An online ship motion prediction method based on X-band navigation radar

Abstract

The application discloses an online ship motion prediction method based on an X-band navigation radar, and comprises the following steps: wave information of a sea area around a ship is monitored and recorded by a wave radar; complete wave field data and one-dimensional wave time history data in a specific direction are obtained through an inversion algorithm; six-degree-of-freedom motion time history data of the ship are combined, and are input into a neural network prediction model together, so that a future motion change trend of the ship is finally obtained. The ocean environment information is introduced into the prediction model, so that the model can better learn the excitation mechanism of the environmental factors on the ship motion, and the performance of the model is improved, and the training period is shortened.

Description

Technical Field

[0001] This invention relates to the field of ship motion prediction technology, and in particular to an online ship motion prediction method based on X-band navigation radar. Background Technology

[0002] In the field of shipbuilding and ocean engineering, the prediction of the motion of ships and structures has always been a research direction worthy of exploration and in-depth study. Accurate prediction can provide a reference for motion compensation devices and provide safety guidance for aircraft landing and cargo transfer.

[0003] Since the mid-20th century, experts and scholars have been studying the interaction between offshore platforms and the environment. This research focuses on the motion of offshore platforms in waves. The motion of offshore platforms can be divided into two main categories. The first category includes roll, pitch, and heave motions, which are influenced by the hydrostatic restoring force provided by the free surface; therefore, their natural periods are relatively short, close to the wave period, and are mainly driven by wave excitation. The second category includes sway, roll, and pitch motions, whose restoring force comes from the mooring system, and therefore their natural periods are typically longer. In this case, in addition to wave-frequency motion, slow drift motion may also occur, making the motion more complex.

[0004] Motion forecasting methods can generally be divided into traditional forecasting methods based on hydrodynamic theory and methods utilizing classical time series analysis. With the rapid development of computing power, artificial neural networks have become a new option for building forecasting models, providing a new approach to solving this problem by fully mining the rich information contained in the data itself.

[0005] Typically, neural network-based models belong to the category of purely data-driven forecasting methods. This means they ignore the physical mechanisms of motion and rely solely on the characteristics of the data itself. Through continuous iterative calculations, the model autonomously "learns" features such as data trends, cycles, and autocorrelation. The advantages of building models in this way are obvious: because physical mechanisms are ignored, modeling is relatively simple. However, due to the lack of prior knowledge, forecasting models built in this way have a relatively low upper limit of accuracy and a long training period.

[0006] Therefore, those skilled in the art are dedicated to developing an online ship motion prediction method based on X-band navigation radar. By introducing marine environmental information into the prediction model, the model can better learn the excitation mechanism of environmental factors on ship motion, thereby improving model performance and shortening the training cycle. Summary of the Invention

[0007] In view of the above-mentioned shortcomings of the prior art, the technical problem to be solved by the present invention is that: the modeling of a prediction model using a neural network is simple, but due to the lack of prior knowledge support, the upper limit of the model's accuracy is relatively low and the training cycle is long.

[0008] To achieve the above objectives, this invention provides an online ship motion prediction method based on X-band navigation radar, comprising the following steps:

[0009] Step 1: Monitor and record wave information in the sea area surrounding the ship using wave radar;

[0010] Step 2: Obtain complete wave field data and one-dimensional wave history data in a specific direction through the inversion algorithm;

[0011] Step 3: Combine the ship's own six-degree-of-freedom motion history data with the neural network prediction model to obtain the future motion trend of the ship.

[0012] Furthermore, the inversion algorithm in step 2 is trained using a deep learning algorithm based on a convolutional neural network autoencoder, with radar images as input and actual wave height images as output, to achieve the inversion from radar images to wave height images.

[0013] Furthermore, the convolutional neural network autoencoder includes an encoder, a fully connected layer, and a decoder.

[0014] Furthermore, the neural network prediction model in step 3 is a real-time ship motion prediction model under multiple modes constructed by analyzing the ship motion response characteristics.

[0015] Furthermore, in the neural network prediction model, the wave field data, wave time history data, and motion time history data are extracted using different feature extraction methods.

[0016] Furthermore, for the wave time history data and motion time history data, a recurrent neural network is used for feature extraction, and a first vector is obtained through a first fully connected layer.

