Intelligent forecasting method for ship movement

Abstract

The invention specifically discloses a ship motion intelligent forecasting method, and relates to the technical field of ship engineering. The method comprises the following steps: S1, acquiring ship navigation data including ship basic information and sea area wave spectrum basic information; s2, acquiring time sequence data including ship motion historical time sequence data and environment historical time sequence data; s3, constructing a ship motion intelligent forecasting model based on an improved Informer algorithm; s4, inputting ship motion time sequence data and environment time sequence data based on the ship motion intelligent forecasting model, and outputting a ship motion speed forecasting result of six degrees of freedom; and S5, carrying out anti-standardization operation on the output forecasting result to obtain a final ship motion forecasting result. According to the method, the ship motion forecasting precision and universality of different ship types in the complex marine environment are remarkably improved, and reliable technical support is provided for safety guarantee of ship navigation, improvement of operation efficiency, marine operation collaboration and the like.

Description

Technical Field

[0001] This invention relates to the field of marine engineering technology, and in particular to an intelligent prediction method for ship motion. Background Technology

[0002] During their voyages at sea, ships are constantly subjected to the combined effects of complex marine environmental factors such as ocean currents, waves, and winds, resulting in six degrees of freedom of motion: yaw, pitch, roll, heave, sway, and swell. These motions not only significantly increase the difficulty of ship navigation control but also pose a great threat to maritime safety and operational efficiency. At best, they disrupt the normal life and work rhythm of the crew and reduce the stability of equipment operation; at worst, they can cause ship attitude imbalance, and in extreme cases, even lead to catastrophic accidents such as capsizing, causing irreparable damage to lives, ship property, and the marine environment.

[0003] Conducting precise research on ship motion forecasting has become a key support for ensuring ship navigation safety and improving shipping efficiency. On the one hand, it can provide real-time and reliable auxiliary information for ship maneuvering decisions and motion control, helping crew members to avoid risks in advance. On the other hand, it also plays an irreplaceable role in marine operations support, emergency response optimization, and ship engineering research and standard setting, and is crucial to the entire ship operation process.

[0004] With the advancement of science and technology and the deepening of research, research on ship motion prediction has led to the development of various motion response prediction methods, including statistical prediction methods, grey prediction methods, Kalman filtering methods, convolutional neural network methods, and traditional neural network methods. However, existing technologies still have significant limitations in practical applications. Poor adaptability to complex environments: Existing methods can achieve good forecasting results under specific and stable motion patterns, but when faced with complex scenarios involving the coupling effects of waves, ocean currents, and sea breezes in the marine environment, the forecasting accuracy drops significantly and it is difficult to meet the actual navigation needs. Insufficient ship type universality: The model parameters and training data of most methods are highly dependent on specific ship types. When applied to ships with different lengths, beams, drafts or hull structures, the prediction results deviate significantly, requiring a large amount of data collection and model training to be carried out again, which is costly and inefficient. Lack of consideration for multivariate coupling: Previous studies have mostly adopted a univariate prediction approach, relying solely on the historical curve characteristics of each degree of freedom motion to predict future motion states. This approach has neither fully considered the inherent coupling relationship between the six degrees of freedom motion of a ship, nor effectively introduced the direct effects of key environmental variables such as bow wave height and wave direction on ship motion, resulting in an incomplete characterization of the actual motion laws by the model.

[0005] In summary, existing technologies suffer from limited forecast accuracy, poor adaptability to different ship types, and insufficient consideration of multi-factor coupling. There is an urgent need to propose an intelligent forecasting method that can comprehensively integrate ship characteristics, environmental variables, and multi-degree-of-freedom motion correlations to improve the accuracy and universality of ship motion forecasting for different ship types in complex marine environments. Summary of the Invention

[0006] The purpose of this invention is to propose an intelligent ship motion prediction method that effectively integrates ship characteristics, environmental variables, and multi-degree-of-freedom motion coupling correlations to improve the accuracy and universality of ship motion prediction for different ship types in complex marine environments, and to provide reliable technical support for ship navigation safety assurance and maneuvering decision assistance.

[0007] To achieve the above objectives, this invention proposes an intelligent prediction method for ship motion, the specific steps of which are as follows: Step S1: Obtain ship navigation data, including basic ship information and basic information of the sea wave spectrum; wherein, the basic ship information includes ship length, and the basic information of the sea wave spectrum includes wave frequency and wave energy density. Step S2: Obtain time-series data, including historical time-series data of ship movement and historical time-series data of the environment; Step S3: Construct a ship motion intelligent prediction model based on the improved Informer algorithm, including setting initial hyperparameter values, preprocessing time series data, embedding algorithm input, encoder and decoder processing, model training and hyperparameter updating; Step S4: Based on the intelligent prediction model of ship motion, input the ship motion time series data and environmental time series data, and output the prediction results of the ship's motion speed in six degrees of freedom; Step S5: Perform inverse standardization on the output forecast results to obtain the final ship motion forecast results.

