Research On Hull Form Design Using Machine Learning With Small Sample Size

The increasingly severe energy problems and the introduction of energy-saving and emission reduction policies have led to the favoring of green ship technology and the innovative development of the hull form design field.With the aid of computer technology,the traditional hull form design method based on parent ship or series ship type data has been developed significantly.At present,the efficiency of hull form design is greatly affected by the speed and accuracy of ship performance prediction.Hull form design still follows the traditional design process,and the innovation hull form design needs to be improved.In recent years,with the development of artificial intelligence technology,the application of machine learning in solving engineering problems has gradually become a research hotspot.Ship resistance,which is closely related to hull form,is one of the priority ship performances in hull form design.Therefore,based on machine learning technology,research on ship resistance prediction,ship bow lines optimization method and hull form design method with small sample size is carried out.The main research contents and results are as follows:In the second part,the influencing factors of ship resistance prediction are analyzed from the two aspects of resistance influencing factors and machine learning model construction process.Based on the analysis of ship resistance,the Froude resistance classification method is used as the decomposition method of ship resistance,and the influencing factors of ship resistance are obtained.To address the problems of poor reliability of machine learning models and low accuracy of prediction results with small sample size,an evaluation method of machine learning models with small sample size is proposed.Compared with the commonly methods,the resistance prediction results obtained by the proposed model evaluation method with small sample size have higher accuracy,and the reliability of the model is improved.In the third part,the hull form expression used for machine learning ship resistance prediction with small sample size is proposed.Two hull form expressions of hull form parameters and offset values are used to predict the resistance under two situations.The one is the resistance prediction of a single ship under different draughts,and the other is the resistance prediction of a specified ship using other ship’s data.The prediction results using two hull form expressions are compared.To address the problem of hull form parameters selection,the influence of the dimensionless of hull form parameters on the prediction results is studied,and the rationality of hull form parameters selection is analyzed.The research shows that better prediction results can be obtained by using the dimensionless hull form parameters as the input of the resistance prediction model with small sample size.In the fourth part,the method for constructing a machine learning model for resistance prediction is proposed.Based on the data of the same type of ship,to address the problem of low prediction accuracy using the basic machine learning model,the ensemble learning resistance prediction model is used to improve the resistance prediction accuracy.To address the problem of small sample size for ship resistance prediction,transfer learning is used to construct resistance prediction models based on the resistance prediction models used for different types of ship or existing empirical formulas,and the accurate prediction of ship resistance is realized.In the fifth part,an optimization method of bow lines based on ensemble learning neural network is proposed.The surface deformation technology of radial basis function interpolation and CFD(Computational Fluid Dynamics)numerical simulation technology are used to obtain the ship resistance with different bow lines in this study,and the ensemble learning is applied to construct a network model reflecting the relationship between the bow line parameters and the resistance.Then the bow lines with the best resistance performance are obtained using evolution strategies,and are verified by CFD numerical simulation technology.The study shows that the total resistance of the optimized ship is reduced by2.924%In the sixth part,a hull form design method based on generative neural network is proposed.The research uses the data of parent ships to construct a conditional variational autoencoder generative neural network model,and generates the hull form parameters under the given resistance conditions.The design results are verified by the resistance prediction model based on ensemble learning and CFD simulation technology.The proposed method breaks the traditional hull form design process,and reduces the dependence of hull form design on computing resources,shortens the design cycle,and improves design efficiency.Machine learning is used to realize the research on ship resistance prediction and hull form design methods with small sample size,which can provide important guidance for the application of machine learning in the field of ships.