Submarine cable construction ship extremely-short-term motion response prediction method and system

Abstract

The invention discloses a submarine cable construction ship extremely-short-term motion response prediction method and system, and relates to the technical field of ship motion control in ocean engineering. Comprising the following steps: acquiring six-degree-of-freedom time sequence data based on a ship-borne inertial measurement unit of a submarine cable construction ship, and constructing a data set; the data set is preprocessed; constructing a model CNN-RSBU-BiLSTM (Convolutional Residual Shrinkage-Bidirectional Long Short Term Memory) with fusion of the convolutional residual shrinkage; the model CNN-RSBU-BiLSTM is trained by means of the data set obtained after preprocessing is carried out on the model CNN-RSBU-BiLSTM; and inputting the six-degree-of-freedom time sequence data acquired and preprocessed in real time into the trained model, and outputting an extremely short-term six-degree-of-freedom motion response prediction result of the submarine cable construction ship. According to the method, the multi-scale feature extraction capability of the one-dimensional convolutional neural network, the noise reduction enhancement effect of the residual shrinkage building unit and the digging capability of the bidirectional long-short-term memory network on the time sequence dependency relationship are combined, so that accurate prediction of the extremely-short-term six-degree-of-freedom motion response of the submarine cable construction ship is realized, and the prediction accuracy is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of ship motion control technology in marine engineering, specifically to a method and system for predicting the very short-term motion response of a submarine cable construction vessel. Background Technology

[0002] With the rapid development of offshore wind power and other marine energy resources, submarine cables, as the core carriers of energy transmission, have become a critical aspect of marine engineering construction in terms of the safety and efficiency of their laying operations. Submarine cable laying vessels, as the core equipment for these operations, must operate continuously in complex and ever-changing sea conditions, inevitably facing interference from the combined effects of wind, waves, and currents, which can easily generate six degrees of freedom motions, including sway, pitch, heave, roll, and bow roll. These motions directly affect the accuracy of submarine cable laying and can even pose a serious threat to construction safety.

[0003] To ensure operational stability, active compensation and control technology for ship motion adjusts propeller thrust in real time through a dynamic positioning system (DP) to counteract hull drift and deflection caused by environmental disturbances. The core prerequisite for this technology to function effectively is the accurate prediction of the construction vessel's extremely short-term motion response: millimeter-level displacement and milliradian-level angle prediction must be achieved within twice the system delay time window, providing a valuable "decision buffer period" for the compensation mechanism.

[0004] Currently, ship motion response prediction technology is mainly based on numerical simulation, data-driven methods, or a hybrid approach combining both. Numerical simulation methods are based on the dynamic equations of ship motion, treating the ship as an elastic or rigid body. By analyzing the forces acting on it under external excitations such as waves, wind, and currents, a mathematical expression for the motion response is established. These equations typically include physical parameters such as ship mass, moment of inertia, damping coefficient, and restoring force coefficient, as well as quantitative models of environmental loads (such as wave forces, wind forces, and current forces). By solving these equations, the changes in displacement, velocity, or acceleration of the ship in six degrees of freedom over time can be obtained, thus enabling the prediction of future motion states. This type of method has a certain generalization ability for new ship types and new sea states, but the equation solving is relatively complex and computationally time-consuming. Furthermore, under complex sea states, such as multi-directional waves and strongly nonlinear waves, the quantification of environmental loads is difficult, and model assumptions are prone to deviating from reality, leading to decreased prediction accuracy. Additionally, the equations for multi-degree-of-freedom coupled motion are complex to solve and have poor real-time performance, making it difficult to meet the requirements of very short-term real-time prediction in the scenario of submarine cable construction vessels.

[0005] Data-driven approaches rely on massive amounts of historical or real-time monitoring data, using algorithms such as machine learning and deep learning to uncover hidden patterns and regularities within the data, constructing a mapping relationship between input and output, and thus achieving motion response prediction. This includes traditional machine learning and deep learning. Its core advantage lies in the fact that it does not rely on complex physical mechanism modeling, but rather completes the prediction task through feature learning from the data itself, making it applicable to ship motion prediction in complex sea conditions. However, traditional deep learning models are highly dependent on data quality. If the data contains too much noise or the sea state range covered by the samples is insufficient, it can easily lead to overfitting or poor generalization ability. Furthermore, the black-box nature of the model makes the physical interpretability of the prediction results weak.

