Two-ship wave force rapid forecasting method based on machine learning
Through the rapid forecast method of wave force of two ships based on machine learning, the machine learning model is used to separate symbol labels and numerical feature quantities, and combined with the CFD method, the wave load overflow caused by wave interference in ship lateral recharge is solved, achieving high precision, Fast wave force forecasting reduces forecast error and response time.
Patent Information
- Application Number
- CN202411002865.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-25
- Publication Date
- 2024-11-08
AI Technical Summary
The prior art during the lateral recharge process of ships due to wave interference is excessively recharged, resulting in difficult recharge or risk of accidents. The existing forecasting methods are not accurate and accurate, and have strong dependence on computing resources, making it difficult to quickly. response.
Using the fast forecast method of wave force of two ships based on machine learning, by obtaining multiple sets of sample data, separating symbol labels and numerical feature quantities, training classification and regression models, using navigation conditions parameters to predict, combining CFD method to calculate wave frequency, and using Various machine learning models are validated and optimized.
It realizes high-precision and fast wave force forecasting for two ships, reduces forecast errors, improves response speed and efficiency, and is suitable for medium wave force forecasting for ships at sea lateral supply.
Smart Images

Figure CN118917195A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of ship technology, and in particular to a method for rapid prediction of wave forces of two ships based on machine learning. Background Art
[0002] When ships are sailing in the open sea and conducting maritime operations, they will continuously consume their stored food, fresh water, fuel, weapons and ammunition and other supplies. Therefore, in order to ensure that ships, especially small and medium-sized ships with relatively weak self-sufficiency, can operate normally, they must quickly replenish the consumed supplies. At present, maritime replenishment is an important logistical support means that ships must take when conducting offshore operations, escort and patrol at sea, which can provide effective protection for the combat readiness of ship equipment.
[0003] There are three main ways of replenishment during ship navigation: horizontal replenishment, vertical replenishment and vertical replenishment. Each replenishment method has its own advantages and application scope. Among them, horizontal replenishment occupies a dominant position in the replenishment method due to its advantages such as high efficiency and wide variety. When using the horizontal replenishment method, the receiving ship and the supply ship maintain a certain relative distance in the horizontal direction and sail in parallel. While maintaining the same speed and the same direction, the materials are transferred through the supply device connected between the two ships.
[0004] When a ship is replenishing laterally, the action of waves will cause the hull to sway, and at the same time, radiation waves will be formed on the water surface. After the incident wave encounters the ship, it will generate diffraction waves. At this time, the radiation waves and diffraction waves of each ship will generate new excitation effects after being transmitted to the adjacent hull. This new excitation will inevitably affect the swaying motion of the hull, and at the same time, new diffraction will be formed, resulting in wave interference between hulls. Due to the existence of this wave interference, the wave load and motion response of each ship may be more intense, resulting in difficulty or failure in replenishment, and may even cause collisions between hulls, resulting in unnecessary accidents and casualties.
[0005] Therefore, rapid prediction of wave forces when two ships approach each other is of great significance in the study of lateral replenishment of ships at sea. At present, the main methods for solving the hydrodynamic interference between close-range hulls in waves are: slicing method, hybrid method combining potential flow and viscous flow, three-dimensional potential flow method, three-dimensional potential flow time domain method, etc. The CFD method is mainly used for two ships in parallel in still water. Most prediction methods based on empirical formulas mainly focus on the influence of certain factors on the wave forces between ships, which have certain limitations, and the prediction accuracy and precision are not high. The CDF method has high prediction accuracy, but it is highly dependent on computing resources, has high time cost, and is difficult to respond quickly. Summary of the invention
[0006] In response to the above problems and technical requirements, this application proposes a method for rapid prediction of wave forces between two ships based on machine learning. The technical solution of this application is as follows:
[0007] A two-ship wave force rapid prediction method based on machine learning, the two-ship wave force rapid prediction method comprising:
[0008] Acquire multiple sets of sample data in a two-ship lateral replenishment scenario, each set of sample data includes navigation condition parameters of the two-ship lateral replenishment scenario and wave force data of the two ships;
[0009] Separate and obtain the symbol labels and numerical feature quantities of the wave force data of the two ships in each group of sample data;
[0010] Taking the sailing condition parameters of each set of sample data as input and the symbol labels of the corresponding wave force data of the two ships as output, a symbol label prediction model is obtained based on machine learning classification model training;
[0011] Taking the sailing condition parameters of each set of sample data as input and the corresponding numerical feature quantities of the wave force data of the two ships as output, a numerical prediction model is obtained based on machine learning regression model training;
[0012] The navigation condition parameters of the two-ship lateral replenishment scenario to be predicted are input into the trained symbolic label prediction model and the numerical prediction model respectively. According to the symbolic label prediction results output by the symbolic label prediction model and the numerical feature quantity prediction results output by the numerical prediction model, the wave force prediction results of the two ships in the scenario to be predicted are obtained.