[0008] Preferably, in step S2, the ship's motion history time series data includes the bow roll at the ship's center of gravity. , swaying Horizontal rocking Rise and fall , unrestrained , sweeping The speed of motion in six degrees of freedom, and the ship's speed. And heading; environmental historical time series data includes wave height and wave direction at the bow.

[0009] Preferably, in step S3, the hyperparameter setting adopts a gradient descent multi-parameter intelligent update algorithm based on ship-shaped features, which includes two steps: setting the initial hyperparameter values ​​and updating the hyperparameters. The initial hyperparameter values ​​include the initial input sequence length. Initial label sequence length Initial prediction sequence length Initial data batch size and initial learning rate The calculation formula is as follows: ; ; ; ; ; ; in, The top three wave energy densities are among the highest. , and The sum of , and Wave energy , and The corresponding wave frequency, For the length of the ship, This represents the average speed of the ship. t The timestamp sampling frequency, round This is the rounding function. M This represents the total number of rows contained in the historical time-series data. Given the length of the input sequence, This is the acceleration due to gravity.

[0010] Preferably, in step S3, the specific operation of time series data preprocessing is as follows: the ship motion history time series data and the environmental history time series data are arranged into a two-dimensional matrix in chronological order, missing values ​​are filled using linear interpolation, and each column is standardized separately using the standardized value Z-score, as shown in the following formula: ; in, For the first i The average of the column data, For the first i Standard deviation of the data For the first i The original data of a row in a column.

[0011] Preferably, in step S3, the algorithm input embeds an improved trigonometric function encoding method from the original Informer algorithm to adapt to the time format standard of year, month, day, hour, minute, and second used for timestamps in ship motion forecasting. Specifically, a six-dimensional matrix is ​​established, with the matrix columns corresponding to year, month, day, hour, minute, and second, respectively. The matrix values ​​remain constant for the year. , , , , Multiplying a 6×512 matrix with a six-dimensional matrix reduces the six-dimensional matrix to a higher dimension. M A 512-dimensional matrix; Historical time-series data on ship movement and environment were arranged chronologically into a two-dimensional matrix. This matrix was then multiplied by a 12×512 matrix randomly distributed between 0 and 1 to increase the dimensionality. M A 512-dimensional matrix, which is the result of upgrading a six-dimensional matrix to a higher dimension. M The 512-dimensional matrix was used to upgrade the historical time-series data of ship motion and environment to a higher dimension. M The 512-dimensional matrices are added and concatenated to serve as the high-dimensional input for the encoder and decoder processing steps.

[0012] Preferably, in step S3, the basic methods for encoder processing and decoder processing are as follows: the encoder employs a multi-head attention mechanism to process the embedded data... M The 512-dimensional matrix is ​​processed to capture long-term dependencies and internal correlations among multiple variables; the decoder uses the embedded... M The system takes a 512-dimensional matrix as input, masks the values ​​to be predicted in the matrix, and processes them using a self-attention mechanism. Then, it performs cross-attention calculations in conjunction with the encoder output sequence. Finally, it generates the prediction results through a feedforward network and a linear layer.

[0013] Preferably, in step S3, during model training, the ship motion history time series data and the environmental history time series data are divided into a training set, a test set, and a validation set in a ratio of 7:2:1.

[0014] Preferably, the specific method for updating hyperparameters in step S3 is as follows: hyperparameters 、 、 The calculation relationships are as follows: ; ; in, For the first n Length of the input sequence during the next hyperparameter update For the first Label sequence length during hyperparameter update 、 For the first Length of the predicted sequence during the next hyperparameter update; The updated value is greater than or equal to the initial value. 0.5 times the initial value, and less than or equal to the initial value. 1.5 times; hyperparameters The update method is as follows: When completing training When it is between 0.1 and 0.2: ; When completing training When greater than 0.2: ; in, For the first i Length of the input sequence during the next hyperparameter update For the first i The length of the input sequence during +1 hyperparameter updates. The maximum root mean square error between the model's predicted results and the original results; The calculation formula is as follows: ; in, The number of types of six-degree-of-freedom motion at the ship's center of gravity. The total number of time steps for predicting motion, and the current predicted sequence length. equal, For the first j Motion prediction value at any time, For the first j The actual value of motion at any given moment; hyperparameters and The update method is as follows: When training is completed Greater than 0.1 and When the hyperparameter is greater than 0.05 or less than 0.01, and The updated formula is as follows: ; ; ; in, For the first n Data batch size during each hyperparameter update For the first n Learning rate during hyperparameter update This represents the total number of times hyperparameter updates have been performed. For the first i Maximum root mean square error during +1 hyperparameter update This represents the difference in root mean square error between two consecutive hyperparameter updates. After completing the hyperparameter update, retrain the model until the hyperparameters are satisfied. The value is between 0.01 and 0.05.

