by using the MSA-CNN-LSTM model and the dynamic normalization sequence.

[0009] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the extremely short-term ship motion prediction method provided above.

[0010] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the extremely short-term ship motion prediction method provided above.

[0011] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the extremely short-term ship motion prediction method described above.

[0012] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method, device, medium, and product for very short-term ship motion forecasting. By dynamically normalizing the acquired time series data, it can adapt to the differences in data distribution under different sea states, thus solving the problem of poor generalization under high sea states. By constructing three sets of convolutional kernels at different scales, it can solve the problem that a single convolutional kernel cannot simultaneously capture both high-frequency fluctuations and low-frequency trends in ship motion, resulting in incomplete feature capture. Based on the three sets of convolutional kernels at different scales, an MSA-CNN-LSTM model is constructed by combining a ship motion attention mechanism and a bidirectional LSTM network. This allows for obtaining ship motion forecasting results based on dynamically normalized sequences. This addresses the problems of lacking targeted weighting of key motion parameters, weak resistance to wind and wave noise interference, and the inability of unidirectional LSTM to capture historical data dependencies and utilize the correlation between time series data, resulting in insufficient prediction continuity. Consequently, it can achieve high-precision predictions at different time scales, improving the safety of autonomous take-off and landing of helicopters and the reliability of mission execution under complex sea conditions. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 A flowchart illustrating an extremely short-term ship motion prediction method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the MSA-CNN-LSTM model structure provided in an embodiment of this application; Figure 3A schematic diagram showing the comparison between the predicted and actual oscillation velocity curves of the MSA-CNN-LSTM model within 5 seconds, provided in an embodiment of this application. Figure 4 A schematic diagram showing the comparison between the predicted curve and the actual curve of the sway velocity of the MSA-CNN-LSTM model within 5 seconds provided in an embodiment of this application; Figure 5 A schematic diagram showing the comparison between the predicted curve and the true curve of the roll angular velocity of the MSA-CNN-LSTM model within 5 seconds provided in an embodiment of this application; Figure 6 A schematic diagram showing the comparison between the predicted curve and the actual curve of the bow roll angular velocity within 5 seconds provided by the MSA-CNN-LSTM model in an embodiment of this application; Figure 7 A schematic diagram showing the comparison between the predicted curve and the true curve of the roll angle within 5 seconds of the MSA-CNN-LSTM model provided in an embodiment of this application; Figure 8 A schematic diagram showing the comparison between the predicted and actual oscillation velocity curves of the MSA-CNN-LSTM model provided in an embodiment of this application within 10 seconds; Figure 9 A schematic diagram showing the comparison between the predicted and actual sway velocity curves of the MSA-CNN-LSTM model within 10 seconds, provided in an embodiment of this application. Figure 10 A schematic diagram showing the comparison between the predicted curve and the actual curve of the roll angular velocity of the MSA-CNN-LSTM model within 10 seconds provided in an embodiment of this application; Figure 11 A schematic diagram showing the comparison between the predicted curve and the actual curve of the bow roll angular velocity within 10 seconds provided by the MSA-CNN-LSTM model in an embodiment of this application; Figure 12 A schematic diagram showing the comparison between the predicted curve and the actual curve of the roll angle of the MSA-CNN-LSTM model within 10 seconds provided in an embodiment of this application; Figure 13 A schematic diagram showing the comparison between the predicted and actual oscillation velocity curves of the MSA-CNN-LSTM model provided in an embodiment of this application over 20 seconds; Figure 14 A schematic diagram showing the comparison between the predicted and actual sway velocity curves of the MSA-CNN-LSTM model provided in an embodiment of this application over 20 seconds; Figure 15 A schematic diagram showing the comparison between the predicted curve and the actual curve of the roll angular velocity of the MSA-CNN-LSTM model within 20 seconds provided in an embodiment of this application; Figure 16A schematic diagram showing the comparison between the predicted curve and the actual curve of the bow roll angular velocity within 20 seconds provided by the MSA-CNN-LSTM model in an embodiment of this application; Figure 17 A schematic diagram showing the comparison between the predicted curve and the actual curve of the roll angle of the MSA-CNN-LSTM model provided in an embodiment of this application within 20 seconds; Figure 18 This is a block diagram of an LSTM cell structure provided in an embodiment of this application; Figure 19 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0015] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0016] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] Under high sea states, ship sway motion exhibits strong randomness, nonlinearity, and time-varying characteristics, and its dynamic response directly affects the safety and feasibility of helicopter landing. Helicopter takeoff and landing safety is closely related to the ship's key degrees of freedom, such as roll and pitch, and these motions can change drastically within a second-level timescale. This application addresses the practical need for safe helicopter landings under different sea states by providing a very short-term ship motion prediction method. This method is executed by computer equipment, specifically by a terminal or server alone, or by both. In this embodiment, the method is illustrated using a server as an example. Figure 1 As shown, the method includes: Step 100: Obtain ship motion parameters, environmental parameters, and wave spectrum parameters over a very short period of time to construct a time series. "Very short period of time" refers to 1-20 seconds or 5-20 seconds.