[0006] Hybrid methods are based on the physical laws governing ship motion. They use the output of a physical model (e.g., theoretical predictions based on hydrodynamic equations) and the residuals from actual monitoring data as input to a data-driven model, correcting biases in the physical model through machine learning or deep learning algorithms. Conversely, they use a data-driven model for initial predictions, then optimize the results using constraints on the physical model (e.g., the conservation of motion equations). The core is a collaborative mechanism of "physical modeling + data correction," which retains the explanatory power of physical methods for motion mechanisms while utilizing data-driven methods to handle complex factors such as noise and coupling effects, achieving more accurate predictions. Compared to pure numerical simulation methods, hybrid methods better handle nonlinearities, coupling effects, and noise interference under complex sea conditions, resulting in higher prediction accuracy. Compared to pure data-driven methods, the introduction of physical constraints reduces reliance on massive amounts of data, enhances generalization ability, and yields results that better match actual motion mechanisms, facilitating interpretation and verification in engineering applications. However, this type of method has a complex model structure, requiring simultaneous optimization of physical parameters and data-driven model parameters, which increases the training difficulty. The fusion strategy of physical model and data-driven model (such as weight allocation and error propagation method) has a significant impact on the prediction effect and needs to be customized for specific scenarios, resulting in weak universality.

[0007] In summary, existing ship motion response prediction methods all have their limitations when facing the need for very short-term motion response prediction of submarine cable laying vessels. How to improve the efficiency and accuracy of motion response prediction for construction vessels to adapt to the operational needs under complex sea conditions has become an urgent problem to be solved in this field. Summary of the Invention

[0008] The purpose of this invention is to propose a method and system for predicting the extremely short-term motion response of submarine cable construction vessels, thereby improving the accuracy and stability of the prediction.

[0009] According to a first aspect of the present disclosure, a method for predicting the very short-term motion response of a submarine cable laying vessel is provided, comprising the following steps: A dataset was constructed by acquiring six-degree-of-freedom time-series data using the shipborne inertial measurement unit of a submarine cable construction vessel. The dataset is preprocessed, including time-series windowing and normalization. A fusion model of convolutional residual shrinkage bidirectional long short-term memory network CNN-RSBU-BiLSTM is constructed. The model includes a one-dimensional convolutional neural network CNN residual shrinkage building unit RSBU and a bidirectional long short-term memory network BiLSTM connected in sequence. The model CNN-RSBU-BiLSTM was trained using the preprocessed dataset until the model converged. The real-time collected and preprocessed six-degree-of-freedom time-series data is input into the trained model, and the very short-term six-degree-of-freedom motion response prediction results of the submarine cable construction vessel are output.

[0010] In one embodiment, the temporal windowing process divides the input data into windows containing data from multiple historical moments, uses the data within the window as input samples, and uses the motion response data corresponding to future moments as labels, and generates input-output mapping pairs through a sliding window.

[0011] In one embodiment, a one-dimensional convolutional neural network (CNN) is used to extract multi-scale features from the input six-degree-of-freedom time-series data. The residual shrinkage building unit (RSBU) is used to denoise and enhance the extracted feature data. A bidirectional long short-term memory (BiLSTM) network is used to perform temporal deep learning on the processed features to mine the temporal dependencies.

[0012] In one embodiment, the residual shrinkage building unit (RSBU) includes an SE module. The SE module is used to explicitly model the correlation between neural network channels. It includes four parts: transformation, squeezing, excitation, and scaling, and is implemented as follows: Transformation: Mapping the input X to a feature map , where R represents the set of real numbers, and H, W, and C are the height, width, and number of feature channels of the feature map, respectively; Squeezing: By aggregating information within the same channel in a spatial dimension, a multidimensional statistic focusing on the correlation between channels is obtained. This process is achieved through global average pooling, i.e., given an unweighted feature map... U , along the spatial dimension H × W Multidimensional statistics are obtained after compression. ; Incentive: An incentive mechanism with sigmoid activation is used to perform the incentive operation, capturing inter-channel dependencies and obtaining weights. g； Scaling: The obtained weightsg Compared with unweighted feature maps U Perform channel-by-channel weighting to form the final output. .