[0013] A further technical solution is to separate the symbol labels and numerical feature quantities of the wave force data of the two ships in each set of sample data, including:
[0014] When the wave force data of the two ships in the sample data are positive values, the symbol label of the sample data is determined to be a positive label; when the wave force data of the two ships in the sample data are negative values, the symbol label of the sample data is determined to be a negative label;
[0015] The numerical characteristic quantities of the wave force data of the two ships are obtained based on the absolute values of the wave force data of the two ships in the sample data.
[0016] A further technical solution is to obtain numerical characteristic quantities of the wave force data of the two ships based on the absolute values of the wave force data of the two ships in the sample data, including: performing logarithmic processing on the absolute values of the wave force data of the two ships in the sample data to obtain the numerical characteristic quantities of the wave force data of the two ships;
[0017] The wave force forecast results of the two ships in the scenario to be forecasted include determining the wave force forecast results of the two ships as Among them, when the symbol label prediction result is a positive label, x1=1, when the symbol label prediction result is a negative label, x1=-1, and x2 is the numerical feature quantity prediction result.
[0018] A further technical solution is that the wave force data of the two ships in each set of sample data includes the first-order wave force data of the two ships and the second-order wave force data of the two ships, and the obtained wave force forecast results of the two ships in the scene to be forecast include the first-order wave force forecast results of the two ships and the second-order wave force forecast results of the two ships;
[0019] The rapid prediction method of wave forces between two ships also includes:
[0020] According to the output first-order wave force prediction results of the two ships, the wave frequency response curve of the first-order wave force in the frequency domain is obtained. The response transfer function in the frequency domain is written in the form of real and imaginary parts, and then transformed through the impulse response function method to output the first-order wave force time history curve in the time domain.
[0021] Its further technical solution is that the navigation condition parameters include motion parameters of the two ships and wave parameters. The motion parameters of the two ships include the lateral spacing between the two ships, the longitudinal spacing between the two ships, the bow angles of the two ships, and the speeds of the two ships. The wave parameters include wave direction and wave frequency. Among them, the lateral spacing between the two ships is the spacing between the two ships along the direction of the ship width, and the longitudinal spacing between the two ships is the spacing between the two ships along the direction of the ship length.
[0022] Its further technical solution is to obtain multiple sets of sample data in the scenario of two-ship lateral replenishment, including:
[0023] The CFD method is used to calculate the wave force data of the two ships under different navigation conditions in the two-ship lateral replenishment scenario, and the navigation conditions parameters with wave frequencies within the predetermined medium frequency band and the corresponding wave force data of the two ships are retained as sample data.
[0024] Its further technical solution is that the symbol label prediction model obtained by training the machine learning classification model includes:
[0025] A variety of machine learning classification models are used to train the models using sample data to obtain multiple candidate label prediction models. The machine learning classification models used include Bayesian classification, decision tree classification, gradient boosting tree, support vector machine, and K-nearest neighbor.
[0026] The k-fold cross-validation method was adopted for verification, and the candidate label prediction model with the highest classification accuracy was taken as the trained symbol label prediction model.
[0027] Its further technical solution is to obtain a numerical prediction model based on machine learning regression model training, including:
[0028] A variety of machine learning regression models are used to train the models using sample data to obtain multiple candidate numerical prediction models. The machine learning regression models used include linear regression, gradient regression decision tree, random forest, support vector regression, and adaptive boosting.
