Wave method and system for ship motion inversion based on artificial neural network

Abstract

The present application belongs to the technical field of ship and ocean engineering, and discloses a ship-following oscillation motion inversion wave method and system based on an artificial neural network. The method performs inversion of significant wave height, characteristic period and wave direction in wave statistical characteristics based on ship motion in a data-driven manner, simultaneously acquires ship motion data through a pose sensor arranged on the ship, and further acquires surrounding sea wave information of the ship. The present application has good robustness to different sea conditions. Meanwhile, this kind of method can realize acquisition of ship motion data through a pose sensor arranged on the ship, has relatively simple hardware requirements, is high in cost-effectiveness, and can realize ship-following wave monitoring. This kind of method provides a new solution path for current acquisition of surrounding sea wave information of the ship, and can provide wave environment data support for ship navigation and operation decision-making.

Description

Technical Field

[0001] This invention belongs to the field of shipbuilding and marine engineering technology, and in particular relates to a method and system for inverting waves based on ship rolling motion using artificial neural networks. Background Technology

[0002] The seakeeping capability of a ship has become a crucial indicator of its navigation performance. During sea voyages, ships inevitably experience six-degree-of-freedom rolling motion due to wave excitation, significantly impacting operational efficiency and navigational safety. Over the years, the average number of containers lost overboard from ships has reached numerous. A large proportion of these accidents are caused by ship rolling and swaying during severe weather. Furthermore, helicopters conducting search and rescue or take-off and landing operations in high sea states are also affected by ship rolling motion, increasing the risk of collisions. Therefore, the rolling motion of ships under high winds and waves is a critical issue in shipping and maritime operations. To address this, real-time perception of the ship's wave environment allows crew members to make informed decisions, reducing the risk of encountering critical waves. Moreover, with the increasing autonomy and intelligence of ship navigation, unmanned surface vessels (USVs) are gradually gaining prominence. USVs can complete a series of tasks without requiring the expertise and participation of crew members. For unmanned surface vessels (USVs) to achieve autonomous, low-energy navigation and operations, they must use information about their external environment as a key input for perception, thereby enabling reasonable navigational control. Therefore, perception of the surrounding ocean wave environment is of significant research importance in practical applications.

[0003] Currently, perception methods for ocean wave environments mainly focus on wave monitoring and wave frequency domain inversion techniques. Wave monitoring primarily relies on direct observation by crew members or monitoring using equipment such as buoys, satellites, and radar. However, due to human or equipment factors, these techniques still face numerous problems and limitations in practical applications: subjective misjudgments from direct crew monitoring, limitations imposed by the monitoring range on buoy wave measurements, and the negative impact of spatiotemporal resolution and meteorological conditions on remote sensing satellite wave measurements. Wave frequency domain inversion, as a method for acquiring wave information using transfer functions and spectral analysis, is primarily suitable for relatively stable sea states. In actual ocean wave environments, there are significant non-stationary characteristics between ship motion and waves. Furthermore, under nonlinear hydrodynamic conditions, the accuracy of transfer functions is difficult to guarantee. Therefore, frequency domain inversion methods still face limitations in practical applications.

[0004] Existing research on wave sensing mainly focuses on wave monitoring technology and wave frequency domain inversion technology, while research combining machine learning with wave sensing is still in its early stages. Current patents and literature on wave sensing primarily involve two schemes: buoy-based wave monitoring and radar-based wave inversion.

[0005] The principle of buoy wave monitoring is to measure ocean waves by utilizing the wave-following property of the buoy body; therefore, most buoys are spherical or cylindrical. Sensors (accelerometers, gyroscopes, etc.) built into the buoy can record the motion state of the ocean waves, such as displacement, acceleration, and deflection angle in different directions. This allows for the calculation of basic wave data such as significant wave height and period, and also the acquisition of wave spectra required for engineering projects through spectral estimation methods.

[0006] Radar-based wave inversion first preprocesses the radar-acquired echo signal data (filtering, denoising, etc.), then analyzes the echo signal characteristics and calculates the inverted wave parameters, such as wave height and period, based on relevant theoretical models. This method can also obtain wave spectrum information data. However, there is still a lack of in-depth research on wave perception methods based on machine learning.

[0007] Wave buoys, as a relatively reliable tool for wave estimation, can monitor waves at fixed locations. However, due to their fixed measurement location, covering vast ocean surfaces with buoys presents significant challenges. Furthermore, the need for buoys to remain afloat for extended periods inevitably leads to wear and tear and damage caused by complex wind and wave environments.

[0008] Although shipboard radar can acquire wave information encountered during ship navigation, its wave measurement accuracy is still affected by resolution and antenna rotation speed. Furthermore, from a cost perspective, not all ships can be equipped with radar systems, making shipboard radar not the most cost-effective option for shipboard wave sensing.

[0009] In summary, obtaining wave information from ships still presents certain challenges in practical engineering applications.

[0010] Based on the above analysis, the problems and shortcomings of existing technologies are as follows: While buoy wave measurement can monitor wave information, this method, due to its fixed measurement location, cannot acquire large-scale wave information. Furthermore, this method is costly to deploy and cannot avoid buoy losses due to adverse sea conditions. Shipborne radar can acquire wave information around the ship, but radar deployment is expensive, and not all ships can be equipped with radar. Additionally, the accuracy of radar wave measurement is limited by antenna rotation speed and measurement distance. Summary of the Invention

[0011] To overcome the problems existing in related technologies, the embodiments disclosed in this invention provide a method and system for inverting waves based on ship rolling motion using artificial neural networks. This invention, which uses ship motion response data measured by sensors to achieve ship-borne wave inversion, is considered a research method capable of improving the aforementioned shortcomings. Furthermore, the machine learning-based method constructs a complex mapping relationship between ship motion and waves, exhibiting good robustness against complex ocean wave environment characteristics and ship hydrodynamic characteristics. It can, to some extent, replace the role of the transfer function, making it more suitable for practical applications.