[0013] In one embodiment, the residual shrinkage building unit (RSBU) is specifically implemented in the following ways: The six-degree-of-freedom motion features extracted by convolution are standardized and nonlinearly processed by batch normalization layers and ReLU activation function, and then refined into deep features by one-dimensional convolution. After calculating the absolute value of the features, weight coefficients α are generated through global average pooling, fully connected layers, batch normalization layers, ReLU activation function and Sigmoid activation function, and the optimal noise reduction threshold τ is dynamically obtained in combination with the features. The features are processed using a soft threshold based on a noise reduction threshold τ, and then input into the residual module to optimize the feature representation.

[0014] In one embodiment, soft thresholding is implemented using a non-linear function, as follows:

[0015] in, This refers to a noise reduction threshold that is greater than zero. x Features before soft thresholding y This refers to the features after soft thresholding.

[0016] In one embodiment, the model training process includes: dividing the preprocessed dataset into a training set and a test set, setting the batch size and inputting it into the model, outputting the prediction result through forward propagation, calculating the loss function, optimizing the model parameters using the backpropagation algorithm and gradient descent method, and repeating the training until the loss function converges.

[0017] According to a second aspect of the present disclosure, a very short-term motion response prediction system for a submarine cable construction vessel is provided, comprising: The data acquisition module acquires six-degree-of-freedom time-series data based on the shipborne inertial measurement unit of the submarine cable construction vessel and constructs a dataset. The data preprocessing module preprocesses the dataset, including time-series windowing and normalization. The model building module constructs a fusion convolutional residual shrinking bidirectional long short-term memory network model CNN-RSBU-BiLSTM, which includes a one-dimensional convolutional neural network CNN residual shrinking building unit RSBU and a bidirectional long short-term memory network BiLSTM connected in sequence. The model training module trains the CNN-RSBU-BiLSTM model using the preprocessed dataset until the model converges. The prediction output module inputs the real-time collected and preprocessed six-degree-of-freedom time-series data into the trained model and outputs the prediction results of the very short-term six-degree-of-freedom motion response of the submarine cable construction vessel.

[0018] According to a third aspect of the present disclosure, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and running on the memory, wherein the processor executes the program to implement the aforementioned method for predicting the very short-term motion response of a submarine cable construction vessel.

[0019] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the aforementioned method for predicting the very short-term motion response of a submarine cable construction vessel.

[0020] The advantages of the above technical solutions adopted in this invention compared with the prior art are as follows: 1. By constructing a fusion convolutional residual contraction bidirectional long short-term memory network model CNN-RSBU-BiLSTM, combining the multi-scale feature extraction capability of one-dimensional convolutional neural networks, the noise reduction and enhancement effect of residual contraction building units, and the ability of bidirectional long short-term memory networks to mine temporal dependencies, we achieved accurate prediction of the ultra-short-term six-degree-of-freedom motion response of submarine cable construction vessels, effectively improving prediction accuracy.

[0021] 2. The Residual Shrinkage Building Unit (RSBU) integrates three major mechanisms: adaptive threshold calculation, soft threshold processing, and residual structure optimization. It can selectively filter out noise in ship motion data under complex sea conditions while retaining key motion features. This forms a complete noise reduction process of "noise identification → threshold adaptation → accurate filtering → feature retention", enabling the model to maintain stable performance even in strong noise environments.

[0022] 3. During model training, a reasonable dataset partitioning method was adopted, and parameters were optimized by combining backpropagation algorithm and gradient descent method, which enhanced the model's adaptability to different sea conditions and different operating scenarios and reduced the occurrence of overfitting.

[0023] 4. By using data preprocessing methods such as time-series windowing and combining the efficient collaborative work of various modules of the model, the model can quickly process real-time collected data and output prediction results, meeting the real-time requirements of the very short-term motion response prediction of submarine cable construction vessels, and providing timely decision-making basis for vessel dynamic positioning and motion compensation. Attached Figure Description

[0024] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application.