[0029] The k-fold cross-validation method is adopted for verification, and the candidate numerical prediction model with the smallest mean absolute error is taken as the trained numerical prediction model.
[0030] Its further technical solution is, when candidate numerical prediction models are respectively trained using gradient regression decision tree, random forest, support vector regression and adaptive enhancement, the model parameters with the least negative mean square error are found through random search parameter search method.
[0031] A further technical solution is that the determined symbolic label prediction model is obtained based on gradient boosting tree training, and the determined numerical prediction model is obtained based on random forest training.
[0032] The beneficial technical effects of this application are:
[0033] The present application discloses a method for rapid prediction of wave forces of two ships based on machine learning. The method separates the symbolic labels and numerical feature quantities of the wave force data of the two ships according to the data characteristics of the wave force data of the two ships, and then performs model training based on a classification model and a regression model respectively according to different prediction requirements of the symbolic labels and the numerical feature quantities. The symbolic label prediction model and the numerical prediction model obtained by training can be used to rapidly obtain the wave force prediction results of the two ships based on the navigation condition parameters. The method has high prediction accuracy and fast response speed, and has good prediction efficiency. It can rapidly predict the wave force when the two ships are close to each other, and is of great significance in the study of lateral replenishment of ships at sea.
[0034] Aiming at the characteristics of the wave forces of the two ships with changeable signs and large differences in magnitude, this method can better characterize the data characteristics of the wave forces of the two ships to serve model training and forecasting, and help reduce forecast errors by separating the symbol labels and extracting numerical features through logarithmic processing of the absolute values.
[0035] This method takes into account various motion parameters and wave parameters of the two ships that affect the wave forces of the two ships during model training, and can more comprehensively consider the impact of various factors on the wave forces of the two ships, which is conducive to improving the forecast precision and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 This is a method flow chart of a method for rapid prediction of wave forces of two ships according to an embodiment of the present application.
[0037] Figure 2The wave force data of two ships calculated by CFD in an example, as well as the data curve graph after taking the absolute value and further logarithmic processing.
[0038] Figure 3 The figure is a comparison curve diagram of the prediction result obtained by using the two-ship wave force rapid prediction method of the present application and the CFD calculation result in an example. DETAILED DESCRIPTION
[0039] The specific implementation of the present application is further described below in conjunction with the accompanying drawings.
[0040] This application discloses a method for rapid prediction of wave forces between two ships based on machine learning. Please refer to Figure 1 As shown in the flowchart, the two-ship wave force rapid prediction method includes the following steps:
[0041] Step 1: obtain multiple sets of sample data in the scenario of two-ship lateral replenishment, each set of sample data includes the navigation condition parameters of the two-ship lateral replenishment scenario and the wave force data of the two ships.
[0042] When constructing sample data, the navigation condition parameters of the two ships in the lateral replenishment scenario are obtained, and the wave force data of the two ships under different navigation condition parameters are calculated using the CFD method. According to expert knowledge in the field of wave force calculation between two ships, the influence of high and low frequency data on the actual wave force is not large. However, the value of this part of the data is relatively small, which will interfere with the numerical prediction of the main frequency band. Therefore, in one embodiment, the navigation condition parameters with wave frequencies in the predetermined medium frequency band and the corresponding wave force data of the two ships are retained as sample data, that is, the main frequency band data is retained and the high and low frequency band data are filtered out to reduce data interference and data processing. In one embodiment, the range of retained wave frequencies is 0.2 to 1.2.
[0043] In addition, the rapid prediction method of wave forces between two ships of the present application comprehensively considers various factors that affect the wave forces between ships. The navigation condition parameters in each scenario include the motion parameters of the two ships and the wave parameters. The motion parameters of the two ships include the lateral spacing between the two ships, the longitudinal spacing between the two ships, the bow angles of the two ships, and the speeds of the two ships. The wave parameters include the wave direction and the wave frequency. Among them, the lateral spacing between the two ships is the spacing between the two ships along the width of the ship, and the longitudinal spacing between the two ships is the spacing between the two ships along the length of the ship. The wave force data of the two ships in each set of sample data includes the first-order wave force data of the two ships and the second-order wave force data of the two ships. Regardless of
[0044] Step 2: Separate and obtain the symbolic labels and numerical feature quantities of the wave force data of the two ships in each group of sample data.