[0012] The technical solution is as follows: A wave inversion method based on ship-borne rolling motion using artificial neural networks. This method uses a data-driven approach to invert meaningful wave height, characteristic period, and wave direction based on the statistical characteristics of wave motion from ship motion. Simultaneously, it acquires ship motion data using pose sensors deployed on the ship, thereby obtaining information about the surrounding ocean waves. Specifically, it includes the following steps:

[0013] S1, preprocess the six-degree-of-freedom motion data of the ship acquired by the sensor, and after data extraction and classification, the preprocessed motion data is transformed into single-degree-of-freedom motion data and multi-degree-of-freedom motion data of the ship.

[0014] S2, reconstruct the obtained single-degree-of-freedom motion data and multi-degree-of-freedom motion data of the ship to obtain a dataset that meets the input requirements of the artificial neural network model;

[0015] S3, the reconstructed ship single-degree-of-freedom motion data and multi-degree-of-freedom motion data are mixed and washed separately, and then the mixed data is divided into training set and test set;

[0016] S4. Based on the divided single-degree-of-freedom motion dataset and multi-degree-of-freedom motion dataset of the ship, construct artificial neural network models respectively, and train and test the artificial neural network models. Use the artificial neural network models to obtain the inversion results of the meaningful wave height, characteristic period, and wave direction based on the ship motion.

[0017] In step S1, the six-degree-of-freedom motion data of the ship acquired by the sensors is preprocessed. After data extraction and classification, the preprocessed motion data is divided into single-degree-of-freedom motion data and multi-degree-of-freedom motion data of the ship, specifically including:

[0018] The six-degree-of-freedom motion time history data acquired by the sensor is as follows: In the formula, Let t1, t2, ..., t be respectively. n The motion attitude data of the ship at each degree of freedom at time x1, x2…x nWhere n represents the length of the data and m represents the number of degrees of freedom, the six degrees of freedom of the ship's motion are sway, pitch, heave, roll, pitch, and bow, and are numbered 1-6 respectively; in the motion time history data collected by the sensors, the raw data is translated and filtered; the translation process first obtains the mean of the motion time history data of each degree of freedom. In the formula Let t1, t2, ..., t be respectively. n Ship motion attitude data for each degree of freedom at time x1, x2…x n The mean of the time-series data, after being shifted, is represented as follows: fn2-fn2′…fn6-fn6′=fn1″,fn2″…fn6″, where fn1″,fn2″…fn6″ represent the ship's motion attitude data x1,x2…x1,x2…x6′ after translation processing at times t1, t2…tn, respectively. n The translated data still needs to be processed by Fourier filtering;

[0019] The data after translation and filtering preprocessing are extracted and classified according to their respective uses, and divided into single-degree-of-freedom motion data and multi-degree-of-freedom motion data;

[0020] The preprocessed single-degree-of-freedom motion data of the ship is heave motion data. Ship multi-degree-of-freedom motion data includes ship roll and pitch motion data.

[0021] In step S2, the obtained single-degree-of-freedom motion data and multi-degree-of-freedom motion data of the ship are reconstructed, including:

[0022] First, the time history data of the single-degree-of-freedom heave motion... Processing is divided into blocks based on a certain time duration, specifically as follows:

[0023]

[0024] Where d is the time length;

[0025] Secondly, statistical processing is performed on each time segment, and meaningful values ​​are taken for each time segment; the reconstructed single-degree-of-freedom heave motion data is represented as follows:

[0026]

[0027] Furthermore, the meaningful value is determined as follows: multiple wave heights are obtained by wave-by-wave analysis of a time series data, the wave heights are arranged in descending order, and the average value of the first 1 / 3 of the wave height data is taken as the meaningful value.

[0028] In step S2, a dataset that meets the input requirements of an artificial neural network model is obtained, including:

[0029] The hull roll and pitch motion data obtained by segmentation and statistical analysis are expressed as follows: and Considering that the relative motions of the ship's roll and pitch are closely related to the wave direction, the roll and pitch motion data collected and processed simultaneously are ratio-processed to obtain the relative change values ​​of the roll and pitch motions, specifically expressed as follows:

[0030] Furthermore, in the reconstruction processing of the obtained ship single-degree-of-freedom motion data and multi-degree-of-freedom motion data, the meaningful wave height H matched with the time segment is... s Characteristic period T p and waves towards D m Wave data corresponds one-to-one, represented as H s =[Y1,Y2…Y n / d ],T p =[Y1,Y2…Y n / d ],D m =[Y1,Y2…Y n / d The meaningful wave height dataset is H. s Y1, Y2…Y n / d These represent the reconstructed ship motion data segments X1, X2…X, respectively. n / d The corresponding wave has a meaningful wave height value; similarly, in the characteristic periodic dataset T p In the case of Y1, Y2…Y n / d These represent the reconstructed ship motion data segments X1, X2…X, respectively. n / d The corresponding wave characteristic period value; the corresponding wave direction dataset D m Y1, Y2…Y n / d These represent the reconstructed ship motion data segments X1, X2…X, respectively. n / d The corresponding wave direction.

[0031] In step S3, the reconstructed ship single-degree-of-freedom motion data and multi-degree-of-freedom motion data are mixed and washed separately, including:

[0032] The reconstructed ship single-degree-of-freedom motion data And the corresponding wave data H s =[Y1,Y2…Y n / d First, assign uniform sequence numbers, then shuffle the numbers to ensure a one-to-one correspondence between data and numbers. Then, divide the shuffled data into training and test sets in a 4:1 ratio. The input and output of the training set are represented as Train. X= [X1, X2…X j ] and Train Y = [X1, X2…X j The corresponding test set can be represented as Text. X = [X1, X2…X k ] and Text Y =[Y1,Y2…Y k The lengths of the training and test sets are set according to actual needs;

[0033] The reconstructed ship multi-degree-of-freedom motion data And the corresponding wave data D m =[Y1,Y2…Y n / d A unified sequence number is assigned, and then the numbering is shuffled to ensure a one-to-one correspondence between the data and the number. The shuffled data is then divided into training and testing sets, with the lengths of the training and testing sets set according to actual needs.