[0017] Furthermore, the recurrent neural network is a long short-term memory network or a gated recurrent unit.

[0018] Furthermore, for the wave field data, a convolutional neural network is used for feature extraction, and a second vector is obtained through a second fully connected layer.

[0019] Furthermore, when using a convolutional neural network for feature extraction, a max pooling operation is then superimposed.

[0020] Furthermore, the future motion trend of the ship is obtained by splicing the first vector and the second vector through a third fully connected layer.

[0021] The beneficial effects of this invention are that by incorporating marine environmental information into the forecasting model, the model can better learn the excitation mechanism of environmental factors on ship motion, thereby improving the model's performance and shortening the training cycle.

[0022] The following will further explain the concept, specific structure, and technical effects of the present invention in conjunction with the accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the forecasting method module structure according to a preferred embodiment of the present invention;

[0024] Figure 2 This is a schematic diagram of a radar scan with a shadowed area according to a preferred embodiment of the present invention;

[0025] Figure 3 This is a schematic diagram of a radar image inversion model structure according to a preferred embodiment of the present invention;

[0026] Figure 4 This is a schematic diagram of a recurrent neural network structure according to a preferred embodiment of the present invention;

[0027] Figure 5 This is a schematic diagram of two-dimensional convolution calculation according to a preferred embodiment of the present invention;

[0028] Figure 6 This is a schematic diagram of a multimodal forecasting model structure according to a preferred embodiment of the present invention.

[0029] Among them, 1-wave radar, 2-inversion algorithm, 3-wave field data, 4-wave time history data, 5-ship, 6-motion time history data, 7-neural network prediction model, 8-motion change trend, 9-radar antenna, 10-sea level, 11-shaded area. Detailed Implementation

[0030] The preferred embodiments of the present invention are described below with reference to the accompanying drawings to make the technical content clearer and easier to understand. The present invention can be embodied in many different forms, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.

[0031] In the accompanying drawings, components with the same structure are indicated by the same numerical designation, and components with similar structures or functions are indicated by similar numerical designations. The dimensions and thicknesses of each component shown in the drawings are arbitrary, and the present invention does not limit the dimensions and thicknesses of each component. To make the illustrations clearer, the thickness of some components has been appropriately exaggerated in the drawings.

[0032] Based on existing related technologies and theories, this invention proposes an online ship motion prediction method based on X-band navigation radar. It introduces multimodal data such as wave field and one-dimensional wave to construct a multimodal scale ship motion response prediction model.

[0033] The data measurement, processing, and motion prediction modules involved in this invention are as follows: Figure 1 As shown, wave radar 1 first monitors and records wave information in the sea area surrounding ship 5. The complete wave field data 3 and one-dimensional wave time history data 4 in a specific direction are obtained through inversion algorithm 2. Combined with the six-degree-of-freedom motion time history data 6 of ship 5, they are input into the neural network prediction model 7 to finally obtain the future motion trend 8 of the ship.

[0034] To build a multimodal ship motion prediction model based on X-band radar, the first problem to solve is how to obtain wave and wave field information through wave radar.

[0035] The radar obtains wave information on sea level 10 by rotating and continuously scanning a fan-shaped area of ​​the sea surface using the radar antenna 9. However, the radar is affected by factors such as shadow modulation when scanning sea surface waves. Shadow modulation occurs because the electromagnetic waves emitted by the X-band radar almost skim the sea surface, making a very small angle with sea level 10. Since sea surface waves are undulating, some higher waves can block the propagation of electromagnetic waves, creating an area behind which electromagnetic waves cannot penetrate. This results in no echo or a weak echo signal in the covered area, thus creating some undetectable shadow areas 11, such as... Figure 2 As shown.

[0036] The latest technical methods have been developed, which use deep learning algorithms based on convolutional neural network autoencoders to train radar images as input and actual wave height images as output, and have achieved the inversion from radar images to wave height images. Figure 3 This is a diagram of the model structure proposed in this method, which links radar images with wave information through an autoencoder (AE). The method has been proven effective even with radar signals subject to noise interference, and can output one-dimensional wave history data for specific points (e.g., the bow of a ship), further enhancing the forecast model's perception of the surrounding marine environment.