[0025] Figure 1 This is a diagram illustrating the overall framework of the CNN-RSBU-BiLSTM model. Figure 2 This is a diagram illustrating data windowing. Figure 3 This is a schematic diagram of the SE module structure. Figure 4 This is a schematic diagram of the soft threshold function; Figure 5 This is a schematic diagram of the soft threshold derivative; Figure 6 Here is a diagram of the CNN-RSBU-BiLSTM model structure; Figure 7 Flowchart for training a prediction model of the very short-term motion response of a construction vessel. Detailed Implementation

[0026] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0027] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0028] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0029] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and systems according to various embodiments of this disclosure. It should be noted that each block in a flowchart or block diagram may represent a module, segment, or portion of code, which may include one or more executable instructions for implementing the logical functions specified in the various embodiments. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutively represented blocks may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, may be implemented using a dedicated hardware-based system that performs the specified functions or operations, or using a combination of dedicated hardware and computer instructions.

[0030] Example 1: Cable laying operations on submarine cable laying vessels place stringent requirements on hull stability, necessitating the construction of a dynamically stable operating environment. The core of this approach is controlling the hull's motion amplitude using a motion compensation system. To ensure the motion compensation system receives the necessary buffer and anticipation periods, predicting the vessel's extremely short-term motion response becomes crucial. Furthermore, under complex sea conditions, the prediction system must maintain high robustness, accurately capturing complex operating conditions such as abrupt changes in large roll damping and multi-directional wave coupling effects, and outputting uncertainty boundaries in real time to provide a reliable basis for the dynamic switching strategy of the control system. The prediction of the extremely short-term motion response of submarine cable laying vessels is based on historical or real-time data to predict the six degrees of freedom of the vessel within a very short timeframe. Based on the predicted six degrees of freedom data, the control system assesses the hull's motion trend in real time and implements dynamic compensation measures accordingly. This invention integrates the Residual Shrinking Unit (RSBU) into the CNN-BiLSTM network architecture, such as... Figure 1 As shown, the preprocessed data from the shipboard sensors of the construction vessel is used as input to extract and learn data features, thereby enabling the prediction of the very short-term motion response of the submarine cable construction vessel.

[0031] This embodiment provides a method for predicting the very short-term motion response of a submarine cable laying vessel, including the following steps: S1. Acquire six-degree-of-freedom time-series data based on the shipborne inertial measurement unit of the submarine cable construction vessel, and construct a dataset; S2. Preprocess the dataset, including time-series windowing and normalization. Specifically, time-series windowing is used because the degrees of freedom data used in this invention are time-series sequences. Therefore, the data is windowed, and the model input samples contain data from multiple historical time points, such as... Figure 2 As shown in the figure w_len Indicates the window length. w_step Indicates the step size of the sliding window. w_pre To predict the step size, 1 t 2 t , ..., mt Data representing multiple features at a given time. t Indicates the time. Window size represents the amount of sample data, that is, the number of data samples within the window. xt The input is the motion response data at the next time step. Y The "label" is used to input data into the model to obtain predicted values ​​of motion responses in subsequent time periods; the window can be clicked. w_step The step length slides backward, and so on.

[0032] To reduce the impact of the magnitude and unit of the six-degree-of-freedom motion response of the submarine cable construction vessel on the accuracy of data prediction, this invention employs a minimum-maximum normalization (Min-MaxNormalization) data processing method on the six-degree-of-freedom motion characteristic data of the vessel. This method maps the original data in the sample set to the [0,1] interval through a linear transformation. The specific formula is as follows: (1) In the formula, The data is after normalization. The original data, The minimum value of the original data. This represents the maximum value of the original data.

[0033] S3. Construct a fusion convolutional residual contraction bidirectional long short-term memory network model CNN-RSBU-BiLSTM. The model includes a one-dimensional convolutional neural network CNN residual contraction building unit RSBU and a bidirectional long short-term memory network BiLSTM connected in sequence. The one-dimensional convolutional neural network CNN extracts multi-scale features from the input six-degree-of-freedom temporal data. The residual contraction building unit RSBU is used to denoise and enhance the extracted feature data. The bidirectional long short-term memory network BiLSTM is used to perform temporal deep learning on the processed features to explore the temporal dependencies. Specifically, the Residual Shrinking Building Unit (RSBU) includes an SE module. The SE module can explicitly model the correlations between neural network channels. Using techniques such as global average pooling and fully connected layers, it adaptively learns the weights of each channel, enhancing features beneficial to the target task while suppressing information in features that is useless for the mapping relationship. The SE module is easy to use and can be directly embedded into existing network structures without significantly increasing the model's complexity. The SE module mainly consists of four parts: transformation, squeezing, activation, and scaling. Figure 3 As shown, the specific description is as follows: A transformation is a process that, for any given input, changes its shape. X Mapping to feature map ,in R Represents the set of real numbers. H , W , C These represent the height, width, and number of feature channels of the feature map, respectively.