[0034] Furthermore, the shuffling process is based on data reconstruction. The original data is divided into blocks, and the statistical values ​​of the corresponding sub-blocks are taken. Then, each statistical value is sorted out.

[0035] In step S4, the inversion results of the meaningful wave height, characteristic period, and wave direction based on ship motion are obtained using an artificial neural network model, including:

[0036] The input to the artificial neural network model is a statistical sample X of ship motion data. input = [X1, X2…X i The output is the corresponding wave statistics Y. output =[Y1,Y2…Y i ], where i is the number of samples;

[0037] When the ship motion statistics input to the artificial neural network model are X input When the time is right, the corresponding output is Y. output =f(W×X) input +b), where W is the weight of each input element in the neuron, b is the bias of each input element in the neuron, and f(x) is the activation function, specifically the ReLU function, expressed as follows:

[0038] During data transmission, activation functions are used to map the data; the output Y of the neural network is then obtained. outputNext, the output is compared with the actual data, and the error between the two is measured using a loss function. The weights are then adjusted to make the output closer to the actual result. The mean squared error (MSE) is chosen as the loss function, specifically expressed as MSE = mean(∑(Y)). output -Y real )) 2 , where Y real The true value is used; thus, the training of the artificial neural network model is completed through iterative calculation.

[0039] Another objective of this invention is to provide a wave inversion system based on ship rolling motion using an artificial neural network, and to implement the wave inversion method based on ship rolling motion using an artificial neural network. The system includes:

[0040] The single-degree-of-freedom motion data and multi-degree-of-freedom motion data acquisition module is used to preprocess the six-degree-of-freedom motion data of the ship acquired by the sensor. After the preprocessing, the motion data is extracted and classified to form two types of data: single-degree-of-freedom motion data and multi-degree-of-freedom motion data.

[0041] The reconstruction processing module is used to reconstruct the obtained single-degree-of-freedom motion data and multi-degree-of-freedom motion data of the ship to obtain a dataset that meets the input requirements of the artificial neural network model.

[0042] The mixed washing module is used to mix the reconstructed ship single-degree-of-freedom motion data and multi-degree-of-freedom motion data, and then divide the mixed data into training set and test set.

[0043] The wave inversion module is used to construct artificial neural network models based on the obtained single-degree-of-freedom motion dataset and multi-degree-of-freedom motion dataset of the ship, and to train and test the artificial neural network models to obtain wave inversion results based on ship motion.

[0044] Combining all the above technical solutions, the advantages and positive effects of this invention are as follows: This invention proposes a wave inversion method based on ship rolling motion using an artificial neural network model. It aims to utilize ship motion data and, with the help of an artificial neural network model, construct a complex mapping relationship between ship motion and waves, thereby obtaining the meaningful wave height, characteristic period, and wave direction, providing wave environment data support for ship navigation and operational decisions.

[0045] This invention proposes a wave inversion method based on ship-borne rolling motion using an artificial neural network model. This data-driven method enables the inversion of wave statistical characteristics (sense wave height, characteristic period, and wave direction, etc.) based on ship motion, exhibiting good robustness across various sea states. Furthermore, this method can acquire ship motion data using pose sensors deployed on board, resulting in simple hardware requirements, high cost-effectiveness, and the ability to monitor waves while on board. This approach provides a novel solution for acquiring wave information around a ship, offering wave environment data support for ship navigation and operational decisions. Attached Figure Description

[0046] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure;

[0047] Figure 1 This is a flowchart of the wave inversion method based on artificial neural network for ship rolling motion provided in this embodiment of the invention;

[0048] Figure 2 This is a schematic diagram of the simulation data of heave motion time history under a certain working condition provided in an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of the wave inversion method based on artificial neural network for ship rolling motion provided in this embodiment of the invention;

[0050] Figure 4 This is a schematic diagram of the reconstructed heave motion data sample provided in an embodiment of the present invention;

[0051] Figure 5 This is a sample image of heave motion data obtained after mixing and washing, provided in an embodiment of the present invention.

[0052] Figure 6 This is a comparison chart of the inverted values ​​of the meaningful wave height based on the artificial neural network model provided in this embodiment of the invention with the true values;

[0053] Figure 7 This is a schematic diagram of a wave inversion system based on artificial neural networks for ship swaying motion provided in an embodiment of the present invention;

[0054] In the diagram: 1. Single-degree-of-freedom motion data acquisition module; 2. Reconstruction processing module; 3. Mixed washing processing module; 4. Wave inversion module. Detailed Implementation

[0055] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0056] The wave inversion method based on ship rolling motion provided in this embodiment of the invention uses a data-driven approach to invert meaningful wave height, characteristic period, and wave direction based on the statistical characteristics of wave motion of the ship motion. At the same time, it acquires ship motion data through pose sensors deployed on the ship, thereby obtaining information on the sea waves around the ship.

[0057] Specifically, a data-driven approach can achieve the inversion of wave statistical characteristics (sense wave height, characteristic period, and wave direction, etc.) based on ship motion, exhibiting good robustness across various sea states. Furthermore, this method can acquire ship motion data using onboard pose sensors, requiring relatively simple hardware, offering high cost-effectiveness, and enabling onboard wave monitoring. This approach provides a new solution for acquiring information on the surrounding ocean waves, providing wave environment data support for ship navigation and operational decisions.

[0058] Example 1, such as Figure 1 As shown, the wave inversion method based on artificial neural network for ship-borne swaying motion provided in this embodiment of the invention includes:

[0059] S1 involves preprocessing the six-degree-of-freedom motion data of the ship acquired by the sensors. After data extraction and classification, the preprocessed motion data is divided into two types: single-degree-of-freedom motion data and multi-degree-of-freedom motion data. Both types of data are used in subsequent deep learning model inversion algorithms.