[0045] The wave forces of the two ships vary greatly, and the data range spans multiple orders of magnitude. The input navigation condition parameters are relatively simple, and the degree of change is far less than that of the wave forces of the two ships. At the same time, the wave forces of the two ships are manifested at low frequencies and in some areas, and direct forecasting is very difficult. In addition, in the forecast of the wave forces between the two ships, there are a large number of such forecast points: the forecast point and its adjacent forecast point have opposite signs, and the order of magnitude of the difference between the two points reaches 10 5 ~10 8 , which will make the prediction error of the machine learning model at these points very large. Therefore, before model training, it is necessary to preprocess the sample data according to the characteristics of the data set, including separating the symbolic features and numerical features of the wave force data of the two ships:
[0046] First, add a symbol label, including: when the wave force data of the two ships in the sample data is positive, determine the symbol label of the sample data as a positive label; when the wave force data of the two ships in the sample data is negative, determine the symbol label of the sample data as a negative label. In actual use, 1 can be directly used as a positive label and -1 can be used as a negative label.
[0047] After adding the symbol label, the absolute value of the wave force data of the two ships is taken. This process will make the wave force data of the two ships no longer cross the coordinate axis, which can reduce the prediction error caused by the change of positive and negative signs. Then, the numerical characteristic quantity of the wave force data of the two ships can be obtained based on the absolute value of the wave force data of the two ships in the sample data.
[0048] One approach is to directly use the absolute values of the wave force data of the two ships as numerical feature quantities. However, after separating the signs and values of the wave force data of the two ships in the sample data, the curve change trend of the absolute values of the wave force data of the two ships is still very complex, and the difference between the minimum and maximum values reaches multiple orders of magnitude. The drastically changing data structure will have a great impact on the prediction effect of the machine learning model. Therefore, in another embodiment, the absolute values of the wave force data of the two ships are not directly used as numerical feature quantities, but the absolute values of the wave force data of the two ships in the sample data are logarithmically processed as the numerical feature quantities of the wave force data of the two ships. The logarithmic function is a monotonically increasing function within its domain. Taking the logarithm of the absolute values of the wave force data of the two ships will not change the relative relationship between the wave force data of the two ships, but compress the data scale, making the wave force data of the two ships more stable, which can be better used for subsequent model training, and also weakening the heteroscedasticity of the subsequent model.
[0049] In an example, taking the second-order longitudinal force of the two ships in the wave force data of the two ships as an example, the second-order longitudinal force curves of the two ships at different wave frequencies obtained by direct CFD calculation are shown as follows: Figure 2 As shown in (a) of the figure, the curve diagram after separating the signs of the second-order longitudinal forces of the two ships and taking the absolute values is as follows Figure 2As shown in (b) of Figure 2 (b) It can be seen that the absolute value difference of the wave force data of the two ships is still large. Further taking the logarithm of the absolute value of the wave force data of the two ships, the curve obtained is as follows Figure 2 As shown in (c) in Figure 2 (c) It can be seen that the relative relationship of the data does not change after logarithmic processing, but the data scale is compressed.
[0050] Step 3, taking the navigation condition parameters of each set of sample data as input and the corresponding symbol labels of the wave force data of the two ships as output, a symbol label prediction model is obtained based on machine learning classification model training.
[0051] The prediction of symbolic labels is a typical binary classification problem, so a machine learning classification model is used to train the symbolic label prediction model. There are endless machine learning methods that can solve binary classification problems, and the scope of application and characteristics of different methods are also different. Only by using appropriate methods in a targeted manner can the ability of machine learning be fully utilized. Therefore, when training to obtain a symbolic label prediction model, first use a variety of different machine learning classification models to perform model training using sample data to obtain multiple candidate label prediction models. The machine learning classification models used in this embodiment include Bayesian classification, decision tree classification, gradient boosting tree, support vector machine, and K-nearest neighbor.