[0015] Preferably, in step S4, the motion speed prediction result is obtained by inputting... Step-size ship motion time-series data and environmental time-series data, and through The step size sequence data is used as the data to be predicted. Prior data for step size data.

[0016] Preferably, in step S5, the destandardization operation method uses the standardized value Z-score for destandardization operation.

[0017] Therefore, this invention proposes an intelligent prediction method for ship motion, which has the following advantages: (1) This invention effectively captures the coupling relationship between the six degrees of freedom motion of a ship and the intrinsic connection between motion and environmental variables through the improved Informer algorithm and multi-head attention mechanism, which solves the limitations of traditional single-variable forecasting, greatly improves the accuracy of ship motion forecasting in complex marine environments, and reduces the forecast fluctuation risk under extreme sea conditions.

[0018] (2) This invention relies on a gradient descent multi-parameter intelligent update algorithm based on ship type characteristics. The initial value of the hyperparameter is deeply bound to the characteristics such as ship length and speed and sea wave information. It can be dynamically adjusted according to the training error, thus getting rid of the dependence of traditional methods on specific ship types. It is applicable to different types of ships and reduces the model reconstruction cost when applying across ship types.

[0019] (3) This invention ensures data quality through linear interpolation completion and Z-score standardization. It innovatively proposes a six-dimensional time matrix embedding and multi-source data dimensionality upsizing splicing, making full use of ship motion, environment and time stamp data. The output accurate forecast results can directly support ship maneuvering decisions, motion control and marine operation coordination, providing reliable technical support for ship navigation safety and operational efficiency improvement.

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0021] Figure 1 This is a flowchart of a smart prediction method for ship motion according to the present invention; Figure 2 This is a schematic diagram of the JONSWAP spectrum curve in an embodiment of the present invention; Figure 3 This is a comparative diagram of ship motion prediction results in an embodiment of the present invention; Figure 4 This is a schematic diagram comparing error indices of a ship motion prediction result in an embodiment of the present invention. Detailed Implementation

[0022] To make the technical solutions, advantages, and objectives of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below. The described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0023] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0024] Example This embodiment uses a standard KVLCC vessel model navigating in a certain sea area as an example to conduct a simulation and verify the effectiveness of the invention. Figure 1 As shown, this invention provides a method for intelligent prediction of ship motion, the steps of which are as follows: S1. Obtain vessel navigation data, including basic vessel information and basic wave spectrum information for the sea area; among which, vessel length... l The value is 322m. For vessels navigating the North Sea, the JONSWAP spectrum is used, with a sea state peak factor of 3.3, a significant wave height of 4m, and a peak period of 8s. Figure 2 As shown.

[0025] S2. Acquire time-series data, sampling at timestamp frequency. t The time series data of ship motion and environmental history was collected for 3000 seconds, with a time series of 1 second; the average ship speed was 7.22 m / s.

[0026] S3. Construct a ship motion intelligent prediction model based on the improved Informer algorithm, including setting initial hyperparameter values, time series data preprocessing, algorithm input embedding, encoder and decoder processing, model training and hyperparameter updating; S31. Initial hyperparameter values: In the JONSWAP spectrum sea state of the example, the wave energies of the top three wave densities are quite close. Therefore, a wave frequency of 0.005 Hz is taken as the step size, and the wave energy densities are taken as 126420588.28, 108567321.55, and 105238146.32, respectively, corresponding to frequencies of 0.125, 0.120, and 0.130. At this time, the initial hyperparameter values ​​can be calculated as follows: , , , , .

[0027] S32. Time series data preprocessing: The historical time series data of ship movement and environmental history are arranged into a two-dimensional matrix in chronological order. Missing values ​​are filled in by linear interpolation, and each column is standardized by Z-score. The mean and variance of all columns are recorded.