[0034] Compression, by aggregating information within the same channel spatially, eliminates interference from spatial distribution, resulting in a multidimensional statistic focused on the correlation between channels. This process is achieved through global average pooling, given an unweighted feature map. U , along the spatial dimension H ×W Multidimensional statistics are obtained after compression. , of which c The calculation method for each element is as follows: (2) in, The first in the input data i Line number j Column elements, This refers to compression transformation.

[0035] To fully utilize the information aggregated during squeezing, further excitation operations are needed to capture the dependencies between channels. The excitation function must meet two criteria: it must be flexible enough to capture the nonlinear interactions between channels; and it must be able to learn non-exclusive relationships to ensure consideration of multiple channels. A gating mechanism with sigmoid activation is employed. (3) in, and These are the sigmoid and ReLU activation functions, respectively. and These are the parameters for the two fully connected layers; the generated... g That is, to depict U The weights of importance of each channel.

[0036] Scaling, the obtained weights g With unweighted output U Perform channel-by-channel weighting to form the final output. : (4) (5) Noise reduction of the six-DOF time-series data of the submarine cable construction vessel was performed by the Residual Shrinking Building Unit (RSBU). This unit integrates a Squeeze-and-Excitation module (SE module), a Residual module, and a Soft Thresholding Function on top of its basic functional components. By constructing an identity mapping path from the module's inception to subsequent addition steps, RSBU ensures smooth gradient backpropagation during motion response prediction, making shallow network parameter updates more effective. This maintains model training stability when predicting complex multi-DOF coupled motions, effectively mitigating the gradient problem common in deep networks. With the SE module, RSBU can adaptively derive the optimal threshold based on sample features. Simultaneously, by introducing a soft thresholding mechanism, RSBU effectively amplifies information in the input features that significantly contributes to the motion mapping relationship (i.e., six-DOF prediction) while compressing irrelevant or noisy information to zero, thereby effectively filtering noise from the time-series data and significantly improving the accuracy of six-DOF motion prediction.

[0037] Soft thresholding is implemented using nonlinear functions, such as... Figure 4 The principle is as follows: (6) in, This refers to a noise reduction threshold that is greater than zero. x Features before soft thresholding y Features after soft thresholding; Soft thresholding will be located in the range Features within a certain range are set to 0, while features far from 0 are adjusted to shrink towards 0. The soft thresholding function does not directly set all negative feature values ​​to 0, preserving potentially valuable negative feature information. Furthermore, the derivative of the soft thresholding function's output with respect to the input only has two possible values: 0 or 1, preventing gradient vanishing and gradient exploding problems. Figure 5 As shown.

[0038] Residual shrinkage building unit RSBU, etc. Figure 6 As shown, its main functions can be divided into three core parts: 1) First, the six-degree-of-freedom motion features of the construction vessel extracted through convolution operations are standardized in batches using a batch normalization (BN) layer, effectively improving the distribution stability of the feature data and laying the foundation for subsequent feature processing. Simultaneously, the ReLU activation function is used to introduce an activation mechanism into the nonlinear expression of the cable construction vessel's motion features, enhancing the model's ability to fit complex motion relationships and thus deeply exploring the nonlinear pattern features of the six-degree-of-freedom motion. Subsequently, one-dimensional convolution operations are used to further extract and refine the deep features related to the six-degree-of-freedom motion of the construction vessel, thereby more fully exploring the motion law characteristics.

[0039] 2) Absolute values ​​are calculated for the deep features of the six-degree-of-freedom motion to highlight information features that contribute to the motion of the construction vessel. Through global average pooling and flattening, the original features are transformed into new feature representations, which are then processed sequentially through two fully connected layers, a batch normalization layer, and a ReLU activation function. Finally, a weight coefficient α (αϵ[0,1]) is generated using the Sigmoid activation function. This coefficient can be adaptively adjusted according to the actual situation of the construction vessel's six-degree-of-freedom motion features to determine the weight allocation of different motion features in the threshold calculation. Combining α and the construction vessel's motion features, the optimal noise reduction threshold τ for the sample's six-degree-of-freedom motion can be dynamically obtained, achieving accurate identification and adaptive filtering of noise under different construction vessel motion states.