[0018] Among these, ship motion parameters include surge velocity. Sway Velocity Roll rate Yaw Rate Roll Angle Environmental parameters include wind speed. and angle of attack (Including sine components) Cosine component Wave spectrum parameters include wave height. Wave period Wave force response amplitude operator .

[0019] Step 101: Perform dynamic normalization on the time series to obtain a dynamically normalized series.

[0020] Step 102: Construct three sets of convolutional kernels of different scales, and combine them with the ship motion attention mechanism and a bidirectional LSTM network to build the MSA-CNN-LSTM model. The structure of the MSA-CNN-LSTM model is as follows: Figure 2 As shown. MSA stands for Multi-Scale Attention.

[0021] Step 103: Using the MSA-CNN-LSTM model, ship motion prediction results are obtained based on the dynamically normalized sequence.

[0022] By implementing steps 100-103 above, this application can accurately capture the multi-scale characteristics of ship motion, adapt to complex sea conditions, and strengthen the weight of key information to improve the safety and efficiency of helicopter landing.

[0023] In an exemplary embodiment of this application, in order to take into account both the high-frequency fluctuations and low-frequency trends of ship motion and further improve the comprehensiveness of feature capture, the implementation process of step 101 above may include: Step 101-1: Determine the fluctuation coefficient based on wind speed and wave height in the time series. The fluctuation coefficient is expressed as... : .

[0024] Step 101-2: When the volatility coefficient is less than the set value, perform batch normalization on the time series. The formula used for batch normalization (BatchNorm) is: .

[0025] .

[0026] In the formula, This is the batch average. For batch variance, To prevent constants with a denominator of 0, and For learnable parameters, For the i-th parameter in the time series, These are the normalized eigenvalues. This is the final output value after scaling and translation using the learned parameters. The number of samples in each batch.

[0027] Step 101-3: When the volatility coefficient is greater than or equal to the set value, perform instance normalization on the time series. The formula used for instance normalization (InstanceNorm) is: .

[0028] .

[0029] In the formula, This is the instance mean. For instance variance, The length of a single sample sequence. This is the j-th parameter in the time series.

[0030] Step 101-3: Generate a dynamic normalized sequence based on the batch normalization results and the instance normalization results.

[0031] In an exemplary embodiment of this application, in order to capture the multi-periodic features of ship motion, the three sets of convolutional kernels of different scales constructed in step 102 above are as follows: Short-period convolution kernels: used to capture high-frequency fluctuation features (such as instantaneous changes in roll angular velocity).

[0032] Medium-period convolution kernel: used to capture medium-period features (such as the swaying motion corresponding to the wave cycle).

[0033] Long-period convolution kernels: used to capture long-period trends (such as oscillation displacement changes caused by ocean currents).

[0034] Based on the three sets of convolutional kernels of different scales constructed above, the implementation process of step 103 in this application can be described as follows: Step 103-1: Three sets of convolutional kernels of different scales are used to perform multi-scale convolutional feature extraction on the dynamically normalized sequence to obtain fused features. Wherein: Step 11: Perform convolution operations on the dynamically normalized sequence using three sets of convolution kernels of different scales, resulting in: In the formula, , , These are the weight matrices for the three sets of convolutional kernels. , , For bias terms, This is the convolution operation function. , , The features are output by three sets of convolution kernels of different scales.