[0060] The specific method for step S1 is as follows: The six-degree-of-freedom motion time history data acquired by the sensor is: In the formula, Let t1, t2, ..., t be respectively. n The motion attitude data of the ship at each degree of freedom at time x1, x2…x n Where n represents the length of the data and m represents the number of degrees of freedom, the six degrees of freedom of the ship's motion are sway, pitch, heave, roll, pitch, and bow, and are numbered 1-6 respectively; in the motion time history data collected by the sensors, the raw data is translated and filtered; the translation process first obtains the mean of the motion time history data of each degree of freedom. In the formula Let t1, t2, ..., t be respectively. n Ship motion attitude data for each degree of freedom at time x1, x2…x n The mean of the time-series data, after being shifted, is represented as follows: In the formula, Let t1, t2, ..., t be respectively. n The motion attitude data of the ship in each degree of freedom obtained after translation processing under time conditions are x1, x2…x n The translated data still needs to be processed by Fourier filtering;

[0061] The data after translation and filtering preprocessing are extracted and classified according to their respective uses, and divided into single-degree-of-freedom motion data and multi-degree-of-freedom motion data;

[0062] The preprocessed single-degree-of-freedom motion data of the ship is heave motion data. Ship multi-degree-of-freedom motion data includes ship roll and pitch motion data.

[0063] It is understandable that data translation and FFT smoothing are common data preprocessing methods for the six-degree-of-freedom motion data of ships. These methods can make the data input to the model fluctuate around zero and make the data smooth and continuous, which facilitates subsequent model training.

[0064] This method for extracting and classifying six-degree-of-freedom (DOF) motion data of ships is based on the theoretical foundation of ship motion and waves. The ship's heave and pitch motions are closely related to wave height and wave period; the relative motion between the ship's roll and pitch is closely related to wave direction. Therefore, before constructing the model, the raw data is classified according to different wave feature acquisition methods: wave height and wave period are inverted using the ship's heave or pitch motion; wave direction is inverted using the ship's roll and pitch motions. Typically, both motion prediction and wave inversion refer to the theory of seakeeping, mainly focusing on the analysis of the ship's roll, pitch, and heave motions. The innovation of this invention lies in the wave direction inversion, which mainly focuses on the relative motion between the ship's roll and pitch—an idea that has not been mentioned or disclosed in existing technologies.

[0065] S2, the single-degree-of-freedom motion data and multi-degree-of-freedom motion data of the ship obtained in step S1 are reconstructed to obtain a dataset that meets the input requirements of the artificial neural network model.

[0066] The specific method of step S2 is as follows: For the reconstruction of single-degree-of-freedom motion data, taking heave motion data as an example, firstly, the single-degree-of-freedom heave motion time history data is... Processing is divided into blocks based on a certain time duration, specifically as follows:

[0067]

[0068] Where d is the time length;

[0069] Secondly, statistical processing is performed on each time segment, and meaningful values ​​are taken for each time segment; the reconstructed single-degree-of-freedom heave motion data is represented as follows:

[0070]

[0071] In step S2, the reconstruction processing of the multi-degree-of-freedom motion data obtained in step S1 includes:

[0072] The hull roll and pitch motion data obtained by segmentation and statistical analysis are expressed as follows: and Considering that the relative motions of the ship's roll and pitch are closely related to the wave direction, the roll and pitch motion data collected and processed simultaneously are ratio-processed to obtain the relative change values ​​of the roll and pitch motions, specifically expressed as follows:

[0073] In the reconstruction process of the obtained single-degree-of-freedom and multi-degree-of-freedom motion data of the ship, the meaningful wave height H matched with the time segment is... s Characteristic period T p and waves towards D m Wave data corresponds one-to-one, represented as H s =[Y1,Y2…Y n / d ],T p =[Y1,Y2…Y n / d ],D m =[Y1,Y2…Y n / d The meaningful wave height dataset is H. s Y1, Y2…Y n / d These represent the reconstructed ship motion data segments X1, X2…X, respectively. n / d The corresponding wave has a meaningful wave height value; similarly, in the characteristic periodic dataset T p In the case of Y1, Y2…Y n / d These represent the reconstructed ship motion data segments X1, X2…X, respectively. n / d The corresponding wave characteristic period value; the corresponding wave direction dataset D m Y1, Y2…Y n / d These represent the reconstructed ship motion data segments X1, X2…X, respectively. n / d The corresponding wave direction.

[0074] It is understandable that data segmentation usually involves dividing the time-series data into segments with a certain window length. This type of processing is a common method in time series algorithms based on deep learning and machine learning.

[0075] The innovation of this invention lies in using meaningful values ​​as statistical values ​​for each time segment. Meaningful values, as important statistical parameters for measuring ship motion and waves, characterize the overall horizontal magnitude of ship motion and waves, and are mentioned in existing seakeeping principles. Based on an intelligent model, the statistical values ​​of ship motion are used as input to invert the meaningful wave height, characteristic period, and wave direction. This type of method has not been mentioned or disclosed in existing technologies.

[0076] S3. The single-degree-of-freedom motion data and multi-degree-of-freedom motion data of the ship obtained in step S2 are mixed and washed, and then the mixed data is divided into training set and test set.

[0077] The reconstructed ship single-degree-of-freedom motion data And the corresponding wave data H s =[Y1,Y2…Y n / d First, assign uniform sequence numbers, then shuffle the numbers to ensure a one-to-one correspondence between data and numbers. Then, divide the shuffled data into training and test sets in a 4:1 ratio. The input and output of the training set are represented as Train. X = [X1, X2…X j ] and Train Y = [X1, X2…X j The corresponding test set can be represented as Text. X = [X1, X2…X k ] and Text Y =[Y1,Y2…Y k The lengths of the training and test sets are set according to actual needs;

[0078] The reconstructed ship multi-degree-of-freedom motion data And the corresponding wave data D m =[Y1,Y2…Y n / d A unified sequence number is assigned, and then the numbering is shuffled to ensure a one-to-one correspondence between the data and the number. The shuffled data is then divided into training and testing sets, with the lengths of the training and testing sets set according to actual needs.