[0052] Then, the k-fold cross-validation method is adopted for verification, and the classification accuracy is used as the model evaluation index of the candidate label prediction model, so that the candidate label prediction model with the highest classification accuracy is used as the trained symbol label prediction model.
[0053] The classification accuracy is the proportion of correctly classified samples in all samples. The higher the classification accuracy, the better the classification effect. After comparing multiple machine learning classification models, the final symbolic label prediction model was obtained based on gradient boosting tree training. Gradient boosting tree is a machine learning method that can be used for classification problems. It uses an additive model and a forward step-by-step algorithm. In machine learning methods, it is generally necessary to calculate the residual of the loss function, but the calculation of the residual of the regression tree is not convenient. In this regard, the gradient boosting tree provides a new method, which is to estimate the residual of the boosting tree model through the negative gradient of the loss function. The gradient boosting tree is one of the most widely used algorithms with excellent generalization and expression capabilities. It has good interpretability and robustness, can automatically discover high-order relationships within features, and does not require normalization of complex data.
[0054] Step 4, using the navigation condition parameters of each set of sample data as input and the corresponding numerical feature values of the wave force data of the two ships as output, a numerical prediction model is obtained based on machine learning regression model training.
[0055] The prediction of numerical feature quantity is a typical regression problem, so a machine learning regression model is used to train the numerical prediction model. Similarly, there are endless machine learning methods that can solve regression problems, and the scope of application and characteristics of different methods are also different. Only by using appropriate methods in a targeted manner can the ability of machine learning be fully utilized. Therefore, when training to obtain a numerical prediction model, first use a variety of different machine learning regression models to use sample data for model training to obtain multiple candidate numerical prediction models. The machine learning regression models used in this embodiment include linear regression, gradient regression decision tree, random forest, support vector regression and adaptive enhancement. When the candidate numerical prediction models are trained using gradient regression decision tree, random forest, support vector regression and adaptive enhancement, these four types of models need to be adjusted for hyperparameters. The problem of wave force prediction between two ships is very complex in itself, and the scale of parameters that the model needs to search is relatively large. Therefore, in order to adapt to the data characteristics of wave force prediction between two ships, the model parameters with the least negative mean square error are found by the parameter search method of random search. Random search will not try all parameter combinations, but extract a fixed number of parameter combinations from a specified distribution. When the number of randomly sampled samples is large enough, the optimal hyperparameter combination can be found or approximated.
[0056] Then, the k-fold cross-validation method is adopted for verification, and the mean absolute error is used as the model evaluation index of the candidate numerical prediction model, so that the candidate numerical prediction model with the smallest mean absolute error is taken as the trained numerical prediction model.
[0057] After comparing multiple machine learning regression models, the final numerical prediction model was obtained based on random forest training. Random forest is a type of machine learning algorithm that is widely used to deal with regression and classification problems and has achieved good application results in many fields. The random forest model is also a classic ensemble learning method. The ensemble algorithm generally constructs multiple weak learners and then uses a certain strategy to combine weak learners to complete the learning task, thereby obtaining better performance than a single weak learner, taking into account both accuracy and generalization ability. The base learner used by the random forest algorithm is the decision tree. It uses the bagging method to integrate decision trees and further introduces random selection into the decision tree formation process through multiple decision trees. In the regression problem, the classification label of the decision tree will become a continuous value. In order to fit the label, the decision tree needs to be very lush to perform well on the training set, but it will also make the model prone to overfitting. However, the random forest effectively reduces the adverse effects by introducing a random process. During training, multiple training samples will be randomly and with replacement extracted from the training set for each tree as the training set of the tree; at the same time, in the process of tree branching, features will be randomly selected to control the depth of the tree. These two random processes make random forests less likely to fall into overfitting and have good noise resistance.
[0058] Step 5, after the symbol label prediction model and the numerical prediction model are trained, the two models can be used to quickly predict the wave force of the two ships, including first obtaining the navigation condition parameters in the two-ship lateral replenishment scenario, and the parameter content of the navigation condition parameters is as in step 1 above. Then, the navigation condition parameters to be predicted in the two-ship lateral replenishment scenario are respectively input into the trained symbol label prediction model and the numerical prediction model, and the prediction results of the wave force of the two ships in the scenario to be predicted are obtained according to the symbol label prediction results output by the symbol label prediction model and the numerical feature quantity prediction results output by the numerical prediction model.