[0035] Step 12: Pad and concatenate the features output by the three sets of convolutional kernels at different scales to obtain fused features. For example, zero-padding the features output by the three sets of convolutional kernels at different scales to make them the same length. Then, feature splicing is performed, resulting in: .

[0036] In the formula, As a feature of fusion, Zero-filling function, For the fill length, This is the feature splicing function.

[0037] Step 103-2: Multiply the fused features and the attention weights determined based on the ship motion attention mechanism element-wise to obtain the weighted features. Wherein: Step 21: Determine the feature importance score based on the ship's motion physical characteristics, including: .

[0038] In the formula, For feature importance scoring, For fusion features The One dimension, Initial weights (dimensions corresponding to roll angle and heave velocity) Other dimensions ), This represents the maximum value of the fused features.

[0039] Step 22: Obtain the attention weights by normalizing the feature importance scores using the Softmax function, resulting in: .

[0040] In the formula, For the first Attention weights in each dimension, satisfying . For the ship motion attention mechanism, exp() is an exponential function.

[0041] Step 23: Multiply the fused features element-wise with the attention weights to obtain the weighted features, which are: .

[0042] In the formula, , These are weighted features.

[0043] Step 103-3: Use a bidirectional LSTM network to obtain output features based on weighted features.

[0044] The bidirectional LSTM network includes forward LSTM and backward LSTM, with 128 hidden layer neurons and 2 layers.

[0045] The state update formula for a forward LSTM is: .

[0046] The state update formula for the reverse LSTM (time step reverse traversal) is: .

[0047] In the formula, , , These are the input gate, forget gate, and output gate, respectively. This refers to the cellular state (i.e., the memory state). , These are the hidden states of the forward and reverse LSTM, respectively. It is the Sigmoid activation function. For element-wise multiplication, This is the weight matrix. This is a bias term. Let be the input features at time t. For gating units, Let T be the hidden state vector at different times, where T is the extremely short-term state.

[0048] To mitigate gradient vanishing, a residual connection is made between the output features and weighted features of the bidirectional LSTM network. .

[0049] In the formula, This is a vector concatenation operation. For output features.

[0050] Step 103-4: Perform dropout regularization and activation processing on the output features to obtain the ship motion prediction results. Wherein: Step 41: Perform Dropout processing on the output features to prevent overfitting, resulting in the processed features: .

[0051] In the formula, This represents the neuron inactivation probability, which is the probability of randomly setting the output of 20% of neurons to 0. To process features.

[0052] Step 42: Activate the features using the ReLU function to obtain the activated features, which are: .

[0053] In the formula, These are the features after activation.

[0054] Step 43: Map the activated features to the predicted degrees of freedom of ship motion through a fully connected layer to obtain the ship motion prediction result, which is: .

[0055] In the formula, This is the weight matrix of the fully connected layer. For bias terms, These are the predicted values ​​for 7 key parameters.

[0056] In an exemplary embodiment of this application, in order to further improve the accuracy and robustness of the MSA-CNN-LSTM model constructed in step 102, this application uses mean squared error (MSE) as the loss function of the MSA-CNN-LSTM model, as follows: .

[0057] In the formula, To predict the time domain length (5-20 seconds). for Time of the first The actual values ​​of each parameter For the corresponding predicted value, This represents the loss function value.

[0058] Based on the above loss function, the construction method of each dataset during the actual training process of the model is described as follows: The lengths of the collected parameter sequences (i.e., time series samples) are standardized. Let the original motion time series be... The wave cycle sequence is By padding with leading zeros, the length of all sequences is made uniform. To obtain the standardized sequence and .

[0059] The standardized sequence is generated by an open-loop controller, executing different maneuvers such as turning and circling. In addition to the standardized sequence sent to the actuators, assuming wind measurement data including intensity and angle of attack is available, the unknown systems to be identified include... =6 inputs That is, angle of attack ( and ), wind speed ( ), propeller speed ( ) and propeller angle (left side) and the right side The predicted output is... =5 system states That is, two-dimensional velocity (oscillation) and sweeping ), roll angular velocity Bow roll rate and roll angle Dataset The dataset includes 96 hours of regular samples and 29 hours of out-of-distribution (OOD) samples, with measurements taken once per second. The regular samples comprise 60% of the training set. 10% of the validation set and 30% of the test set Out-of-distribution samples For evaluation purposes only, it has a wider propeller speed range and more frequent rudder angle changes.