[0079] It's understandable that the data shuffling process for ship motion data follows data reconstruction. This involves dividing the original data into blocks, extracting the statistical values ​​from each block, and then shuffling the order of these values. The data shuffling is implemented using built-in Python functions. The shuffled data serves as input to the intelligent model, facilitating training across all datasets and improving model training and prediction. Regarding dataset partitioning, the ratio of training to test set lengths is typically 4:1, but this can be adjusted based on the actual dataset size in the application.

[0080] S4. Based on the ship single-degree-of-freedom motion dataset and multi-degree-of-freedom motion dataset divided in step S3, construct artificial neural network models respectively, and train and test the artificial neural network models. Use the artificial neural network models to obtain inversion results such as meaningful wave height, characteristic period, and wave direction based on ship motion.

[0081] The specific method for step S4 is as follows:

[0082] The input to the artificial neural network model is a statistical sample X of ship motion data. input = [X1, X2…X i The output is the corresponding wave statistics Y. output =[Y1,Y2…Y i ], where i is the number of samples;

[0083] When the ship motion statistics input to the artificial neural network model are X input When the time is right, the corresponding output is Y. output =f(W×X) input +b), where W is the weight of each input element in the neuron, b is the bias of each input element in the neuron, and f(x) is the activation function, specifically the ReLU function, expressed as follows:

[0084] During data transmission, activation functions are used to map the data; the output Y of the neural network is then obtained. output Next, the output is compared with the actual data, and the error between the two is measured using a loss function. The weights are then adjusted to make the output closer to the actual result. The mean squared error (MSE) is chosen as the loss function, specifically expressed as MSE = mean(∑(Y)). output -Y real )) 2 , where Y real The true value is used; thus, the training of the artificial neural network model is completed through iterative calculation.

[0085] Therefore, the training of an artificial neural network (ANN) model can be completed through iterative calculations. The neural network hierarchy, number of neurons, activation function, and loss function can be adjusted according to the specific circumstances.

[0086] Example 2, as another implementation of the present invention, focuses on the inversion of meaningful wave height and characteristic period based on the single-degree-of-freedom heave motion of a ship for ease of demonstration. The data source is simulation data obtained based on the slicing method and linear superposition theory. The operating conditions are relatively simple, mainly considering sea states of 3-5, zero speed, and 180° wave direction. In real-world ocean environments, sea states, ship speeds, and headings are complex and variable; therefore, datasets can be constructed and model effectiveness verified from three different perspectives: changes in sea state, speed, and heading. Furthermore, the inversion of meaningful wave height and characteristic period can also be achieved using the ship's pitching motion.

[0087] Furthermore, the dataset used in this example is simulation data obtained based on the slicing method and linear superposition theory. However, different data sources such as pool data and actual ship data can also be used.

[0088] This example utilizes ship heave data and an artificial intelligence model to invert the meaningful wave height and characteristic period of waves based on ship motion. The data consists of ship heave motion time-history simulation data and meaningful wave height and characteristic period data under corresponding operating conditions, obtained through the slicing method and linear superposition theory. The operating conditions are set within sea states 3-5, corresponding to meaningful wave heights of [0.5m-5.0m] and wave characteristic periods of [6.5s, 9.25s]. Twelve sea states are selected at equal intervals within this range. In each operating condition, the ship's speed is set to 0, the wave direction is top-on (wave direction angle is 180°), and the ship motion simulation data duration is 30,000s with a time interval of 0.5s. To achieve wave inversion based on ship motion, the above operating conditions need to be uniformly stitched together to ensure full coverage of all data within the set operating condition range. The ship heave time-history data under one of these operating conditions is shown below. Figure 2 As shown.

[0089] like Figure 3 As shown, the wave inversion method based on artificial neural network for ship swaying motion provided in this embodiment of the invention includes the following steps:

[0090] Step 1: Preprocess the six-degree-of-freedom motion data of the ship acquired by the sensors. After preprocessing, the motion data is extracted and classified into two types: single-degree-of-freedom motion data and multi-degree-of-freedom motion data. Both types of data are used in the subsequent deep learning model inversion algorithm. The specific method of Step 1 is as follows: The six-degree-of-freedom motion time history data acquired by the sensors is as follows: In the formula, Let t1, t2, ..., t be respectively. n The motion attitude data of the ship at each degree of freedom at time x1, x2…x n Where n represents the length of the data and m represents the number of degrees of freedom, the six degrees of freedom of the ship's motion are sway, pitch, heave, roll, pitch, and bow, and are numbered 1-6 respectively; in the motion time history data collected by the sensors, the raw data is translated and filtered; the translation process first obtains the mean of the motion time history data of each degree of freedom. In the formula Let t1, t2, ..., t be respectively. n Ship motion attitude data for each degree of freedom at time x1, x2…x n The mean of the time-series data, after being shifted, is represented as follows: In the formula, Let t1, t2, ..., t be respectively. n The motion attitude data of the ship in each degree of freedom obtained after translation processing under time conditions are x1, x2…x n The translated data still needs to be processed by Fourier filtering;

[0091] The data after translation and filtering preprocessing are extracted and classified according to their respective uses, and divided into single-degree-of-freedom motion data and multi-degree-of-freedom motion data;

[0092] The preprocessed single-degree-of-freedom motion data of the ship is heave motion data. Ship multi-degree-of-freedom motion data includes ship roll and pitch motion data.

[0093] It is understandable that step 1 achieves the preprocessing and classification of ship motion data.