[0028] S33. Algorithm Input Embedding: The time lists in the ship motion history time series data and the environmental history time series data are kept constant year by year. , , , , The calculation is performed in this way, and then multiplied by a random number of 0 to 1 generated by the random function. 3D matrix, generate 3000 A 12-dimensional matrix. Simultaneously, a 12-dimensional matrix composed of random numbers between 0 and 1 generated by the `random` function. A 3D matrix is ​​generated by multiplying it by the normalized time-series data from step S32, resulting in a 3000-dimensional matrix. A 3000-dimensional matrix, combining two 3000-dimensional matrices. The dimensional matrices are added together and used as high-dimensional inputs for both encoder and decoder processing.

[0029] S34. Encoder and Decoder Processing: The 3000×512 matrix is ​​used as the encoder input. A multi-head attention mechanism is adopted to capture the long-term dependencies and internal correlations between multiple variables, so as to realize the mutual influence between motions of different degrees of freedom or between motion and environmental historical time series data. The decoder takes the 3000×512 matrix and the encoder output as input. First, the items that need to be predicted in the 3000×512 matrix are masked to avoid data leakage. Then, a self-attention mechanism is adopted to perform cross-attention calculation on the masked part of the 3000×512 matrix and the encoder output. Finally, the prediction result is generated through a feedforward network and a linear layer. The 6×512 matrix and the 12×512 matrix in step S33 are adjusted according to the prediction value.

[0030] S35. Model Training: The historical time-series data of ship motion and the historical time-series data of the environment were divided into training set, test set and validation set in a ratio of 7:2:1. Model training was carried out, and a total of 57 training sessions were completed, with 5 early stops.

[0031] S36, Hyperparameter Update: The model has completed training. , This result indicates that the algorithm's hyperparameters are poorly configured, leading to local optima and slow convergence. The hyperparameters need to be updated. , , , , Update the hyperparameters and return to step S35 to retrain the model.

[0032] S37. Conduct and complete model training. , This indicates that the hyperparameter settings are reasonable and the algorithm's prediction results have high accuracy.

[0033] S4. Based on the intelligent prediction model of ship motion, input ship motion time series data and environmental time series data, and output the prediction results of the ship's motion speed in six degrees of freedom; S5. The output forecast results are de-standardized using the standardized Z-score to obtain the final ship motion forecast results.

[0034] like Figures 3-4 As shown, the results predicted by this algorithm have a high degree of consistency with the actual motion of the ship and small error fluctuations. Compared with the traditional LSTM and Transformer algorithms, this method performs best under error indices such as MAE, MSE and RMSE.

[0035] It is worth noting that all contents not described in detail in this invention are existing technologies and are well known to those skilled in the art.

[0036] Therefore, this invention provides an intelligent ship motion prediction method. Through the improved Informer algorithm, intelligent hyperparameter update based on ship type characteristics, and efficient utilization of multi-source data, it significantly improves the accuracy and universality of ship motion prediction for different ship types in complex marine environments, while enhancing model stability. This provides reliable technical support for ensuring the safety of ship navigation, improving operational efficiency, and promoting collaborative marine operations.

[0037] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for intelligent prediction of ship motion, characterized in that, The specific steps are as follows: Step S1: Obtain ship navigation data, including basic ship information and basic information of the sea wave spectrum; wherein, the basic ship information includes ship length, and the basic information of the sea wave spectrum includes wave frequency and wave energy density. Step S2: Obtain time-series data, including historical time-series data of ship movement and historical time-series data of the environment; Step S3: Construct a ship motion intelligent prediction model based on the improved Informer algorithm, including setting initial hyperparameter values, preprocessing time series data, embedding algorithm input, encoder and decoder processing, model training and hyperparameter updating; Step S4: Based on the intelligent prediction model of ship motion, input the ship motion time series data and environmental time series data, and output the prediction results of the ship's motion speed in six degrees of freedom; Step S5: Perform inverse standardization on the output forecast results to obtain the final ship motion forecast results.

2. The intelligent prediction method for ship motion according to claim 1, characterized in that, In step S2, the ship's motion history time series data includes the bow roll at the ship's center of gravity. , swaying Horizontal rocking Rise and fall , unrestrained , sweeping The speed of motion in six degrees of freedom, and the ship's speed. And heading; environmental historical time series data includes wave height and wave direction at the bow.