[0059] Corresponding to the above data processing of the wave force data of the two ships, after obtaining the symbol label forecast results and the numerical characteristic quantity forecast results, the wave force forecast results of the two ships for the scene to be forecasted are obtained, including determining the wave force forecast results of the two ships as
[0060] Among them, when the symbol label prediction result is a positive label, x1=1, when the symbol label prediction result is a negative label, x1=-1, and x2 is the numerical feature quantity prediction result.
[0061] Corresponding to the data content of the wave force data of the two ships in the sample data, the wave force prediction results of the two ships here include the first-order wave force prediction results of the two ships and the second-order wave force prediction results of the two ships. According to the first-order wave force prediction results of the two ships and the second-order wave force prediction results of the two ships combined with the wave parameters, the response curve of the first-order wave force in the frequency domain with the wave frequency and the response curve of the second-order wave force with the wave frequency can be obtained.
[0062] In practical applications, the response curve of the first-order wave force with wave frequency cannot be used directly, and it needs to be converted from the frequency domain to the time domain. In order to ensure that the phase information in the time domain results is not lost, the response transfer function in the frequency domain is written in the form of real and imaginary parts, and then converted through the impulse response function method. By inputting the frequency domain response transfer function and wave parameters, the first-order wave force time history curve in the time domain is output.
[0063] The method of this application can be used to quickly and accurately predict the wave forces of two ships. After comparison and verification of multiple sets of test data, the prediction accuracy of this application is much higher than the prediction results of empirical formulas. Taking the second-order longitudinal force between two ships as an example, the model prediction results obtained by using this application to predict the second-order longitudinal force between two ships during the replenishment of two types of ships are compared with the CFD calculation results. Figure 3 As shown in the comparison results, it can be seen that the prediction accuracy of the method of the present application is reliable. At the same time, compared with the traditional CFD method, the response time of the prediction method of the present application can reach the millisecond level.
[0064] The above is only a preferred embodiment of the present application, and the present application is not limited to the above embodiments. It is understood that other improvements and changes directly derived or associated by those skilled in the art without departing from the spirit and concept of the present application should be considered to be included in the protection scope of the present application.
Claims
1. A method for rapid prediction of wave forces between two ships based on machine learning, characterized in that: The two-ship wave force rapid prediction method comprises: Acquire multiple sets of sample data in a two-ship lateral replenishment scenario, each set of sample data includes navigation condition parameters of the two-ship lateral replenishment scenario and wave force data of the two ships; Separate and obtain the symbol labels and numerical feature quantities of the wave force data of the two ships in each group of sample data; Taking the sailing condition parameters of each set of sample data as input and the symbol labels of the corresponding wave force data of the two ships as output, a symbol label prediction model is obtained based on machine learning classification model training; Taking the sailing condition parameters of each set of sample data as input and the corresponding numerical feature quantities of the wave force data of the two ships as output, a numerical prediction model is obtained based on machine learning regression model training; The navigation condition parameters of the two-ship lateral replenishment scenario to be predicted are respectively input into the trained symbolic label prediction model and the numerical prediction model, and the wave force prediction results of the two ships in the scenario to be predicted are obtained according to the symbolic label prediction results output by the symbolic label prediction model and the numerical feature quantity prediction results output by the numerical prediction model.
2. The method for rapid prediction of wave forces between two ships according to claim 1 is characterized in that: The separation obtains the symbol labels and numerical feature quantities of the wave force data of the two ships in each group of sample data, including: When the wave force data of the two ships in the sample data are positive values, the symbol label of the sample data is determined to be a positive label; when the wave force data of the two ships in the sample data are negative values, the symbol label of the sample data is determined to be a negative label; The numerical feature quantities of the wave force data of the two ships are obtained based on the absolute values of the wave force data of the two ships in the sample data.