[0094] Step 2: Reconstruct the single-degree-of-freedom motion data and multi-degree-of-freedom motion data obtained in Step 1 to obtain a dataset that meets the input requirements of the artificial neural network model. The specific method of Step 2 is as follows: For the reconstruction of single-degree-of-freedom motion data, taking heave motion data as an example, firstly, the single-degree-of-freedom heave motion time-history data... Processing is divided into blocks based on a certain time duration, specifically as follows:

[0095]

[0096] Where d is the time length;

[0097] Secondly, statistical processing is performed on each time segment. Typically, meaningful values ​​are taken from each time segment (the method for determining meaningful values ​​is: for a time series of data, multiple wave heights are obtained through wave-by-wave analysis, then the wave heights are arranged in descending order, and the average of the first 1 / 3 of the wave height data is taken as the meaningful value). Therefore, the reconstructed single-degree-of-freedom heave motion data can be represented as follows: For the reconstruction of multi-degree-of-freedom motion data, the hull's roll and pitch motion data are usually reconstructed in a similar way to single-degree-of-freedom motion data, which can be specifically represented as follows: and Considering that the relative motions of the ship's roll and pitch are closely related to the wave direction, the roll and pitch motion data collected and processed simultaneously are ratio-processed to obtain the relative change values ​​of the roll and pitch motions, specifically expressed as follows:

[0098] In addition, the meaningful wave height H matched with the time segment s Characteristic period T p and waves towards D m Wave data corresponds one-to-one, represented as H s =[Y1,Y2…Y n / d ],T p =[Y1,Y2…Y n / d ],D m =[Y1,Y2…Y n / d The meaningful wave height dataset is H. s Y1, Y2…Y n / d These represent the reconstructed ship motion data segments X1, X2…X, respectively. n / d The corresponding wave has a meaningful wave height value; similarly, in the characteristic periodic dataset T p In the case of Y1, Y2…Y n / d These represent the reconstructed ship motion data segments X1, X2…X, respectively. n / d The corresponding wave characteristic period value; the corresponding wave direction dataset D m Y1, Y2…Y n / d These represent the reconstructed ship motion data segments X1, X2…X, respectively. n / d The corresponding wave direction.

[0099] The reconstructed heave motion data is as follows Figure 4 As shown.

[0100] It is understandable that step 2 realizes the segmentation of ship motion data and the statistical processing of each sub-block, thus achieving data reconstruction.

[0101] Step 3: The ship's single-DOF motion data and multi-DOF motion data reconstructed in Step 2 are mixed, and then the mixed data is divided into training and testing sets. To maintain the one-to-one correspondence between motion and waves after data mixing, the reconstructed ship single-DOF motion data... And the corresponding wave data H s =[Y1,Y2…Y n / d First, assign uniform sequence numbers, then shuffle the numbers to ensure a one-to-one correspondence between data and numbers. Then, divide the shuffled data into training and test sets in a 4:1 ratio. The input and output of the training set are represented as Train. X = [X1, X2…X j ] and Train Y = [X1, X2…X j The corresponding test set can be represented as Text. X = [X1, X2…X k ] and Text Y =[Y1,Y2…Y k The lengths of the training and test sets are set according to actual needs;

[0102] The reconstructed ship multi-degree-of-freedom motion data And the corresponding wave data D m =[Y1,Y2…Y n / d A unified sequence number is assigned, and then the numbering is shuffled to ensure a one-to-one correspondence between the data and the number. The shuffled data is then divided into training and testing sets, with the lengths of the training and testing sets set according to actual needs.

[0103] The heave motion data after mixing and washing are as follows Figure 5 As shown.

[0104] It is understandable that step 3, based on the data reconstruction, implements the shuffling process of the reconstructed data and divides the training set and test set to facilitate model training and testing.

[0105] Step 4: Construct Artificial Neural Network (ANN) models based on the dataset obtained in Step 3, and train and test the models. Use the ANN models to obtain wave inversion results based on ship motion. Specifically, Step 4 involves using the ANN models to obtain inversion results for meaningful wave height, characteristic period, and wave direction based on ship motion, including:

[0106] The input to the artificial neural network model is a statistical sample X of ship motion data. input = [X1, X2…X i The output is the corresponding wave statistics Y.output =[Y1,Y2…Y i ], where i is the number of samples;

[0107] When the ship motion statistics input to the artificial neural network model are X input When the time is right, the corresponding output is Y. output =f(W×X) input +b), where W is the weight of each input element in the neuron, b is the bias of each input element in the neuron, and f(x) is the activation function, specifically the ReLU function, expressed as follows:

[0108] During data transmission, activation functions are used to map the data; the output Y of the neural network is then obtained. output Next, the output is compared with the actual data, and the error between the two is measured using a loss function. The weights are then adjusted to make the output closer to the actual result. The mean squared error (MSE) is chosen as the loss function, specifically expressed as MSE = mean(∑(Y)). output -Y real )) 2 , where Y real The true value is used; thus, the training of the artificial neural network model is completed through iterative calculation.

[0109] It is understandable that step 4 involves constructing an artificial neural network model based on the model input and output, and realizing wave inversion based on ship motion.

[0110] like Figure 6 The comparison results between the inverted values ​​and the true values ​​of the meaningful wave height based on the artificial neural network model are shown. Figure 6 The middle section compares the meaningful values ​​of the actual waves with the model inversion results. The closer the model inversion results are to the actual values, the better the model's inversion effect. The actual values ​​in this section were obtained synchronously in the preceding numerical simulation of ship motion, that is, using the slicing method and linear superposition theory to obtain ship motion and wave data under different operating conditions. This result is the inversion result at the numerical level.

[0111] Example 3: The artificial neural network (ANN) used in this embodiment of the invention constructs a wave inversion model to realize the mapping relationship between ship motion and wave information data. More precisely, other time-series prediction neural networks, such as Long Short-Term Memory (LSTM) networks and Temporal Convolutional Networks (TCN), can also be used.

[0112] Example 4, as Figure 7 As shown, this embodiment of the invention provides a wave inversion system based on artificial neural networks for ship-borne swaying motion, comprising:

[0113] The single-degree-of-freedom motion data and multi-degree-of-freedom motion data acquisition module 1 is used to preprocess the six-degree-of-freedom motion data of the ship acquired by the sensor. After the preprocessing, the motion data is extracted and classified to form two types of data: single-degree-of-freedom motion data and multi-degree-of-freedom motion data.

[0114] Reconstruction processing module 2 is used to reconstruct the obtained single-degree-of-freedom motion data and multi-degree-of-freedom motion data of the ship to obtain a dataset that meets the input requirements of the artificial neural network model;

[0115] The mixed washing module 3 is used to mix and process the reconstructed ship single-degree-of-freedom motion data and multi-degree-of-freedom motion data, and then divide the mixed data into training set and test set.