[0037] After acquiring crucial wave field information, a multimodal real-time ship motion prediction model can be constructed by analyzing the ship's motion response characteristics. The model's input consists of the aforementioned wave field information, one-dimensional wave history data, and the ship's six degrees of freedom motion. Different feature extraction methods are employed based on the characteristics of these data.

[0038] For wave time-history data and motion time-history data, which are continuous variables in the time dimension, the motion changes at each moment depend on the environment and motion at past moments, and can be simplified as follows:

[0039] motion t =F(motion) t-1 wave t-1 )

[0040] =F(F(motion) t-2 wave t-2 ), wave t-1 )

[0041] =...=F(...F(motion0,wave0),wave1)

[0042] Here, F is a nonlinear function. Therefore, we cannot simply input all past information equally into the model. The model should focus more on changes in the "recent" moment, while appropriately "discarding" information from a long time ago.

[0043] In classic neural network algorithms, Recurrent Neural Networks (RNNs) satisfy the assumption that the model "memorizes" past information. However, classic RNN models suffer from the vanishing / exploding gradient problem, making convergence difficult. Therefore, many variants of RNNs have emerged, such as Long Short-Term Memory (LSTM) networks and Gated Recurrent Units (GRUs). Figure 4 As shown in the figure below, a classic LSTM unit includes a forget gate, a remember gate, and an output gate. These three gate structures allow the LSTM to selectively remember or forget data. The GRU unit is simpler than the LSTM, combining the forget gate and the output gate into a single gate, resulting in fewer parameters. Their performance is roughly equivalent; the choice depends on the model's performance.

[0044] Wave field data, representing wave information within the sea area surrounding a ship, is two-dimensional. Its two dimensions represent the offset relative to the ship's length and beam, respectively. This data records the wave height at that location.

[0045] wave field∈R n×n wave field = {h ij}, 0≤i,j≤n

[0046] Such two-dimensional data can be used for feature extraction through methods similar to image processing, such as convolutional neural networks (CNNs). Figure 5 Typically, after performing convolution operations, max pooling can be applied to extract more crucial information.

[0047] By combining the different feature extraction methods described above, the structure of the final multimodal ship motion prediction model can be obtained, such as... Figure 6 As shown in the figure, LSTM is used to extract features from wave time history data 4 and motion time history data 6, and a vector is obtained through a fully connected (FC) layer. For wave field data 3, key information is extracted through two-dimensional convolution and max pooling, and a vector is also obtained through a fully connected layer. The two are then concatenated and passed through a fully connected layer to obtain the final model prediction.

[0048] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for online ship motion prediction based on X-band navigation radar, characterized in that, Includes the following steps: Step 1: Monitor and record wave information in the sea area surrounding the ship using wave radar; Step 2: Obtain complete wave field data and one-dimensional wave history data in a specific direction through the inversion algorithm; The inversion algorithm in step 2 is trained using a deep learning algorithm based on a convolutional neural network autoencoder, with radar image as input and actual wave height image as output, to achieve the inversion from radar image to wave height image. Step 3: Combine the ship's own six-degree-of-freedom motion history data with the neural network prediction model to obtain the future motion trend of the ship. The neural network prediction model in step 3 is a real-time ship motion prediction model under multiple modes constructed by analyzing the ship motion response characteristics. In the neural network prediction model, the wave field data, wave time history data, and motion time history data adopt different feature extraction methods; For the wave time history data and motion time history data, a recurrent neural network is used for feature extraction, and a first vector is obtained through a first fully connected layer; For the wave field data, a convolutional neural network is used for feature extraction, and a second vector is obtained through a second fully connected layer; The future motion trend of the ship is obtained by splicing the first vector and the second vector through the third fully connected layer.

2. The online ship motion prediction method based on X-band navigation radar as described in claim 1, characterized in that, The convolutional neural network autoencoder includes an encoder, a fully connected layer, and a decoder.

3. The online ship motion prediction method based on X-band navigation radar as described in claim 1, characterized in that, The recurrent neural network is a long short-term memory network or a gated recurrent unit.

4. The online ship motion prediction method based on X-band navigation radar as described in claim 1, characterized in that, The method employs a convolutional neural network for feature extraction, followed by a max pooling operation.

Images