[0060] Based on the dataset constructed above, this application employs a stochastic gradient descent (SGD) optimizer to minimize the loss function, with the initializer having a length of Train for 400 epochs on a sequence with a learning rate of 50, using the same hyperparameters as the predictor. The training hyperparameters are set as follows: learning rate. Batch size Training cycle Sequence length .

[0061] In one exemplary embodiment of this application, in order to evaluate the prediction performance of the analysis model, three prediction evaluation metrics are selected to evaluate the ultra-short-term model (i.e., the MSA-CNN-LSTM model).

[0062] Mean Absolute Error ), which measures the average absolute deviation between predicted and actual values, is defined as: .

[0063] In the formula, This is the actual value. For predicted values, Let T be the number of future times.

[0064] Mean Squared Error The squared mean of the prediction error is defined as: .

[0065] Root Mean Squared Error The standard deviation of the prediction error, which measures the overall prediction accuracy and the dispersion of the error distribution of the model (i.e., the MSA-CNN-LSTM model), is defined as: .

[0066] The above three indicators are all general evaluation indicators in the field of time series forecasting and have a certain degree of universality.

[0067] Multiple working conditions were randomly selected from a large dataset to verify the prediction accuracy of the proposed model (i.e., the MSA-CNN-LSTM model) at 5 seconds, 10 seconds, and 20 seconds. Specific experimental results are as follows: Figures 3-17 As shown in Tables 1-3, the results demonstrate that the MSA-CNN-LSTM model is highly feasible for ultra-short-term ship motion prediction. It accurately predicts ship motion, thus providing precise timing for helicopter landings during the ship's resting period, validating its effectiveness and feasibility in ship resting period prediction. This also proves the effectiveness of deep learning technology in processing complex nonlinear time series data and its applicability in ship motion and resting period prediction. It provides real-time landing window warnings for helicopter pilots, assisting them in completing landing operations during the ship's stable resting period, thereby improving landing efficiency and pilot safety. Figures 3-17 In this text, sway speed, roll speed, pitch speed, and yaw rate are all expressed in English. "true" represents the actual value, "prediction" represents the predicted value, and "time-step" represents the time step.

[0068] Table 1. Root mean square error of MSA-CNN-LSTM model predictions within 5 seconds

[0069] Table 2. Root mean square error of MSA-CNN-LSTM model predictions within 10 seconds.

[0070] Table 3. Root mean square error of MSA-CNN-LSTM model predictions within 20 seconds.

[0071] Based on the above description, the method provided in this application has the advantages of high computational efficiency and high prediction accuracy. Its ultra-short-term ship motion prediction mainly targets the changes in the ship's motion state over the next 5-20 seconds, aiming to provide real-time motion trend information for helicopter flight control systems and shipboard auxiliary landing systems. Within this timescale, the ship's sway response is mainly determined by current and recent wave excitations, exhibiting significant temporal continuity and local stationarity. Traditional statistical modeling methods (such as autoregressive AR and Kalman filtering) struggle to fully capture the nonlinear and phase-coupling characteristics of ship attitude signals, resulting in limited prediction accuracy. To address these issues, this application employs a long short-term memory recurrent neural network for ultra-short-term ship motion prediction.

[0072] The LSTM network structure captures the dynamic correlation between consecutive time steps in sequence data through its recurrent structure. It possesses memory capabilities and temporal feature extraction abilities, and can retain historical state information in the time series through its internal recurrent structure, thus effectively modeling the dynamic evolution of ships over ultra-short time steps. The LSTM network structure diagram is shown below. Figure 18 As shown, LSTM adds a memory state unit to the hidden layer neural nodes to store past information, and uses three gating structures (input gate, forget gate, and output gate) to control the forgetting and updating of historical information.

[0073] The LSTM network updates its time state according to the following rules:

[0074] in , , , These represent the weight matrices of each connection layer. , , , These represent the bias terms of each connection layer. This is the sigmoid function. The input at the current time step. The hidden state of the previous moment and the memory state of the previous moment. Passing through the above three gating units in sequence , and Perform calculations to obtain the hidden state at the current time. and memory state The weight matrix, hidden state, and memory state within the gated unit are continuously updated as input for the next time step, thereby enabling training and learning of the sequence information.