3. The intelligent prediction method for ship motion according to claim 2, characterized in that, In step S3, the hyperparameter setting adopts a gradient descent multi-parameter intelligent update algorithm based on ship-shaped features, which includes two steps: setting the initial hyperparameter values ​​and updating the hyperparameters. The initial hyperparameter values ​​include the initial input sequence length. Initial label sequence length Initial prediction sequence length Initial data batch size and initial learning rate The calculation formula is as follows: ； ； ； ； ； ； in, The top three wave energy densities are among the highest. , and The sum of , and Wave energy , and The corresponding wave frequency, For the length of the ship, This represents the average speed of the ship. t The timestamp sampling frequency, round This is the rounding function. M This represents the total number of rows contained in the historical time-series data. Given the length of the input sequence, This is the acceleration due to gravity.

4. The intelligent prediction method for ship motion according to claim 3, characterized in that, In step S3, the specific operations of time series data preprocessing are as follows: the ship motion history time series data and environmental history time series data are arranged into a two-dimensional matrix in chronological order, missing values ​​are filled in using linear interpolation, and each column is standardized using a standardized value Z-score.

5. The intelligent prediction method for ship motion according to claim 4, characterized in that, In step S3, the algorithm input embeds an improved trigonometric function encoding method from the original Informer algorithm to adapt to the time format standard of year, month, day, hour, minute, and second used for timestamps in ship motion forecasting. Specifically, a six-dimensional matrix is ​​established, with the matrix columns corresponding to year, month, day, hour, minute, and second, respectively. The matrix values ​​remain constant for the year. , , , , Multiplying a 6×512 matrix with a six-dimensional matrix reduces the six-dimensional matrix to a higher dimension. M A 512-dimensional matrix; Historical time-series data on ship movement and environment were arranged chronologically into a two-dimensional matrix. This matrix was then multiplied by a 12×512 matrix randomly distributed between 0 and 1 to increase the dimensionality. M A 512-dimensional matrix, which is the result of upgrading a six-dimensional matrix to a higher dimension. M The 512-dimensional matrix was used to upgrade the historical time-series data of ship motion and environment to a higher dimension. M The 512-dimensional matrices are added and concatenated to serve as the high-dimensional input for the encoder and decoder processing steps.

6. The intelligent prediction method for ship motion according to claim 5, characterized in that, In step S3, the basic methods for encoder and decoder processing are as follows: the encoder employs a multi-head attention mechanism to process the embedded data. M Processing a 512-dimensional matrix captures long-term dependencies and internal correlations among multiple variables; The decoder is embedded M The system takes a 512-dimensional matrix as input, masks the values ​​to be predicted in the matrix, and processes them using a self-attention mechanism. Then, it performs cross-attention calculations in conjunction with the encoder output sequence. Finally, it generates the prediction results through a feedforward network and a linear layer.

7. The intelligent prediction method for ship motion according to claim 6, characterized in that, In step S3, during model training, the ship motion history time series data and the environmental history time series data are divided into training set, test set and validation set in a ratio of 7:2:

1.

8. The intelligent prediction method for ship motion according to claim 7, characterized in that, In step S3, the specific method for updating hyperparameters is as follows: hyperparameters 、 、 The calculation relationships are as follows: ； ； in, For the first n Length of the input sequence during the next hyperparameter update For the first Label sequence length during hyperparameter update 、 For the first Length of the predicted sequence during the next hyperparameter update; The updated value is greater than or equal to the initial value. 0.5 times the initial value, and less than or equal to the initial value. 1.5 times; hyperparameters The update method is as follows: When completing training When it is between 0.1 and 0.2: ； When completing training When greater than 0.2: ； in, For the first i Length of the input sequence during the next hyperparameter update For the first i The length of the input sequence during +1 hyperparameter updates. The maximum root mean square error between the model's predicted results and the original results; hyperparameters and The update method is as follows: When training is completed Greater than 0.1 and When the hyperparameter is greater than 0.05 or less than 0.01, and The updated formula is as follows: ； ； ； in, For the first n Data batch size during each hyperparameter update For the first n Learning rate during hyperparameter update This represents the total number of times hyperparameter updates have been performed. For the first i Maximum root mean square error during +1 hyperparameter update This represents the difference in root mean square error between two consecutive hyperparameter updates. After completing the hyperparameter update, retrain the model until the hyperparameters are satisfied. The value is between 0.01 and 0.

05.

9. The intelligent prediction method for ship motion according to claim 8, characterized in that, In step S4, the motion speed prediction result is obtained by inputting... Step-size ship motion time-series data and environmental time-series data, and through The step size sequence data is used as the data to be predicted. Prior data for step size data.

10. The intelligent prediction method for ship motion according to claim 9, characterized in that, In step S5, the denormalization operation method uses the normalized value Z-score for denormalization operation.

Images