[0040] 3) Based on the obtained noise reduction threshold τ, soft thresholding is performed to achieve effective noise reduction. The processed result is then input into the residual module for further optimization of feature representation. This process not only effectively suppresses noise interference but also preserves the feature information crucial for predicting the six degrees of freedom of motion, ensuring the stability and reliability of the model's prediction accuracy.

[0041] S4. Train the CNN-RSBU-BiLSTM model using the preprocessed dataset, such as... Figure 7 As shown, continue until the model converges; Specifically, the training samples are constructed based on historical or real-time six-DOF time-series data of the submarine cable construction vessel, and input-output mapping pairs are generated through a sliding window. Input features Output labels for six-DOF motion data (Surge, Sway, Heave, Roll, Pitch, Yaw) over H time steps within a historical window. For the future h The motion response at each time step is expressed as: (7) (8) in This is the normalized six-degree-of-freedom vector. Samples are arranged according to step size. w_step Sliding generation preserves the dynamic coupling characteristics between degrees of freedom. Key characteristics of the samples are shown in Table 1. End-to-end motion timing mapping is achieved: a long input window captures low-frequency trends, and a short output window focuses on the extremely short-term dynamic response, supporting the real-time control requirements of the ship's dynamic positioning system (DP).

[0042] Table 1. Structure of the learning sample dataset

[0043] S5. Input the real-time collected and preprocessed six-degree-of-freedom time-series data into the trained model, and output the prediction results of the very short-term six-degree-of-freedom motion response of the submarine cable construction vessel.

[0044] The following three indicator systems are adopted: 1) Mean Absolute Error (MAE) and Mean Squared Error (MSE), which are used to intuitively reflect the degree of deviation between the model's prediction results and the actual values. The smaller the indicator value, the higher the model's prediction accuracy; 2) Coefficient of Determination (R²) 2 R is used to characterize how well a model fits the data. 2 The closer the value is to 1, the better the model fits the data and the more reliable the prediction results. The relevant calculation formula is as follows: (9) (10) (11) in, , and They represent the first The actual value, the predicted value, and the average of the actual values ​​at each time point; n This indicates the length of the time series data set.

[0045] The experimental scheme of this invention was designed using a submarine cable construction vessel as background data. The experiment employed ANSYS / AQWA simulation software to construct a ship motion model and build corresponding experimental datasets. Eight experimental working condition parameters were designed (see Table 2 for details), and an LSTM model was selected as the comparison benchmark. Analysis of the six-degree-of-freedom data corresponding to the eight working conditions obtained from the simulation revealed that 90%... o Side waves have a significant impact on the ship's pitching, rolling, and bowing motions; therefore, 90° was subsequently selected. oThree sets of experiments were designed for side wave sea state conditions: 1) The input window length H was set to 720, the sliding step size L was set to 40, the prediction duration h was 5s, and 10 time steps were contained per second. Sea state 5 to 8 were selected, based on 100% of the training dataset. 2) The experimental window length H was set to 720, the sliding step size L was set to 40, the time resolution was set to 0.1 seconds / step, and the complete dataset was selected for testing with a sample size N (N is 100% of the dataset). The sea state was condition 5, and the prediction duration h (hϵ[5s,25s]) was adjusted. 3) The experimental window length H was set to 720, the sliding step size L was set to 40, the time resolution was set to 0.1 seconds / step, the sea state was fixed at condition 5, the prediction duration was fixed at h=5s, the dataset was divided into five equal parts, and the sample size N was adjusted.

[0046] Table 2 Wave load parameters under experimental conditions

[0047] Experiments show that the proposed method for predicting the ultra-short-term motion response of submarine cable construction vessels, which integrates a convolutional residual shrinking bidirectional long short-term memory network, can adapt to various complex sea conditions for six-degree-of-freedom prediction. Furthermore, it exhibits more accurate and stable prediction characteristics as the prediction duration and sample size change. Table 3 shows the model prediction performance under different sea conditions. Table 4 shows the model prediction performance under different prediction durations, and Table 5 shows the model prediction performance under different sample sizes. It can be seen that the average prediction accuracy of the proposed method is more than 30% higher than that of traditional methods, with a maximum improvement of 67%, significantly enhancing the prediction accuracy of the ultra-short-term motion response of submarine cable construction vessels. The method described in this invention achieves better prediction advantages when the prediction duration is short; for example, when the prediction duration is 5 seconds, the prediction accuracy is more than 50% higher than the comparative algorithm. The method described in this invention has low sensitivity to the sample size, exhibiting more stable prediction performance and demonstrating strong robustness and applicability.