[0075] By constructing a multi-scale convolutional feature extraction module using three sets of convolutional kernels of different sizes, the high-frequency fluctuations, medium-period features, and long-period trends of ship motion can be fully captured, improving the completeness of feature capture by 30%.

[0076] The ship motion attention mechanism is based on dynamically weighted key parameters of physical characteristics, which significantly enhances the ability to resist wind, waves and noise interference and improves the recognition accuracy during rest periods by 25%-30%.

[0077] The dynamic normalization module adaptively switches the normalization mode according to the sea state fluctuation coefficient, which improves the generalization of the model under sea states 2-6 and reduces the prediction error by 15%-20% under high sea states.

[0078] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 19 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores very short-term ship motion prediction data. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a very short-term ship motion prediction method.

[0079] Those skilled in the art will understand that Figure 19 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include, but are not limited to, the following: Figure 19 The diagram shows more or fewer components, or combinations of certain components, or different component arrangements.

[0080] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0081] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0082] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0083] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0084] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (RRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0085] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0086] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0087] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting very short-term ship motion, characterized in that, include: Obtain ship motion parameters, environmental parameters, and wave spectrum parameters over a very short period of time to construct a time series. The term "extremely short period" refers to 1 to 20 seconds. The time series is dynamically normalized to obtain a dynamically normalized sequence; Three sets of convolutional kernels of different scales were constructed, and the MSA-CNN-LSTM model was built by combining the ship motion attention mechanism and the bidirectional LSTM network. The ship motion prediction results are obtained by using the MSA-CNN-LSTM model and the dynamic normalization sequence.

2. The method for very short-term ship motion prediction according to claim 1, characterized in that, The time series is dynamically normalized to obtain a dynamically normalized sequence, including: The fluctuation coefficient is determined based on the wind speed and wave height in the time series. When the fluctuation coefficient is less than the set value, the time series is subjected to batch normalization. When the fluctuation coefficient is greater than or equal to the set value, the time series is subjected to instance normalization. The dynamic normalized sequence is generated based on the batch normalization result and the instance normalization result.

3. The method for very short-term ship motion prediction according to claim 1, characterized in that, Using the MSA-CNN-LSTM model, ship motion prediction results are obtained based on the dynamically normalized sequence, including: Three sets of convolutional kernels of different scales are used to perform multi-scale convolutional feature extraction on the dynamically normalized sequence to obtain fused features; The fused features and the attention weights determined based on the ship motion attention mechanism are multiplied element by element to obtain the weighted features; The bidirectional LSTM network is used to obtain output features based on the weighted features; The output features are subjected to dropout regularization and activation processing to obtain the ship motion prediction results.

4. The method for very short-term ship motion prediction according to claim 3, characterized in that, Three sets of convolutional kernels of different scales are used to perform multi-scale convolutional feature extraction on the dynamically normalized sequence to obtain fused features, including: The dynamically normalized sequence is convolved using three sets of convolution kernels of different scales. The features output by three sets of convolution kernels of different scales are filled and concatenated to obtain the fused features.

5. The method for very short-term ship motion prediction according to claim 3, characterized in that, The process of determining attention weights based on the ship motion attention mechanism includes: Feature importance scores are determined based on the physical characteristics of ship motion; The attention weights are obtained by normalizing the feature importance scores using the Softmax function.

6. The method for very short-term ship motion prediction according to claim 1, characterized in that, The output features are subjected to dropout regularization and activation processing to obtain the ship motion prediction results, including: The output features are subjected to Dropout processing to obtain the processed features; The processed features are activated using the ReLU function to obtain the activated features; The activated features are mapped to predicted ship motion degrees of freedom through a fully connected layer to obtain the ship motion prediction result.

7. The method for very short-term ship motion prediction according to claim 1, characterized in that, Mean squared error is used as the loss function for the MSA-CNN-LSTM model.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the very short-term ship motion prediction method according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method for very short-term ship motion prediction as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the method for very short-term ship motion prediction as described in any one of claims 1-7.

Images