[0116] Wave inversion module 4 is used to construct artificial neural network models based on the obtained training and test sets, and to train and test the artificial neural network models to obtain wave inversion results based on ship motion.

[0117] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0118] The information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0119] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this invention. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments.

[0120] Based on the technical solutions described in the above embodiments of the present invention, the following application examples can be further proposed.

[0121] According to embodiments of this application, the present invention also provides a computer device comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the steps in any of the above-described method embodiments.

[0122] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps described in the various method embodiments above.

[0123] This invention also provides an information data processing terminal, which, when executed on an electronic device, provides a user input interface to implement the steps described in the above method embodiments. The information data processing terminal is not limited to mobile phones, computers, or switches.

[0124] This invention also provides a server that, when executed on an electronic device, provides a user input interface to implement the steps described in the above method embodiments.

[0125] This invention also provides a computer program product that, when run on an electronic device, enables the electronic device to implement the steps described in the various method embodiments above.

[0126] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying the computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0127] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0128] To further demonstrate the positive effects of the above embodiments, the present invention conducts the following experiments based on the above technical solutions.

[0129] The numerical simulation results obtained so far are all based on simulated data. The results show that the method is feasible to a certain extent.

[0130] Experiments demonstrate that this invention proposes a wave inversion method based on an artificial neural network model for shipboard rolling motion. This method, driven by data, utilizes shipboard motion response data to acquire information about the surrounding wave environment (sense wave height, characteristic period, and wave direction, etc.). This type of algorithm can be integrated into ship situational awareness software to obtain information about the surrounding ocean wave environment. Acquiring this information can provide auxiliary decision-making for ship navigation, thereby reducing navigation and operational risks caused by critical waves and improving the safety of ship navigation and operations. Furthermore, given the known ocean wave environment encountered by the ship, the ship can reduce fuel consumption and improve navigation and operational efficiency to some extent by rationally adjusting its navigation attitude. In addition, this method involves simple hardware requirements and low equipment specifications, making it a cost-effective new method for wave perception. This method can also be widely adopted by ships, providing a sufficient and effective database for sensing global ocean wave environment information and providing strong support for marine environmental analysis.

[0131] This invention proposes a wave inversion method based on shipboard rolling motion using an artificial neural network model. This method, driven by data, utilizes shipboard motion response data to acquire information about the surrounding wave environment (sense wave height, characteristic period, and wave direction, etc.). Currently, the main means of sensing ocean wave environment information are wave buoys and shipboard radar. However, both buoys and radar face challenges due to fixed measurement locations, resolution, and cost. Using shipboard motion response data to invert ocean wave information around the ship via a data-driven approach is still in its early stages in China. This invention utilizes a neural network to map ship motion to wave statistical characteristics, offering novelty compared to traditional methods. Furthermore, the method of inverting wave direction using the relative motion between the ship's roll and pitch is not mentioned in existing literature, also possessing novelty. From a practical application perspective, this invention aims to provide a new solution for acquiring ocean wave environment information from ships.

[0132] This invention proposes a wave inversion method based on an artificial neural network model for ship-borne rolling motion. This method, driven by data, uses the ship's motion response as input and ship-borne wave statistical characteristics (sense wave height, characteristic period, and wave direction, etc.) as output. The artificial neural network model maps the ship's motion to these wave statistical characteristics. The method aims to acquire the wave statistical characteristics surrounding the ship. The objective achieved by this invention is similar to that of wave information sensed by buoys and shipborne radar, both aiming to obtain information such as the sense wave height, characteristic period, and wave direction around the ship. This invention's method has simple hardware requirements and low equipment demands, aiming to provide a new and cost-effective solution for shipborne wave environment perception.

[0133] This invention proposes a wave inversion method based on shipboard rolling motion using an artificial neural network model. This method differs from wave sensing methods such as buoys and shipboard radar. It combines machine learning with wave environment perception and, through a data-driven approach, establishes a mapping relationship between ship motion and wave statistical characteristics, providing a new solution for shipboard wave environment perception.

[0134] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention and within the spirit and principles of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for inverting waves based on ship-borne rolling motion using artificial neural networks, characterized in that, This method uses a data-driven approach to invert meaningful wave height, characteristic period, and wave direction based on the statistical characteristics of waves from ship motion. Simultaneously, it acquires ship motion data using pose sensors deployed on board, thereby obtaining information about the surrounding ocean waves. Specifically, it includes the following steps: S1, preprocess the six-degree-of-freedom motion data of the ship acquired by the sensor, and after data extraction and classification, the preprocessed motion data is transformed into single-degree-of-freedom motion data and multi-degree-of-freedom motion data of the ship. S2, reconstruct the obtained single-degree-of-freedom motion data and multi-degree-of-freedom motion data of the ship to obtain a dataset that meets the input requirements of the artificial neural network model; S3, the reconstructed ship single-degree-of-freedom motion data and multi-degree-of-freedom motion data are mixed and washed separately, and then the mixed data is divided into training set and test set; S4. Based on the divided single-degree-of-freedom motion dataset and multi-degree-of-freedom motion dataset of the ship, construct artificial neural network models respectively, and train and test the artificial neural network models. Use the artificial neural network models to obtain the inversion results of the meaningful wave height, characteristic period, and wave direction based on the ship motion. In step S1, the six-degree-of-freedom motion data of the ship acquired by the sensors is preprocessed. After data extraction and classification, the preprocessed motion data is divided into single-degree-of-freedom motion data and multi-degree-of-freedom motion data of the ship, specifically including: The six-degree-of-freedom motion time history data acquired by the sensor is as follows: In the formula, They represent The motion attitude data of the ship at each degree of freedom at any given time, among which... Indicates the length of the data. The number of degrees of freedom is indicated by the six degrees of freedom of the ship's motion: sway, pitch, heave, roll, pitch, and bow, which are numbered 1-6 respectively. The motion time-history data collected by the sensors undergoes translation and filtering processing on the raw data. The translation processing first obtains the mean value of the motion time-history data for each degree of freedom. In the formula They represent The mean values ​​of the ship's motion attitude data for each degree of freedom over time, after translation processing, are represented as the time history data as follows: In the formula, They represent The motion attitude data of the ship in each degree of freedom obtained after translation processing under time. The translated data still needs to be processed by Fourier filtering; The data after translation and filtering preprocessing are extracted and classified according to their respective uses, and divided into single-degree-of-freedom motion data and multi-degree-of-freedom motion data; The preprocessed single-degree-of-freedom motion data of the ship is heave motion data. The ship's multi-degree-of-freedom motion data includes the ship's roll and pitch motion data. .