3. The method for rapid prediction of wave forces between two ships according to claim 2 is characterized in that: The step of obtaining the numerical characteristic quantities of the wave force data of the two ships based on the absolute values of the wave force data of the two ships in the sample data comprises: performing logarithmic processing on the absolute values of the wave force data of the two ships in the sample data to obtain the numerical characteristic quantities of the wave force data of the two ships; The wave force forecast results of the two ships in the scenario to be forecasted are obtained by determining the wave force forecast results of the two ships as Among them, when the symbol label prediction result is a positive label, x1=1, when the symbol label prediction result is a negative label, x1=-1, and x2 is the numerical feature quantity prediction result.
4. The method for rapid prediction of wave forces between two ships according to claim 1 is characterized in that: The wave force data of the two ships in each set of sample data include the first-order wave force data of the two ships and the second-order wave force data of the two ships, and the obtained wave force forecast results of the two ships for the scenario to be forecast include the first-order wave force forecast results of the two ships and the second-order wave force forecast results of the two ships; The two-ship wave force rapid prediction method also includes: According to the output first-order wave force prediction results of the two ships, the wave frequency response curve of the first-order wave force in the frequency domain is obtained. The response transfer function in the frequency domain is written in the form of real and imaginary parts, and then transformed through the impulse response function method to output the first-order wave force time history curve in the time domain.
5. The method for rapid prediction of wave forces between two ships according to claim 1 is characterized in that: The navigation condition parameters include motion parameters of the two ships and wave parameters. The motion parameters of the two ships include the lateral spacing between the two ships, the longitudinal spacing between the two ships, the bow angles of the two ships, and the speeds of the two ships. The wave parameters include wave direction and wave frequency. Among them, the lateral spacing between the two ships is the spacing between the two ships along the width of the ship, and the longitudinal spacing between the two ships is the spacing between the two ships along the length of the ship.
6. The method for rapid prediction of wave forces between two ships according to claim 5 is characterized in that: The obtaining of multiple sets of sample data in the scenario of two-ship lateral replenishment includes: The CFD method is used to calculate the wave force data of the two ships under different navigation conditions in the two-ship lateral replenishment scenario, and the navigation conditions parameters with wave frequencies within the predetermined medium frequency band and the corresponding wave force data of the two ships are retained as sample data.
7. The method for rapid prediction of wave forces between two ships according to claim 1, characterized in that: The symbol label prediction model obtained by training the machine learning classification model includes: A variety of machine learning classification models are used to train the models using sample data to obtain multiple candidate label prediction models. The machine learning classification models used include Bayesian classification, decision tree classification, gradient boosting tree, support vector machine, and K-nearest neighbor. The k-fold cross-validation method was adopted for verification, and the candidate label prediction model with the highest classification accuracy was taken as the trained symbol label prediction model.
8. The method for rapid prediction of wave forces between two ships according to claim 1, characterized in that: The numerical prediction model obtained by training the machine learning regression model includes: A variety of machine learning regression models are used to train the models using sample data to obtain multiple candidate numerical prediction models. The machine learning regression models used include linear regression, gradient regression decision tree, random forest, support vector regression, and adaptive boosting. The k-fold cross-validation method is adopted for verification, and the candidate numerical prediction model with the smallest mean absolute error is taken as the trained numerical prediction model.
9. The method for rapid prediction of wave forces between two ships according to claim 8, characterized in that: When the candidate numerical prediction models are trained using gradient regression decision tree, random forest, support vector regression and adaptive boosting respectively, the model parameters with the least negative mean square error are found through random search parameter search method.
10. The method for rapid prediction of wave forces between two ships according to claim 7 or 8, characterized in that: The determined symbolic label prediction model is obtained based on gradient boosting tree training, and the determined numerical prediction model is obtained based on random forest training.
Citation Information
Patent Citations
Risk assessment method based on limited water area ship risk assessment model parameter correction
CN116029554A
Three-degree-of-freedom unmanned surface ship model online identification method
CN117993102A
CFD-based ship wind field rapid forecasting method and system
CN118261079A
Transverse supply simulation method for navigation of two ships in waves
CN118348826A
Cited By
Two-ship sailing transverse supply decision-making method considering inter-ship effect
CN121386755A