[0048] Table 3. Model prediction performance under different sea states

[0049] Table 4. Model prediction performance under different prediction durations

[0050] Table 5. Model predictability under different sample sizes

[0051] Example 2: This embodiment provides a very short-term motion response prediction system for submarine cable construction vessels, including: The data acquisition module acquires six-degree-of-freedom time-series data based on the shipborne inertial measurement unit of the submarine cable construction vessel and constructs a dataset. The data preprocessing module preprocesses the dataset, including time-series windowing and normalization. The model building module constructs a fusion convolutional residual shrinking bidirectional long short-term memory network model CNN-RSBU-BiLSTM, which includes a one-dimensional convolutional neural network CNN residual shrinking building unit RSBU and a bidirectional long short-term memory network BiLSTM connected in sequence. The model training module trains the CNN-RSBU-BiLSTM model using the preprocessed dataset until the model converges. The prediction output module inputs the real-time collected and preprocessed six-degree-of-freedom time-series data into the trained model and outputs the prediction results of the very short-term six-degree-of-freedom motion response of the submarine cable construction vessel.

[0052] Example 3: An electronic device includes a memory, a processor, and a computer program stored in the memory and running thereon. When the processor executes the program, it implements the aforementioned method for predicting the extremely short-term motion response of a submarine cable-laying vessel, comprising: A dataset was constructed by acquiring six-degree-of-freedom time-series data using the shipborne inertial measurement unit of a submarine cable construction vessel. The dataset is preprocessed, including time-series windowing and normalization. A fusion model of convolutional residual shrinkage bidirectional long short-term memory network CNN-RSBU-BiLSTM is constructed. The model includes a one-dimensional convolutional neural network CNN residual shrinkage building unit RSBU and a bidirectional long short-term memory network BiLSTM connected in sequence. The model CNN-RSBU-BiLSTM was trained using the preprocessed dataset until the model converged. The real-time collected and preprocessed six-degree-of-freedom time-series data is input into the trained model, and the very short-term six-degree-of-freedom motion response prediction results of the submarine cable construction vessel are output.

[0053] Example 4: A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method for predicting the very short-term motion response of a submarine cable construction vessel, comprising: A dataset was constructed by acquiring six-degree-of-freedom time-series data using the shipborne inertial measurement unit of a submarine cable construction vessel. The dataset is preprocessed, including time-series windowing and normalization. A fusion model of convolutional residual shrinkage bidirectional long short-term memory network CNN-RSBU-BiLSTM is constructed. The model includes a one-dimensional convolutional neural network CNN residual shrinkage building unit RSBU and a bidirectional long short-term memory network BiLSTM connected in sequence. The model CNN-RSBU-BiLSTM was trained using the preprocessed dataset until the model converged. The real-time collected and preprocessed six-degree-of-freedom time-series data is input into the trained model, and the very short-term six-degree-of-freedom motion response prediction results of the submarine cable construction vessel are output.

[0054] Those skilled in the art will understand that the modules or steps described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, which can then be stored in a storage device for execution by a computer device. Alternatively, they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. This disclosure is not limited to any particular combination of hardware and software.

[0055] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0056] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A method for predicting the very short-term motion response of a submarine cable laying vessel, characterized in that, Includes the following steps: A dataset was constructed by acquiring six-degree-of-freedom time-series data using the shipborne inertial measurement unit of a submarine cable construction vessel. The dataset is preprocessed, including time-series windowing and normalization. A fusion model of convolutional residual shrinkage bidirectional long short-term memory network CNN-RSBU-BiLSTM is constructed. The model includes a one-dimensional convolutional neural network CNN residual shrinkage building unit RSBU and a bidirectional long short-term memory network BiLSTM connected in sequence. The model CNN-RSBU-BiLSTM was trained using the preprocessed dataset until the model converged. The real-time collected and preprocessed six-degree-of-freedom time-series data is input into the trained model, and the very short-term six-degree-of-freedom motion response prediction results of the submarine cable construction vessel are output.