2. The wave inversion method based on artificial neural network for ship-borne rolling motion according to claim 1, characterized in that, In step S2, the obtained single-degree-of-freedom motion data and multi-degree-of-freedom motion data of the ship are reconstructed, including: First, the time history data of the single-degree-of-freedom heave motion... Processing is divided into blocks based on a certain time duration, specifically as follows: ； in, The duration; Secondly, statistical processing is performed on each time segment, and meaningful values ​​are taken for each time segment; the reconstructed single-degree-of-freedom heave motion data is represented as follows: 。 3. The wave inversion method based on artificial neural network for ship-borne rolling motion according to claim 2, characterized in that, The meaningful value is determined as follows: multiple wave heights are obtained by wave-by-wave analysis of a time series data, the wave heights are arranged in descending order, and the average value of the first 1 / 3 of the wave height data is taken as the meaningful value.

4. The wave inversion method based on artificial neural network for ship-borne rolling motion according to claim 2, characterized in that, In step S2, a dataset that meets the input requirements of an artificial neural network model is obtained, including: The hull roll and pitch motion data obtained by segmentation and statistical analysis are expressed as follows: and Considering that the relative motions of the ship's roll and pitch are closely related to the wave direction, the roll and pitch motion data collected and processed simultaneously are compared to obtain the relative change values ​​of the roll and pitch motions, specifically expressed as follows: .

5. The wave inversion method based on artificial neural network for ship-borne rolling motion according to claim 4, characterized in that, In the reconstruction process of the obtained single-degree-of-freedom and multi-degree-of-freedom motion data of ships, the meaningful wave height matched with the time segment is... Characteristic Period and wave direction Wave data corresponds one-to-one, represented as The meaningful wave height dataset is , These represent fragments of ship motion data after reconstruction. The corresponding wave has a meaningful wave height value; similarly, in the characteristic periodic dataset In this case, These represent the wave characteristic period values ​​corresponding to the reconstructed ship motion data segments, respectively; Corresponding wave direction dataset, These represent the wave direction corresponding to the reconstructed ship motion data segment.

6. The wave inversion method based on artificial neural network for ship-borne rolling motion according to claim 5, characterized in that, In step S3, the reconstructed ship single-degree-of-freedom motion data and multi-degree-of-freedom motion data are mixed and washed separately, including: The reconstructed ship single-degree-of-freedom motion data and the corresponding wave data A unified sequence numbering process is performed, followed by shuffling of the numbers to ensure a one-to-one correspondence between data and numbers. The shuffled data is then divided into training and test sets in a 4:1 ratio. The input and output representations of the training set are as follows: and The corresponding test set can be represented as and The lengths of the training and test sets are set according to actual needs. The reconstructed ship multi-degree-of-freedom motion data and the corresponding wave data A unified sequence number is assigned, and then the numbering is shuffled to ensure a one-to-one correspondence between the data and the number. The shuffled data is then divided into training and testing sets, with the lengths of the training and testing sets set according to actual needs.

7. The wave inversion method based on artificial neural network for ship-borne rolling motion according to claim 6, characterized in that, The shuffling process is based on data reconstruction. The original data is divided into blocks, and the statistical values ​​of the corresponding sub-blocks are taken. Then, each statistical value is sorted out.

8. The wave inversion method based on artificial neural network for ship-borne rolling motion according to claim 1, characterized in that, In step S4, the inversion results of the meaningful wave height, characteristic period, and wave direction based on ship motion are obtained using an artificial neural network model, including: The input to the artificial neural network model is a statistical sample of ship motion data. The output is the corresponding wave statistics. ,in, The number of samples; When the ship motion statistics input to the artificial neural network model are When that happens, the corresponding output is ,in, For each input element, the weight in that neuron is... This corresponds to the bias of each input element in the neuron. For the activation function, we choose the ReLU function, with the specific expression as follows: ; During data transmission, activation functions are used to map the data; the output of the neural network is then obtained. Next, the output is compared with the actual data, and the error between the two is measured using a loss function. The weights are then adjusted to make the output closer to the actual result. The mean squared error (MSE) is chosen as the loss function, and its specific expression is as follows: ,in The true value is used; thus, the training of the artificial neural network model is completed through iterative calculation.

9. A wave inversion system based on ship-borne rolling motion using artificial neural networks, characterized in that, The system implementing the wave inversion method based on artificial neural networks for ship-borne swaying motion according to any one of claims 1-8, comprises: The module for acquiring single-degree-of-freedom motion data and multi-degree-of-freedom motion data (1) is used to preprocess the six-degree-of-freedom motion data of the ship acquired by the sensor. After the preprocessing, the motion data is extracted and classified to form two types of data: single-degree-of-freedom motion data and multi-degree-of-freedom motion data. The reconstruction processing module (2) is used to reconstruct the obtained single-degree-of-freedom motion data and multi-degree-of-freedom motion data of the ship to obtain a dataset that meets the input requirements of the artificial neural network model. The mixed washing module (3) is used to mix the reconstructed ship single-degree-of-freedom motion data and multi-degree-of-freedom motion data, and then divide the mixed data into training set and test set. The wave inversion module (4) is used to construct artificial neural network models based on the obtained single-degree-of-freedom motion dataset and multi-degree-of-freedom motion dataset of the ship, and to train and test the artificial neural network models, and to obtain wave inversion results based on ship motion using the artificial neural network models.

Images