2. The method for predicting the very short-term motion response of a submarine cable construction vessel according to claim 1, characterized in that, The temporal windowing process divides the input data into windows containing data from multiple historical moments. The data within each window is used as the input sample, and the motion response data corresponding to future moments is used as the label. Input-output mapping pairs are generated through a sliding window.

3. The method for predicting the very short-term motion response of a submarine cable laying vessel according to claim 1, characterized in that, Multi-scale feature extraction is performed on the input six-degree-of-freedom time-series data using a one-dimensional convolutional neural network (CNN). The extracted feature data is then denoised and enhanced using residual shrinkage building units (RSBU). Finally, a bidirectional long short-term memory (BiLSTM) network is used to perform temporal deep learning on the processed features to uncover the temporal dependencies.

4. The method for predicting the very short-term motion response of a submarine cable laying vessel according to claim 1, characterized in that, The Residual Shrinkage Building Unit (RSBU) includes an SE module, which is used to explicitly model the correlation between neural network channels. The SE module comprises four parts: transformation, squeezing, excitation, and scaling, and is implemented as follows: Transformation: Mapping the input X to a feature map , where R represents the set of real numbers, and H, W, and C are the height, width, and number of feature channels of the feature map, respectively; Squeezing: By aggregating information within the same channel in a spatial dimension, a multidimensional statistic focusing on the correlation between channels is obtained. This process is achieved through global average pooling, i.e., given an unweighted feature map... U , along the spatial dimension H × W Multidimensional statistics are obtained after compression. ; Incentive: An incentive mechanism with sigmoid activation is used to perform the incentive operation, capturing inter-channel dependencies and obtaining weights. g； Scaling: The obtained weights g Compared with unweighted feature maps U Perform channel-by-channel weighting to form the final output. .

5. The method for predicting the very short-term motion response of a submarine cable laying vessel according to claim 1, characterized in that, The specific implementation methods of the residual shrinkage building unit (RSBU) include: The six-degree-of-freedom motion features extracted by convolution are standardized and nonlinearly processed by batch normalization layers and ReLU activation function, and then refined into deep features by one-dimensional convolution. After calculating the absolute value of the features, weight coefficients α are generated through global average pooling, fully connected layers, batch normalization layers, ReLU activation function and Sigmoid activation function, and the optimal noise reduction threshold τ is dynamically obtained in combination with the features. The features are processed using a soft threshold based on a noise reduction threshold τ, and then input into the residual module to optimize the feature representation.

6. The method for predicting the very short-term motion response of a submarine cable construction vessel according to claim 5, characterized in that, Soft thresholding is implemented using a non-linear function, and the principle is as follows: in, This refers to a noise reduction threshold that is greater than zero. x Features before soft thresholding y This refers to the features after soft thresholding.

7. The method for predicting the very short-term motion response of a submarine cable laying vessel according to claim 1, characterized in that, The model training process includes: dividing the preprocessed dataset into training and test sets, setting the batch size and inputting it into the model, outputting the prediction results through forward propagation, calculating the loss function, optimizing the model parameters using the backpropagation algorithm and gradient descent method, and repeating the training until the loss function converges.

8. A short-term motion response prediction system for a submarine cable laying vessel, characterized in that, include: The data acquisition module acquires six-degree-of-freedom time-series data based on the shipborne inertial measurement unit of the submarine cable construction vessel and constructs a dataset. The data preprocessing module preprocesses the dataset, including time-series windowing and normalization. The model building module constructs a fusion convolutional residual shrinking bidirectional long short-term memory network model CNN-RSBU-BiLSTM, which includes a one-dimensional convolutional neural network CNN residual shrinking building unit RSBU and a bidirectional long short-term memory network BiLSTM connected in sequence. The model training module trains the CNN-RSBU-BiLSTM model using the preprocessed dataset until the model converges. The prediction output module inputs the real-time collected and preprocessed six-degree-of-freedom time-series data into the trained model and outputs the prediction results of the very short-term six-degree-of-freedom motion response of the submarine cable construction vessel.

9. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and running thereon, characterized in that, When the processor executes the program, it implements the method for predicting the extremely short-term motion response of a submarine cable construction vessel as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for predicting the very short-term motion response of a submarine cable construction vessel as described in any one of claims 1-7.

Images